Documente Academic
Documente Profesional
Documente Cultură
Regine Choe
A Dissertation
in
2005
Arjun G. Yodh
Supervisor of Dissertation
Randall D. Kamien
by
Regine Choe
Dedication
iv
Acknowledgements
When I was at a crossroads in deciding the graduate school, I ran across the Physics Today article
written by Arjun Yodh and Britton Chance (BC). The multidisciplinary aspect and clinical applica-
bility of diffuse optical imaging immediately seized my attention and became the deciding factor
in the end. I have been truly fortunate to be able to work in a superb environment created by Arjun
and BC with a wonderful ensemble of people. It is my great pleasure to acknowledge all of them
Arjun has always been a constant force which provided me with a big picture and guided me
to be a confident and independent researcher. He has taught me how to identify and approach the
problem with firm grasp of fundamental physical understanding. His critical scientific insights have
challenged and directed me to strive for intellectual advances. He has taught me how to deliever
the presentation with clarity and how to write scientific papers and grants. BC is a legendary figure
in diffuse optical community with full of brilliant innovative ideas and keen sense of research flow.
As my second advisor, he introduced me to the clinical fetal brain oxygen monitoring project in
the beginning years of my research. This exposure was invaluable in understanding the clinical
and physiological motivation, experimental instrument and protocol design. His trust, support and
Besides my advisors, I was lucky to have two mentors who provided the specific research di-
rection and inspired me greatly along the way. I could not have survived my first year of research
without Nirmala Ramanujam, with whom I worked on the fetal project. Nimmi is a great teacher
with uncanny ability to explain the key concept with such clarity even the newcomers can under-
stand. She was always willing to spend time to brainstorm the direction of research in a systematic
manner and to discuss the results with great patience. She has become a role model for me. Joseph
v
Culver is a great experimental physicist who has taught me the ins and outs of the diffuse optical to-
mography of breast cancer. Joe has designed and established the very backbone of current research
instrumentation. Not only he was well-versed in the experiment, but also great in the theoretical
I often admire at the great ensemble of good-natured and brilliant members working in Arjun’s
laboratory. Turgut Durduran is such an invaluable friend and colleague that one may only hope
to meet once in a lifetime. Starting from the very first year of preparing for qualifying exams,
throughout ups and downs of graduate school (both academic and personal), and finally the pro-
cess of writing the thesis, he has given constant encouragement and strength to go on. His keen
insights and brilliat ideas shining through and his non-hesitant active approaches to research al-
ways inspire me. It has been an invigorating experience to collaborate with him on some projects
in the past, which I am looking forward to continuing. I enjoy working with the members of breast
cancer project: Alper Çorlu, Kijoon Lee and Soren Konecky. Alperci ğim has a flair for theoretical
development, providing fundamental basis for breast cancer imaging. He has given me a steady
academic and emotional support with a protectiveness of a younger brother. Kijoon and Soren are
like a breeze of fresh air with new ideas, helping in every possible way with great eagerness. I am
grateful to Chao Zhou who sometimes goes out of his way to help me even though we do not share
a project. Jonathan Fisher is a virtuoso in science and piano who has continuously amazes people
around him. He is my favorite late night snack buddy who gives me great advices on self-esteem
and confidence. I would like to also thank Hsing-wen for being there for me always. Also, I would
like to thank Guoqiang Yu, Ulas Sunar and David Busch for helpful exchange of information and
who has helped me greatly on recruitment and measurements of patients. Her dedication to our
vi
project and eagerness to help are greatly appreciated.
During her short time in our laboratory, Anisa Nayeem established the starting point of clinical
connections when we did not have any clue. Leonid Zubkov, Cecil Cheung, Monica Holboke and
Joseph Giammarco were very helpful in experimental and theoretical side of Arjun’s laboratory.
In BC’s laboratory, I had great interactions with Yu Chen, Vasilis Ntziachristos, Xavier Intes and
Shoko Nioka. Gargi Vishnoi was a post-doc when we were doing fetal project together, who was
a supportive friend. Mary Leonard provided with her beautiful illustration for our presentations.
Dorothea Coleman (Dot) and Glen Fechner were always helpful in administrative aspects of re-
search and grants. Also, I would like to thank the business office in the Department of Physics
and Astronomy for their timely manner in handling the purchase orders and other financial re-
lated tasks, making research go smoothly. The environment of Physics Department was always
favorable. I enjoyed taking physics courses from various wonderful faculty members. Especially
I would like to thank my thesis committee members, Dr. Randall Kamien, Dr. Nigel Lockyer, Dr.
Gino Segre and Dr. Paul Heiney for their time and effort.
Our research could not have been possible without many clinical collaborators. Especially I
would like to thank Dr. Douglas Fraker for helping us launch the breast cancer imaging research in
the Hospital of the University of Pennsylvania and introduce us to numerous enthusiastic collabo-
rators. Dr. Brian Czerniecki and Dr. Julia Tchou have been very supportive and actively referred
the interested breast cancer patients to our study. Dr. Mark Rosen has introduced us to new ap-
plication of neoadjuvant chemotherapy monitoring and connected us with Dr. Angela DeMichele.
He is an enthusiastic and helpful collaborator who understands the strength and weakness of the
new technology. Dr. Mitch Schnall and Dr. Gautham Mallampati have helped us in comparing the
localization of breast tumor between MRI and optical method. I would like to also thank Lauren
vii
Sherman for neoadjuvant chemotherapy patient interface, Nancy O’Connor from Penn Tower, and
Kathleen McCarthy for referring patients from PPG study patient population along with Stephanie
Davis and Tamara April. For the fetal project, the animal model study was not possible without the
help of Dr. Mark Nijland and Dr. Peter Nathanielsz of Cornell University. Most of all, I would like
to thank volunteers and patients who participated in our clinical measurements for their generosity
and courage towards developing research for the good of future generation.
Lastly, I’d like to thank all my family and friends who have shaped who I am now. Among my
friends, I single out devoted friends who were there for me throughout my graduate years. Marky,
Saurav, Turgut and I formed a society µστ ρ exploring authentic ethnic cuisines from all around
the world. Their adventurous spirits towards trying new things made a lot of fond memories. Also,
I would like to thank compassionate Eylem for inviting me to dinners when I was too busy to cook
while writing this thesis. I would like to thank Jamie for giving me a patient and caring instructions
for the flight lessons. Through becoming a pilot, I have learned whole new level of self-control,
patience as well as actual piloting skill and came to terms with my own abilities. My parents have
always believed in me and supported me. My mother encouraged me to enjoy the moment while
trying my best. My father has been the center and driving force for scientific mindset of our family.
My brother Yookyung has shared his experties in medical illustrations by creating exceptionally
fine illustrations used in my paper and thesis. Also, he has been there for me in my time of hardship
I would like to end the acknowledgement with a quote from one of my favorite animations,
viii
Abstract
Diffuse Optical Tomography and Spectroscopy of Breast Cancer and Fetal
Brain
Regine Choe
Arjun G. Yodh
Diffuse optical techniques utilize light in the near infrared spectral range to measure tissue phys-
spatial map of tissue properties such as total hemoglobin concentration, blood oxygen saturation
and scattering can be obtained using model-based reconstruction algorithms. In this thesis, diffuse
optical techniques were applied for in vivo breast cancer imaging and trans-abdominal fetal brain
oxygenation monitoring.
For in vivo breast cancer imaging, clinical diffuse optical tomography and related instrumen-
tation was developed and used in several contexts. Bulk physiological properties were quantified
for fifty-two healthy subjects in the parallel-plate transmission geometry. Three-dimensional im-
ages of breast were reconstructed for subjects with breast tumors and, tumor contrast with respect
to normal tissue was found in total hemoglobin concentration and scattering and was quantified
for twenty-two breast carcinomas. Tumor contrast and tumor volume changes during neoadju-
vant chemotherapy were tracked for one subject and compared to the dynamic contrast-enhanced
MRI. Finally, the feasibility for measuring blood flow of breast tumors using optical methods was
In a qualitatively different set of experiments, the feasibility for trans-abdominal fetal brain
ix
oxygenation monitoring was demonstrated on pregnant ewes with induced fetal hypoxia. Prelimi-
In total, this research has translated diffuse optical tomography techniques into clinical research
environment.
x
Contents
Dedication iv
Acknowledgements v
Abstract ix
1 Introduction 1
1.2.1 Brief history of breast optical imaging: Before the Diffusion Approximation 6
1.2.2 Brief history of breast optical imaging: After the Diffusion Approximation 7
2 Theory 17
xi
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.1 Initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2.3 Calculation of χ2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3 Spectroscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.4 Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3 Experimental Techniques 43
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.2 Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.2.2 Detectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.2.2.1 Photodiode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
xii
3.2.2.6 SNR comparison of detectors . . . . . . . . . . . . . . . . . . . 49
3.6 Instrumentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
xiii
4.2.2 Diagnostic procedures . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
xiv
4.5.2.6 Image correlation analysis between MRI and DOT . . . . . . . . 129
5.2 Trans-abdominal Near Infrared Oximetry of Hypoxic Stress in Fetal Sheep Brain
in utero . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
xv
5.4.3 Sensitivity to fetal signal . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
6 Summary 181
Glossary 183
Bibliography 186
xvi
List of Tables
xvii
List of Figures
xviii
3.8 Laser diode style . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
xix
3.31 µa of water and lipid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.15 3D reconstruction of breast (subject #111) with multiple ductal carcinomas . . . . 119
xx
4.16 Tumor location of subject #95 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
4.31 Tumor volume and THC contrast decrease with chemotherapy . . . . . . . . . . . 139
4.33 Temporal auto-correlation curves of tumor and healthy breast tissue . . . . . . . . 146
4.34 Relative blood flow scans of a healthy breast and a diseased breast . . . . . . . . . 147
xxi
5.5 Two groups of fetal hypoxia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
5.8 Summary of physiological parameters at prior, during and after birth . . . . . . . . 170
5.13 Fractional amplitude and phase difference between two-layer vs homogeneous . . . 175
xxii
Chapter 1
Introduction
Diffuse optical tomography (DOT) and spectroscopy (DOS) are non-invasive techniques used to
measure the optical properties of physiological tissue. In the near-infrared (NIR) spectral window
of 600 - 1000 nm, photon propagation in tissues is dominated by scattering rather than absorp-
tion. Photons experience multiple scattering events as they propagate deeply into tissue (up to
10 cm). The primary chromophores in this spectral window are oxygenated hemoglobin (HbO 2 ),
deoxygenated hemoglobin (Hb), water (H2 O) and lipid. Each chromophore possesses a distinct
spectrum as shown in Fig. 1.1(a). A weighted sum of the contributions from each chromophore
which correponds approximately to the tissue absorption coefficient (µ a ) is also shown. Here,
the concentration of each chromophore is adjusted to a value typically found in breast tissue: the
hemoglobin (CHb ) as ∼6 µM, and the tissue is assumed to contain a 31 % water and a 57 % lipid
1
of 30 µM, and a blood oxygen saturation, StO2 = CHbO2 /T HC, of ∼80% (see Chapter 2 for
more details). Notice that absorption measurements at multiple wavelengths are required to extract
the concentration of each chromophore. Fig. 1.1(b) shows a typical scattering spectra found in the
breast tissue. The tissue scattering depends on the photon random walk step in the medium. The
reduced scatttering coefficient (µ0s ) which is the reciprocal of the photon random walk step length,
is often modeled within a simplified Mie-scattering approximation [185, 188], i.e. µ 0s (λ) = Aλ−b
where λ is the light wavelength. Notice that scattering is 100× larger than the absorption, and the
12
0.2 HbO2
Hb 10
H2O
0.15 lipid 8
µs′ (cm−1)
(cm )
−1
total
6
0.1
µa
4
0.05
2
0 0
650 700 750 800 850 900 950 1000 650 700 750 800 850 900 950 1000
wavelength(nm) wavelength (nm)
(a) (b)
Figure 1.1: Spectrum of absorption and scattering in the NIR. (a) Spectra of major chromophores
with adjusted concentrations found in typical breast tissue: oxygenated hemoglobin (HbO 2 ) and
deoxygenated hemoglobin (Hb), water and lipid. (b) Scattering spectra. See text for details.
Accurate retrieval of tissue properties based on the DOS and DOT measurements requires that
absorption and the scattering be decoupled. A light transport model based on the diffusion approx-
imation [121] is widely used to describe photon propagation in the NIR. Using optical measure-
ments at multiple source-detector positions on the tissue surfaces, one can reconstruct the internal
2
distribution of the absorption coefficient (µa ) and the reduced scattering coefficient (µ0s ) in three-
dimensions based on the transport model. Physiological images of total hemoglobin concentration,
blood oxygenation, water and lipids are then derived from this information. Thus far DOT and DOS
have generated a lot of scientific interest and have been applied in various deep-tissue applications
including breast cancer imaging [62, 69, 102, 116, 119, 144, 165, 191, 194, 213, 222, 300], brain
functional imaging [80], stroke detection [23, 63, 133], muscle functional studies [148, 226, 285],
I worked primarily on two major applications of DOT and DOS: breast cancer imaging and
fetal brain oxygenation monitoring. In the following sections, the clinical motivation and the brief
Approximately one in nine women will develop breast cancer in their lifetime, and of these cancers,
approximately 30% will be fatal [2, 157]. Currently, several clinical methods are used in breast
cancer screening and diagnosis [95, 152]. The most common of these are palpation and X-ray
mammography. Other methods include ultrasound, magnetic resonance imaging (MRI), positron
emission tomography (PET), and surgical and needle biopsy. Some of these techniques rely on
intrinsic characteristics of breast tissue to image the breast or to identify lesions within it, while
others employ exogenous tracers or contrast agents. Similarly, some are useful in screening while
The most effective screening technique at this time is X-ray mammography. X-Ray mammog-
raphy, however, has a ∼ 22% false negative rate in women under 50 [151]. The method cannot
accurately distinguish between benign and malignant tumors [131]; it has been shown in recent
3
studies to have lower sensitivity in pre-menopausal women [152] and to be of limited clinical
value for women under 35 years of age [126]. Furthermore, increased mammographic density can
arise in some post-menopausal women, for example due to hormone replacement therapy [14], or
cyclically during the menstrual cycle of younger women [280]; these effects reduce the effective-
ness of mammographic screening. Techniques such as MRI and ultrasound are sometimes used
in addition to X-Ray mammography, but have limitations such as high cost, low throughput, lim-
ited specificity (MRI) and low sensitivity (ultrasound). Thus, there is still a need to detect cancers
earlier for treatment [1, 32], to detect cancers missed by mammography [21, 110, 274], and to add
specificity to the mix, since the majority of invasive follow-up procedures (e.g. surgical biopsies)
Near-infrared (NIR) diffuse optical tomography (DOT) is based on the study of functional pro-
cesses and provides several unique measurable parameters with potential to enhance breast tumor
sensitivity and specificity. The techniques utilize non-ionizing radiation, are non-invasive, and
are often technologically simple and fast. Tissue optical absorption coefficients provide access to
blood dynamics, total hemoglobin concentration (THC), blood oxygen saturation (StO 2 ), water
concentration (CH2 O ) and lipid content. These tissue properties are often substantially different in
rapidly growing tumors; for example, high concentrations of hemoglobin with low oxygen satura-
tion are suggestive of rapidly growing tumors [267, 268, 276] due to their high metabolic demand
and (sometimes) poor perfusion. In a different vein, an overall increase in organelle population due
to proliferation of cancer cells leads to an increase of optical scattering coefficients for the tumor.
Nuclei and mitochondria are major contibutors to reduced optical scattering coefficients [184] as
well as collagen in the extracellular matrix [255]. Finally, optical absorption, fluorescence, and
scattering of exogenous contrast agents that occupy vascular and extravascular space also provide
4
useful forms of sensitization.
The utility of DOT for tumor monitoring has been recently demonstrated with success for
apy) is an important therapeutic approach for women with locally advanced breast cancer (LABC).
LABC generally refers to lesions larger than 5 cm that may or may not have spread beyond the
breast and axillary lymph nodes. If the patient responds to neoadjuvant chemotherapy, the size
of the primary tumor decreases, facilitating better control through surgery while potentially erad-
patients to undergo breast conservation therapy without negatively impacting local recurrence rates
or long-term outcome when compared with adjuvant chemotherapy [98]. Additionally, the neoad-
juvant setting provides a means to monitor the effectiveness of chemotherapy by observing its
Information about tumor response during chemotherapy may be useful for treatment optimiza-
has limited utility for assessing tumor response due to chemotherapy-induced fibrosis [52, 243,
269]. While MRI has proved useful for defining the extent of residual disease when compared with
pathology [74,92], and while dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI)
has demonstrated ability to monitor tumor size and vascularity during neoadjuvant chemotherapy
using gadolinium contrast agents [74, 156, 261], the high cost and invasiveness of these methods
render them impractical for frequent (e.g. daily) monitoring. Frequent monitoring of vascularity is
tion with hormone therapy has been confirmed by pathologic analysis [171].
5
The most effective clinical role for diffuse optical tomography (DOT) in the screening, di-
agnosis, and treatment monitoring of breast cancer has yet to be determined. However, it is
clear DOT provides exquisite functional information directly related to tumor patho-physiology
(e.g. metabolic activity, angiogenesis, and blood flow/concentration), and complementary to struc-
diffuse optical tomography of breast are critical for exploitation of the advances of molecular
imaging [190, 191, 277], an emerging field of medicine with promise of new generation optical
contrast agents.
1.2.1 Brief history of breast optical imaging: Before the Diffusion Approximation
Optical characterization of the breast has been attempted since 1929 [66] when the term diaphanog-
raphy was applied to shadowgraphs of breast tissue. The use of light to detect tumors in the breast
was first proposed by Cutler in 1929. Cutler hoped to distinguish between solid tumors and cysts in
the breast, but found it difficult to produce the necessary light intensity for diaphanography without
exposing the patient’s skin to extreme heat. Researchers in the 1930s and 1940s had little more
success, and the technique was temporarily abandoned. In the early 1980’s, however, essentially
the same class of transillumination measurement was renewed [16, 35, 108, 131, 178, 183, 209, 225,
247, 272, 275] using light in regions of low tissue absorption (i.e. 600 nm to 1300 nm). Unfortu-
nately the high degree of breast tissue scattering distorted spectroscopic information and blurred
optical images as a result of the large distribution of photon pathways through the tissue, and a
study in 1990 [4] suggested the method was still inferior to traditional methods of breast imaging.
Thus widebeam transillumination proved largely inadequate for clinical use, because it was too
6
difficult to separate the effects of absorption and scattering within the tissue, and because the two-
dimensional “photographic” data was poorly suited for image reconstruction. The mathematical
modeling of light transport in tissues had not been developed sufficiently for optical tomography
to be readily employed.
1.2.2 Brief history of breast optical imaging: After the Diffusion Approximation
As a result of numerous scientific and technological advances in tissue optics since 1990, optical
mammography now appears feasible with levels of specificity and resolution superior to early
work. The most critical advance in the field of photon migration has been the recognition and
widespread acceptance that light transport over long distances in tissues is well approximated as a
diffusive process [42, 290, 291]. Using this physical model it is possible to quantitatively separate
tissue scattering from tissue absorption, and to accurately incorporate the influence of boundaries,
such as the air-tissue interface, into the transport theory [203, 204]. Waves of diffuse light energy
density [113], or their time-domain analogs [18, 72, 137, 203] propagate deeply in tissues (e.g.
∼10 cm) and obey rules such as refraction [197], diffraction [27, 100], interference [238], and
dispersion [245, 260] as they encounter variations in tissue properties. The separation of scattering
and absorption afforded by the diffusion approximation has led to new, quantitative investigations
about the average concentrations of molecular species in a variety of highly scattering media [43,
For the purposes of breast imaging, the most important contribution of the diffusion approxi-
mation has been to provide a tractable mathematical basis for tomographic image reconstruction.
Tomographic methods were not employed in the early transillumination patient studies, and are
7
crucial for recovery of information about breast tumor shapes and optical properties. Several ap-
proaches have been developed for diffuse optical tomography (DOT) [7, 8, 10, 12]. These include:
transport equation based schemes [230, 232, 242],backprojection methods [53, 271], diffraction to-
mography in k-space and variants [46, 167, 180, 240], perturbation approaches [6, 11, 198, 241], the
Taylor series expansion approach [141–143,205,206,216], gradient based iterative techniques [13],
elliptic systems methods (ESM) [118, 155], truncated Newton schemes [233–236] and Bayesian
conditioning [89–91]. Other important and related theoretical advances include the development
and differential [192] methodologies, and advances in the use of a priori information [154, 165].
During the last few years, scientists in the field have begun to experimentally revisit the prob-
lem of optical mammography armed with these new ideas and approaches. Highlights of the first
wave of this experimental work, which did not carry out a rigorous 3D tomographical analysis, in-
clude: (1) Pilot human breast studies [102, 123, 210] employing point-to-point (tandem-geometry)
transillumination with diffuse photon density waves. This work used the phase of the diffusive
wave as a means to correct for edge effects in the optical mammography apparatus. The phase cor-
rection enhanced image contrast and demonstrated potential for improved optical mammography.
(2) Pilot transillumination (tandem-geometry) measurements of human breasts in the time domain
by European groups [47, 115–117, 214, 253, 258]; again, however, the analyses were approximate.
For example, in some cases optical properties were assigned by fitting to bulk Greens function
response curves [117], and in other cases non-linear perturbation theory [258] and random-walk
based theory [47] were used to improve upon the bulk approximation. (3) Pilot studies by Philips
Corp. using a 256×256 continuous wave (CW) instrument in the “pendant” geometry. Preliminary
results using modified back-projection and algebraic reconstruction technique (ART) algorithms
8
have been presented [54,132]. (4) Pilot studies using plane-wave illumination in the frequency do-
main to probe canine breast cancer differentiation following contrast agent injection. In this case,
back-reflected diffuse light was measured using an intensified CCD camera [229], thus attesting to
In the second wave of clinical research, experimenters have finally begun to exploit the full
possibilities of the tomographic approach with emphasis on clinical measurements. We focus here
primarily on the work of other researchers. Pogue and coworkers and Jiang and coworkers have em-
ployed a cylindrical geometry with rings of source-detector pairs wrapped around the breast. Pogue
et al. [70, 221, 222] have reported quantitative hemoglobin images of the female breast acquired
through model-based reconstruction of near-infrared data. Their approach only uses in-plane data
theless, their reconstructions reveal tumor/normal tissue contrast in total hemoglobin concentration
and StO2 [70]. Jiang et al [144] carried out a 3D study on one patient with CW light at 785 nm,
observing two invasive ductal carcinomas at different locations. Their studies with more patients
were essentially 2D reconstructions of absorption (µa ) and scattering (µ0s ) at 785 nm. Again, they
observed absorption contrast in the invasive carcinomas (N=5) and a fibroadenoma (N=1) [145];
they also found evidence for cysts (low µa and µ0s ) [119]. Barbour and coworkers carried out a
study in an essentially cylindrical (or conical) geometry [15]. They utilized three different pertur-
bations, one for each patient (i.e. deep breathing maneuver, breath holding and Valsalva maneuver),
to explore tumor response, and they found tumor contrast in oxygen supply/demand imbalance and
vasomotor response. Several groups have now begun incorporating optical techniques with other
imaging modalities. For example, Boas and coworkers [165] have co-registered DOT with x-ray
mammography in a ‘hard’ compression instrument. Thus far they have shown a priori spatial
9
information from a co-registered x-ray mammogram can constrain DOT regularization in recon-
struction of µa at 780 nm. This approach will be interesting in the future. There has also been
research, combining ultrasound with optical methods, e.g. Zhu/Yodh et al [299], Yodh/Tromberg
et al [130] and Zhu et al [297, 298, 300]. The DOT problem in this case is made more difficult by
the reflection geometry, and by poor definition of breast boundaries compared to other instruments
described above. Nevertheless, Zhu and coworkers have constructed 3D DOT total hemoglobin
concentration maps (assuming only absorption perturbation), identifying an in situ ductal carci-
noma (N=1) [297], and in other studies they have distinguished early-stage invasive carcinomas
(N=2) from benign lesions (N=17) [300]. Yodh, Chance, Schnall and coworkers [194, 195] have
combined time-domain DOT with MRI in a ‘soft’ parallel-plate compressed breast geometry that
Their most important contributions thus far are the direct verification of tumors by both techniques,
and the study of differential uptake of contrast agents in malignant tumors. More groups have be-
gun to incorporate DOT into MRI environment to utilize high spatial resolution information from
MRI to constrain DOT for better quantification [33]. Finally, 3D DOT fluorescence imaging in
phantoms has been accomplished [109]; this may be important for future studies with contrast
In addition to tomographic imaging, there has been significant spectroscopic research recently
on tumors and normal tissues providing more insight about their optically-derived properties [41,
140, 146, 202, 249, 251, 259] by correlating to physiological parameters such as body-mass-index,
age, breast thickness, radiographic density, paranchemal type, etc. Especially, intrasubject and in-
tersubject variability of those properties are reported, which emphasizes the importance of relative
measure of tumor to normal contrast of the same subject [212, 215, 246, 248].
10
The photon migration community, especially in breast cancer imaging, is at the critical point
in identifying tumor contrast. Photon migration breast cancer research is constantly evolving to-
wards (1) better quantification (3D imaging with multiple source detector positions), (2) extensive
[222, 278]), (3) co-registration with other imaging modalities and (4) therapy monitoring.
During the last decade our laboratory has made substantive contributions to problems in dif-
fuse optical tomography. We published the first experiments demonstrating tomographic recon-
struction of absorption and scattering heterogeneities [198], we were among the first groups to
show how to reconstruct the lifetime and concentration of fluorophores in tissues [199], and we
introduced a methodology whereby diffused temporal light field correlation functions are used
to reconstruct heterogeneous dynamical flow properties [24, 29]. In a different vein, we have
investigated the fundamental resolution and characterization limitations of diffuse optical tech-
niques [28, 166, 168], and we have combined the optical methods with other imaging modalities
such as ultrasound [130, 299] and MRI [192–195] so structural information from the second imag-
ing modality may be optimally coupled with the optical technique to improve the accuracy of as-
signed tissue optical properties [200]. We also continue to develop theoretical tools for DOT which
have, for example, elucidated the resolution and sensitivity trade-offs associated with specific ge-
ometries and instrumentation [65], enabled experimenters to choose optimal source wavelengths
for clinical experiments [58], and clarified the conditions for which differential measurements are
well suited for quantitative tomography [192]. Finally we have consistently pushed DOT towards
11
in vivo applications, for example, in brain hemodynamics and metabolism [63, 88], in tumor biol-
Most recently (and relevant for this thesis), we have successfully assembled a parallel-plate
soft-compression apparatus for diffuse optical tomography of breast, and we have begun clinical
studies with the instrument at the Hospital of the University of Pennsylvania (HUP) [49, 62, 81].
Our second generation parallel-plate instrumentation is unique in the community: our sources and
detectors are based predominantly on continuous-wave (CW) light, our CCD-based detection is
simple and massively parallel compared to all other instrumentation, we collect multiple views
of the breast in the device and so execute and produce full three-dimensional reconstructions,
and our soft compression parallel-plate geometry is attractive for comparison to MRI and x-ray
mammography as well as for maximizing hemodynamic contrast. Two sensible figures of merit
for DOT imaging systems are total data, and data rate (defined as a product of source-number times
rates exceed those of all existing instruments, in most cases by 100× (See Table 1.1).
to carry out 3D diffuse optical tomography with CW light. We showed that DOT reconstructions
efficiently and uniquely separate scattering from absorption, as well as the contributions of one
chromophore from another chromophore [57, 58]. Our clinical pilot studies show that tumors
are detectable and characterizable with the diffuse optical method (Chapter 4, Section 4.4); they
can even be tracked during neoadjuvant chemotherapy [49] (Chapter 4, Section 4.5). In light of
searching for additional optical contrast, we show that blood flow measurement by optical method
is feasible and show distinction among different breast tumors (Chapter 4, Section 4.6).
12
Group type Ns Nd Nλ Ntotal τ data rate(Hz)
this thesis (2nd gen.) CW 48 2.28×105 6 6.5×107 8.4 min 6.6×107
Schmitz et al [239] CW 25 32 4 3.2×103 0.32 s 8640
Colak et al [54] CW 225 225 3 1.5×105 6 min 540
Iftimia et al [134] CW 64 64 3 1.2×104 12 min 17
McBride et al [181] FD 16 15 × 3 6 4.3×103 30 s 42
Li et al [165] FD 40 9 1 3.6×102 1.5 min 4
Chen et al [44] FD 12 8 2 1.9×102 NA NA
Godavarty et al [109] FD 27 128 1 3.5×103 NA NA
Franceschini et al [102] FD 1 2340 2 4.6×103 3 min 2.6
Ntziachristoset al [194] TD 24 8 1 1.9×102 80 s 2.4
Cubeddu et al [258] TD 1 3000 4 1.2×104 5 min 40
Rinneberg et al [115] TD 1 2000 2 4.0×103 5 min 13
Schmidt et al [237] TD 32 32 1 1.0×103 ∼10 min 1
Table 1.1: Comparison of instruments used in the photon migration field in terms of information
content. Ns : number of sources, Nd : number of detectors, Nλ : number of wavelength, Ntotal :
total number of data (= Ns Nd Nλ ), τ : acquisition time. Data rate is defined as Ntotal /τ . (CW:
continuous-wave, FD: Frequency-domain, TD: Time-domain method)
Hypoxic-ischemic damage to the fetal brain can result in permanent neurodevelopmental impair-
ment or death [257, 263, 264]. Early detection of fetal cerebral hypoxic-ischemia is thus important
for timely intervention. Current non-invasive antepartum screening and diagnostic tools for fetal
well-being in utero include the non-stress test (fetal heart rate monitoring) and the biophysical pro-
file (fetal heart rate monitoring and ultrasound). Fetal heart rate monitoring, however, probes fetal
cerebral hemodynamics and oxygenation indirectly, and has a high false-positive rate [161, 173].
This has led to an increased number of unnecessary Cesarean sections and premature deliver-
ies [223]. Clearly, the development of devices to directly monitor fetal cerebral oxygenation and
hemodynamics could improve the specificity of antepartum tests. Near-infrared (NIR) diffuse
optical spectroscopy has the potential to non-invasively monitor fetal cerebral oxygenation and
hemodynamics in utero. NIR light is non-ionizing and the power levels used are harmless to the
13
body, making this technology safe under continuous exposure. Additionally, NIR technology can
A few researchers have developed and employed NIR oximeters to monitor neonatal cerebral
oxygenation [45,77,122,127,158]. More recently, a trans-vaginal NIR fetal oximeter [73,103] has
been developed, but can only be used during labor after the amniotic membrane has ruptured.
cessful could provide a direct assessment of fetal cerebral oxygenation and hemodynamics be fore
labor and delivery. The feasibility of transabdominal NIR continuous wave (CW) spectroscopy was
first explored by Ramanujam et al [227, 228] during a non-stress test. Subsequently, Zourabian et
al developed a transabdominal NIR CW oximeter with the capability to detect fetal arterial pulses
in utero [301]. In addition, theoretical and experimental tissue phantom investigations have been
performed to understand NIR photon diffusion through the fetal brain in utero [138, 270]. Col-
lectively, these studies suggest transabdominal NIR photon diffusion measurements through the
fetal brain in utero are possible. However, due to limitations of the instrumentation and analytical
models employed in these studies, it has not as yet been possible to quantify fetal cerebral blood
To demonstrate the feasibility of transabdominal NIR spectroscopy for detecting and quantify-
ing fetal hypoxia in utero, we designed a fetal hypoxia model using pregnant ewe. We have built
a multi-wavelength NIR frequency-domain instrument with the capability to perform NIR pho-
ton diffusion measurements through tissue over a wide range of source-detector separations and a
two-layer numerical diffusion model to accurately quantify the fetal cerebral blood saturation. We
have shown good agreement between fetal blood saturation determined by the transabdominal NIR
method, and arterial and venous fetal blood saturation quantified from fetal blood samples using a
14
hemoximeter (gold standard) [50] (Chapter 5, Section 5.2).
In this thesis, I describe the theory and the experimental DOS/DOT techniques used for breast
cancer imaging and fetal brain oxygenation monitoring. I will also present physiological results
for each application. Chapter 2 describes the theory behind DOS/DOT. Chapter 2, Section 2.2
outlines the general analysis structure common to DOS and DOT, and the remaining sections of
Chapter 2 go into details about DOS and DOT respectively. Section 2.4 describes and compares
Chapter 3 describes experimental techniques starting from design issues to validation with tis-
sue phantoms. Chapter 3, Section 3.2 briefly discusses specification parameters and signal-to-noise
characteristics of light source and detector components for the application. Detailed schematics of
the frequency-domain homodyne system and CCD-based continuous wave system are introduced
in the following sections respectively. In Section 3.5, the electronic and optical characterization
method of the instrument is outlined. The details of specific instruments used for the breast imag-
ing and fetal oximeter are summarized in Section 3.6. These instruments are further characterized
extensively using various tissue phantoms reflecting the specific requirement of the application, in
Section 3.7.
Chapter 4 explores various optical contrast of breast cancer using three-dimensional DOT im-
age reconstruction. The medical terminology regarding breast cancer diagnosis is briefly intro-
duced in the Section 4.2. Healthy breast optical properties are quantified in Section 4.3 to identify
variations due to intrinsic breast heterogeneity and dependence on demographic features. In Sec-
tion 4.4, three-dimensional reconstructed images of THC, StO 2 , and µ0s are presented. Particularly,
15
THC and µ0s show higher values in the tumor than in the surrounding tissue which correlate well
with other imaging modalities such as X-ray mammography, ultrasound or MRI. In Section 4.5,
the utility of DOT to monitor neoadjuvant chemotherapy is demonstrated by comparing with the
MRI. The potential enhancement of the breast cancer contrast based on blood flow measurement
Chapter 5 explores the feasibility of quantitative fetal oximetry in utero. In Section 5.2, the
accurate quantification of NIR fetal oximetry using two-layer model is demonstrated in an induced
fetal hypoxia model using pregnant sheep. In Section 5.3, a preliminary study on human in a
Cesarean section scenario is discussed for issues regarding translating the technology to clinical
situations. In the following section, based on these issues, a method to optimize the instrument for
16
Chapter 2
Theory
2.1 Introduction
The propagation of photons through turbid media such as tissue is often first described by the
Boltzman transport equation [37, 67]. Generally, solutions of the Boltzman equation are computa-
tionally intensive and time consuming. Fortunately, for the most DOT applications the Boltzman
transport equation is well approximated by the photon diffusion equation. The derivation of pho-
ton diffusion equation from the transport equation is given clearly in reference [22]. We adopt
the photon diffuse equation as our starting point. The photon diffusion equation has the following
form:
1 ∂Φ(r, t)
∇ · (D(r)∇Φ(r, t)) − µa (r)Φ(r, t) − = −S(r, t). (2.1)
v ∂t
Here r is the position vector, t is time and v is the speed of light in the medium [cm/s]. D(r) ∼
=
1
3µ0s (r) [87] is the photon diffusion coefficient, µ0s (r) [cm−1 ] is the reduced scattering coefficient,
and µa (r) [cm−1 ] is the absorption coefficient. Φ(r, t) is the photon fluence rate [Watt/cm 2 ] and
17
The behavior of Φ(r, t) near the boundary of the turbid medium is described by partial current
boundary conditions (i.e. mixed Dirichlet-Neuman boundary condition) [121, 150, 159, 296],
∂Φ(r)
= −αΦ(r), (2.2)
∂ n̂
where n̂ is the vector normal to measurement boundary, α is related to the refractive index mis-
³ ´
1−Ref f 3µ0s −1.44 0.71
match at boundary via the following: α = 1+Ref f 2 , Ref f ∼ n2 + n + 0.668 + 0.63n
nin
and n = nout [114, 150]. Typically, the source is modeled as a single isotropic point source placed
CW FD TD
Intensity
Intensity
Intensity
Input
Output
Figure 2.1: Measurement approaches: (a) Continuous-wave, (b) Frequency-domain, (c) Time-
domain measurement types (solid line: input light source, dashed line: output detected signal)
Three types of measurements are used widely in the community: continuous-wave (CW),
frequency-domain (FD), and time-domain (TD) measurements. Figure 2.1 illustrates the input light
source (solid line) and the output signal (dotted line) for each measurement type. Continuous-wave
measurements employ a light source whose intensity does not vary with time; the detector mea-
sures the transmitted intensity which is affected by the medium. Frequency-domain measurements
18
employ a light source which is amplitude modulated in the radio-frequency (RF) range. The de-
tector measures the amplitude of the transmitted diffuse photon density wave and its phase shift
relative to the input. Time-domain measurements employ a short input light pulse (i.e. typically
less than 1 nanosecond) and detect a delayed and temporally broadened output pulse.
Equation 2.1 further reduces to the following equation for the FD case.
µ ¶
iω
FD : ∇ · (D(r)∇Φ(r)) − µa (r) − Φ(r) = −S(r). (2.3)
v
In the FD case, we have assumed Φ(r, t) = Φ(r)e−iωt , and the e−iωt terms have factored
out, since the detected signals are modulated at the same frequency as the light source (ω: light
source modulation frequency). For the CW case, Equation 2.3 simplifies since ω = 0. Equations
2.1 through 2.3 are used to compute Φ(r), given µa (r) and µ0s (r). This is often referred to as the
forward problem.
The relative placement of light sources and detectors define measurement geometry. In a par-
allel plate design similar to X-ray mammography or MRI, in vivo tissue lies between two planes
where light sources residing in one plane and detectors residing in the other plane. This geometry
is called as the trasmission geometry. However, most organs are not accessible to transmission
geometry due to signal-to-noise issue. In this case, remission geometry is utilized where the light
sources and detectors reside in the same plane (i.e. at the surface of the tissue). Multiple scattering
enables photons to deviate from the straight path and reach the detectors placed in the same plane.
Cylindrial geometry is also used for mostly breast and sometimes neonate brain. This geometry
can be thought as the mixture of remission and transmission depending on the relative position of
19
Using Equation 2.1, it is usually desirable to compute optical properties from measurements
of Φ(r) on the tissue surface. This is called the inverse problem. The inverse problem can be
formulated in various ways depending on the application and measurement design. Diffuse optical
spectrscopy (DOS) treats the reconstruction space as either homogeneous or composed of a limited
number of piecewise homogeneous regions. Therefore, the number of unknowns (i.e. the bulk op-
tical properties of each region) is usually much less than the number of measurements. For Diffuse
optical tomography (DOT), the reconstruction space is discretized into large number of volume
elements or voxels. The optical properties of each voxel are the unknowns to be reconstructed.
To deal with the heterogeneous response of light sources and detectors (which we will refer
to as source-detector coupling coefficients), two different approaches are often employed. Differ-
ential reconstruction utilizes a reference measurement to normalize out the variations of source
data before some physiological perturbation. In other cases, a reference measurement is made on
a tissue phantom with well-known optical properties. In an absolute reconstruction, the coupling
When defining the unknowns to reconstruct, it is also desirable to impose additional spectral
constraints based on known properties of physiological tissues. Spectral constraints on the sample
absorption arise from the well-known spectrum of each chromophore. The absorption coefficient
L
X
µa (λ, r) = ²l (λ)Cl (r). (2.4)
l=1
Here ²l (λ) is the extinction coefficient, Cl (r) the concentration of the l th chromophore and L
20
is the total number of chromophores. In physiological case, major chromophores are oxygenated
hemoglobin (HbO2 ), deoxygenated hemoglobin (Hb), water (H2 O) and lipid. The extinction coeffi-
cients, ²Hb , ²HbO2 , ²H2 O , ²lipid only depend on wavelength λ and has the dimension of [cm − 1/µM].
The extinction coefficients are well documented in literature and can be found in the Appendix Sec-
tion 3.8.1 at the end of Chapter 3. The concentrations of these chromophores C Hb , CHbO2 , CH2 O ,
Clipid are given in [µM]. The scattering constraint comes from the observation that scattering in
tissue follows a simplified Mie-scattering approximation [185, 188] reasonably well, i.e.
where A(r) is called the scattering prefactor and b(r) is called the scattering power (and is related
to the particle size and density, index of refraction of scatterers and the medium). In conventional
schemes (which we will denote as the single-spectral method), absorption (µ a (r)) and scattering
coefficients (µ0s (r)) at each wavelength are first determined, and then the resulting absorption co-
efficients are used to extract chromophore concentrations using Equation 2.4. In the multi-spectral
method (or the a priori spectral approach), the chromophore concentrations and scattering proper-
ties are reconstructed directly in a single step using data at all wavelengths with spectral constraints
(Equation 2.4 and 2.5). Since the chromophore concentrations and the scattering amplitudes and
prefactors are treated as unknowns instead of the optical coefficients at each wavelength, the num-
ber of unknowns is reduced significantly. The amount of computer memory increases with multi-
spectral method. However, it depends on the reconstruction scheme; the Jacobian based approach
requires a significant memory, whereas the nonlinear conjugate gradient based approach does not
21
require significant memory. Moreover, Durduran [80] has shown that multi-spectral approach re-
duces the detrimental effects of interparameter crosstalk and misestimation of sample geometry.
be derived, i.e. the total hemoglobin concentration T HC = C Hb + CHbO2 and the tissue blood
In the following section, DOS and DOT are further developed for each of the clinical applica-
tions to be presented. First, a general analysis scheme common to the DOS and DOT is presented
in a flow chart style in Section 2.2. Following the general format, individual DOS algorithms for
various measurement configurations are presented. Then two different reconstruction schemes de-
veloped for DOT breast cancer imaging in the framework of single-spectral and in multi-spectral
The goal of diffuse optical tomography (DOT) is to accurately estimate the distribution of opti-
cal properties in a tissue volume from non-invasive optical measurements on the surface of the
medium. Various algorithms can be utilized to estimate optical properties. These algorithms dif-
fer depending on the choice of photon propagation model, measurement type and geometry, and
the single-spectral method or the multi-spectral method, x can be the optical properties (µ a and
µ0s ) or the chromophore concentrations and scattering factors (C l , A, and b), respectively. The ini-
tialization process consists of reading in the measurement data, defining the reconstruction space,
22
Begin
Initialization
Calculate χ2
Yes
Criteria? End
No
Solve inverse problem
Update x + ∆x → x
Figure 2.2: General analysis flow chart for model-based optical properties reconstruction.
and assigning initial guess for x(r). r denotes position within the sample volume. The forward
problem computes the fluence rate, Φ(r), on the sample surface given light source information
and optical property distribution x(r). χ2 quantifies the discrepancy between the calculated and
measured fluence rate; its value determines whether to update x(r) and iterate again or stop the
calculation. If the stopping criteria is not met, the inverse problem estimates ∆x for the next it-
eration. After updating x, the process is iterated until the stopping criteria for χ 2 is met. In the
2.2.1 Initialization
At this first stage, measurement data are read in. Each piece of measurement data is associated with
23
Nd (s) is the number of detectors associated with the sth source which may be different per source,
since the measurement data is selected to be above the noise level. Usually these vectors are located
on the surface of the medium. The measured fluence rate of the sample, Φ m (rs , rd ), and that of
reference measurement, ΦR
m (rs , rd ), are recorded.
The three-dimensional medium is discretized with nodes or volume elements centered on, r k
(k = 1, . . . Nv ). For the initialization purpose, an initial µ-map (in particular, µ a (rk ) and µ0s (rk ))
of the sample and reference needs to be assigned. Then calculation of sample fluence rate Φ c (rk )
subdivided into analytic solutions and numerical solutions. The analytic solutions are available
for simple geometries such as infinite, semi-infinite, slab, cylinder, and embedded spheres in those
geometries [6, 203, 217]. The numerical solutions are available via the finite difference method
(FDM) or the finite element method (FEM). Numerical approaches are critical for complicated
2.2.3 Calculation of χ2
χ2 describes the discrepancy between measured and calculated fluence rates. The specific form of
χ2 may differ depending on whether the reconstruction scheme is based on differential or absolute
reconstruction, and on whether the reconstruction uses Born or Rytov type [196] approaches. (For
24
2.2.4 Inverse problem
The inverse problem can be thought as the minimization of χ 2 problem with respect to unknown
x. The solution for the DOT inverse problem can vary depending on the methods employed:
backprojection method [53, 271], analytic method [46, 167, 180, 240], linear method [196, 198]
and nonlinear method [13] (See Reference [7, 8, 107] for reviews). In the following sections, two
different types of nonlinear DOT methods which have been developed in our laboratory, will be
outlined.
2.3 Spectroscopy
In our laboratory, frequency-domain, multi-spectral diffuse optical spectroscopy is widely used for
the quantification of bulk optical properties of the medium. At the initialization step, the mea-
sured fluence rate Φm (λ, r) is recorded and the initial guess for the unknowns assigned. For the
frequency-domain measurement, the measurable quantities at each wavelength and detector posi-
tion are the amplitude (Am ) and the phase shift (θm ) with respect to the input source. The measured
where ξs and θs are the amplitude and phase of the source coupling and ξ d and θd are the amplitude
and phase of the detector coupling. In the multi-spectral method, the unknowns are chromophore
concentrations and the scattering factors. For instance, when the optical properties of the homo-
geneous liquid phantom made with India ink and Intralipid (as described in Chapter 3, Section
3.8.1) are to be recovered, the unknowns are Cink , A and b. When the bulk optical properties of
25
physiological tissue (with the assumption of homogeneous medium) is under investigation, the un-
knowns are CHb , CHbO2 , CH2 O , Clipid , A and b. Since the calculated fluence rate Φc (λ, r) is often
described in terms of µa and µ0s , one needs to decompose the initial guess using Equation 2.4 to
solve the forward problem. (See Appendix Section 3.8.1 for extinction coefficients.)
negative image
z=-z p
z 0+ z b detector fiber
source fiber
z=-z b extrapolated
zb n out boundary
z0 z=0
n in
positive source
Figure 2.3: Source and image configurations for extrapolated boundary condition.
The analytic solutions of the forward problem for semi-infinite and slab geometry with extrap-
olated boundary conditions are widely used in most DOS analyses. For the extrapolated boundary
condition [5, 76, 94, 121], the fluence rate is set to zero at an extrpolated boundary located at a
distance zb outside the medium i.e. Φ(ρ, z = −zb ) = 0. The method of images depicted in Figure
2.3 can be used to construct the solution by placing a negative image source at the opposite side
of the extrapolated boundary. In this extrapolated boundary case, the position of negative image
source is at −zp = −(z0 + zb ) since the source is placed at z0 (1/µ0s inside the medium). The
configuration of source and image for the semi-infinite and slab geometry is depicted in Figure 2.4.
The analytic frequency-domain solution for the semi-infinite geometry (Figure 2.4(a)) with
26
z=-3d
z=-2d
z=-z p
z z=-d
z=0 z=0
z=z 0 source
source z=d
z=2d
z=3d
z=4d
positive image
(a) negative image (b)
Figure 2.4: Source and image configurations for measurement Geometry of (a) Semi-infinite
medium and (b) Slab.
à p p !
vS exp(ik ρ2 + (z − z0 )2 exp(ik ρ2 + (z + zp )2
Φ(ρ, z) = p − p , (2.7)
4πD ρ2 + (z − z0 )2 ρ2 + (z + zp )2
0.668 + 0.064n. z is the axis perpendicular to the medium, and ρ is the radial distance parallel
to the medium, as can be seen in Figure 2.3 and Figure 2.4(a). The filled circle represents the
1
light source displaced µ0s inside the medium [203] and the open circle is the imaginary negative
source displaced an equal distance (zb ) away from the extrapolated boundary where the fluence
rate becomes zero. This solution is used in Chapter 3, Section 3.7 for bulk optical properties
assessment of tissue phantom, in Chapter 4, Section 4.3 in calculating bulk optical properties of
healthy breasts, and in Chapter 5, Section 5.3 for preliminary assessment of clinical data.
The analytic frequency-domain solution for the slab geometry (Figure 2.4(b)) with extrapolated
27
boundary conditions is [55, 217]
m=∞ m=∞
à p p !
vS X exp(ik ρ2 + (z − z+,m )2 X exp(ik ρ2 + (z − z−,m )2
Φ(ρ, z) = p − p ,
4πD m=−∞ ρ2 + (z − z+,m )2 m=−∞ ρ2 + (z − z−,m )2
(2.8)
where z+,m = 2m(d + 2zb ) + z0 , z−,m = 2m(d + 2zb ) − 2zb − z0 for m = 0, ±1, ±2, ..., and d is
the slab thickness. The slab solution is constructed with a series of imaginary sources arising from
two air-tissue boundaries (Figure 2.4(b)). In practice, we use up to 5th term for m. The remaining
terms are usually quite small. This solution is used for analysis in Chapter 4, Section 4.3.
The presence of source detector coupling coefficients in the multiple source detector configu-
ration requires attention: differential and absolute approaches will be described in the following
sections. These approaches have distinct forms of χ2 . Once χ2 is defined, various optimiza-
tion functions can be used to find the combination of unknowns which minimize χ 2 . Usually the
Nelder-Mead simplex method [189, 224] is utilized for our spectroscopy approaches.
When perturbative physiological measurements (Chapter 5, Section 5.2) are possible, the source
coupling coefficient problem becomes considerably simpler. Since the probe is fixed in the same
position under the same condition throughout the perturbation, the source detector coupling coef-
Nλ X Nd ¯
Ns X ¯2
2
X ¯ Φm (λw , rs , rd ) Φc (λ w , r s , r d )) ¯
χ = ¯ ΦR (λw , rs , rd )) − ΦR (λw , rs , rd )) ¯ ,
¯ ¯ (2.9)
m
w=1 s=1 d=1 c
where Nλ , Ns , and Nd are the number of wavelengths, sources and detectors respectively.
28
2.3.2 Absolute approach
Where there is no reference measurement, one needs to explicitly consider the effect of source
and detector coupling coefficients. In particular, the retrieval of optical properties of the matching
fluid of the 2nd generation parallel-plate DOT instrument (described in Chapter 3, Section 3.6.2)
relies on the absolute approach. Raw data of amplitude and phase shift is shown with respect to
the source and detector separations in Figure 2.5. The presence of source and detector coupling
coefficients leads to significant deviation of data points from the expected amplitude and phase
(normalized respctively to the measured values) for a homogeneous medium. This deviation does
not arise for the one source and one scanning detector configuration (Chapter 3, Section 3.6.1).
0 2
−1
log(r2*Amplitude)
1.5
Phase (rad)
−2
1
−3
0.5
−4
−5 0
−6 −0.5
1 2 3 4 5 6 1 2 3
ρ (cm) ρ (cm)4 5 6
(a) (b)
Figure 2.5: Effect of coupling coefficients on measured amplitude and phase of homogeneous
medium. (a) Amplitude and (b) Phase plotted versus source detector separations. Amplitude and
phase are measured on a homogeneous matching medium using frequeny-domain, multiple source
detector instrument (Chapter 3, Section 3.6.2). Significant deviation from semi-infinite solution
(solid line) shows the effect of the coupling coefficients on data.
An implicit way to estimate the source and detector coupling coefficients while fitting for the
chromophore concentration and scattering factors was developed. In this method, we defined χ 2
29
to be normalized with the measured fluence rate,
Nλ X Nd ¯
Ns X
¯ Φm − Φc ¯2
X ¯
2
χ = ¯ Φm ¯ .
¯ ¯ (2.10)
w=1 s=1 d=1
We use Nelder-Mead simplex method to update the unknowns. At each iteration, when C l , A
and b are updated, they are decomposed into µa (λ) and µ0s (λ) by Equation 2.4. For these µa (λ)
and µ0s (λ), coupling coefficients are estimated in the following two-step process. First, detector
Ns
1 X Ac (µa , µ0s )
ξd (µa , µ0s ) = (2.11)
Ns Am
s=1
Ns
0 1 X
θd (µa , µs ) = {θc (µa , µ0s ) − θm }, (2.12)
Ns
s=1
where the calculated amplitude is Ac = |Φc | and the calculated phase is θc = arg(Φc ). Then,
source coupling coefficients are estimated by summing the ratio of adjusted fluence rate over all
the detectors.
Nd
1 X Ac (µa , µ0s )
ξs (µa , µ0s ) = (2.13)
Nd ξd · A m
d=1
Nd
1 X
θs (µa , µ0s ) = {θc (µa , µ0s ) − θm − θd }. (2.14)
Nd
d=1
These source detector coupling coefficients depend on µ a (λ) and µ0s (λ) through Φc , which depends
on the updated unknowns at each iteration. The iteration continues until χ 2 reaches the stopping
criterion.
30
2.4 Imaging
Two different imaging software packages were developed and employed in our laboratory for
breast cancer imaging. The first software, which will be referred to as “Nonlinear Rytov Itera-
tive Method (NRIM)”, was developed by Dr. Monica Holboke [128, 129]. NRIM has mainly been
used for analyzing the tissue phantom measurements to characterize DOT instrumentations [62].
For clinical data analysis, use of NRIM turned out to be challenging mainly because of its exces-
sive computational memory requirements; NRIM computes the Jacobian (weight matrix) explic-
itly. Another approach, TOAST (Time-resolved Optical Absorption and Scattering Tomography)
developed originally by Arridge et al [13] is based on the nonlinear conjugate gradient method;
it has significant memory overhead reduction compared to NRIM. Therefore a collaboration with
Dr. Simon Arridge’s group was initiated. The CW version of TOAST was adapted to our imag-
ing geometry and further developed at UPENN to include the multi-spectral approach [57, 58].
In the following section, the details of NRIM, the single-spectral TOAST (original version),
and the multi-spectral TOAST (MTOAST) are described. Explicit forms of equations are presented
NRIM [128, 129] was first developed using a single-spectral approach. The unknown variables to
fit are the absorption coefficient µa and the reduced scattering coefficient µ0s . Furthermore, this
code was customized for breast imaging wherein we take reference measurement of matching fluid
(Intralipid/India ink). It is also suitable for differential measurement before and after contrast agent
31
injection. The following flow chart (Figure 2.6) shows more of the details for the inverse problem
Begin
Initialization
Calculate Φ(x)
Calculate χ2
Yes
criteria? End
No
Calculate G(rd , rk )
Set up Jacobian J
Solve J∆x = y
Update x + ∆x → x
Figure 2.6: Nonlinear Rytov Iterative Method (NRIM) analysis flow chart
In the initialization step, the 3D space is discretized into an n x × ny × nz Cartesian grid (with
position vector given as rk ). There is an option to read in an initial guess for µa (rk ) and D(rk ) as
Using a finite difference method [160], Equation 2.3 in combination with the partial current
boundary condition (Equation 2.2) becomes a matrix equation. For given µ a (rk ) and D(rk ),
Φ(rk ) is calculated using the preconditioned conjugate gradient method (linear conjugate gradient
method, to be described later in this section). This forward problem is solved for each source.
32
Using the reference µ-map, the reference calculated fluence rate Φ R
c is determined. For the
0th iteration, Φ is calculated using the initial guess µ-map. For the following t th iterations, Φc is
The perturbation due to the heterogeneous variation of optical properties from the background
linear superposition of the incident (background, Φ0 ) and scattered (heterogeneous, Φsc ) diffusive
waves : Φ(r, rs ) = Φ0 (r, rs ) + Φsc (r, rs ). In the Rytov approximation, the contributions from
the incident and scattered parts are expressed in exponential fashion : Φ(r, r s ) = exp[φ0 (r, rs ) +
φsc (r, rs )]. This leads to the derivation of Equation 2.16, which provides the update scheme for
Nd · µ
Ns X ¶¸2
2 1X Φm (rs , rd ) ΦR
c (rs , rd )
χ = ln (2.15)
2 ΦR
m (rs , rd ) Φc (rs , rd )
s=1 d=1
Here, Φm is the measured fluence rate and Φc is the calculated fluence rate based on the forward
problem. ΦR R
m and Φc are the measured and calculated fluence for reference measurements.
|χ2t −χ2t−1 |
The stopping criteria is defined as χ2t−1
< STOP CRITERION, where χ2t denotes χ2 at
Using the diffusion equation, Green’s theorem and Rytov expansion [6, 11, 198, 241, 289], one
can show
Φc (rs , rd )
Φm (rs , rd ) ΦR
µ ¶
c (rs , rd )
= −ln (2.16)
ΦRm (rs , rd ) Φc (rs , rd )
33
where the Green’s function is the solution to
µ ¶
iω
∇ · (D(r)∇G(r)) − µa (r) − G(r) = −δ(rd , r), (2.17)
v
for the sample medium, where ∆µa (r) = µa (r) − µ0a (r) and ∆D(r) = D(r) − D 0 (r) with
The discretization of the integrals for dth detector and sth source leads to the following;
Since there is one such equation for each source and detector pair, a matrix equation for all N sd
PN s
(total number of source and detector pairs, i.e. Nsd = s=1 Nd (s)) is
∆µa (r1 )
³ R
´
m Φc
∆µa (r2 ) −ln Φ
ΦRm Φc
s=1,d=1
..
³ ´
GΦ|r1 GΦ|r2 ∇G·∇Φ|r1 ∇G·∇Φ|r2 Φ ΦR
Φ Φ ··· Φ Φ · · · . −ln R Φ m c
Φm c s=1,d=2
=
.. .. .. ..
∆D(r1 ) ..
. . . . .
.. ³
Φ ΦR
´
. −ln ΦR Φc
m c
s=Ns ,d=Nd
m
∆D(rN )
(2.19)
Jx = y (2.20)
34
As can be seen from Equation 2.19, explicitly built-up of Jacobian J is of size 2 × N sd × Nv . This
such as
Detection plane at y = Y
Y
X
Figure 2.7: Reconstruction geometry for breast imaging (Chapter 3, Section 3.6.2).
For our breast cancer imaging case (Chapter 4), we used a spatially variant regularization
" õ ¶2 ! õ ¶2 µ ¶2 ! #
2(yk − yc ) xk − x c zk − z c
γ(rk ) = lc + le exp + exp + −2 (2.22)
Y X Z
2.7) and setting C = I. lc and le are the regularization parameters corresponding to the center and
edge of the image respectively. Note that y is the axis perpendicular to the source and detection
35
planes.
We solve the matrix equation 2.21 using the linear conjugate gradient method. The linear
conjugate gradient method (CGM) is an iterative method to solve linear systems with positive
definite coefficient matrices [189]. Let W = JT J + γCT C and v = JT y to simplfy the Equation
Given x0 ;
Set r0 ← W x0 − v, p0 ← −r0 , t ← 0;
while rt 6= 0
rtT rt
αt ←
pTt W pt
xt+1 ← xt + αt pt
rt+1 ← rt + αt W pt
T r
rt+1 t+1
βt+1 ← T
rt rt
t←t+1
end
for updating xt . βt is the scalar for determining next conjugate direction given by conjugacy
36
property, based on residuals.
TOAST (Time-resolved Optical Absorption and Scattering Tomography) is a DOT software pack-
age developed by Arridge et al [13]. The CW version of TOAST is employed for our breast cancer
DOT application. The flow chart below (Figure 2.8) depicts the outline of the algorithm with
Begin
Initialization
Calculate Φ(x)
Calculate χ2
Yes
criteria? End
No
Calculate η(rs , rk )
∂χ2
Calculate ∂x
Update x + ∆x → x
The initialization step is similar to the previously described Nonlinear Rytov Iterative Method.
An extra feature in the initialization of TOAST is the ability to assign a heterogeneous distribution
37
for the refractive index for rk . The forward problem is solved using the finite element method [11].
|χ2t −χ2t−1 |
The χ2 is defined as in Equation 2.15. The stopping criteria is defined as (χt +χ2t−1 )/2+²
2 ≤
STOP CRITERION, where ² is 1 × 10−10 and STOP CRITERION is a user-defined value (typ-
· ¸T
−5
ically 1 × 10 ). Instead of forming an explicit Jacobian, TOAST solves for ∆µa ∆D by
using the gradient of χ2 as the search direction in nonlinear conjugate gradient method.
Nonlinear conjugate gradient method is used for large-scale nonlinear optimization problem
[189]. The Polak-Ribière nonlinear CGM to solve a general nonlinear function f (x) is as follows.
Given x0
while ∇ft 6= 0
Set xt+1 ← xt + αt pt
Evaluate ∇ft+1
T (∇f
∇ft+1
PR t+1 − ∇fk )
βt+1 ←
||∇fk | |2
PR
pt+1 ← −∇ft+1 + βt+1 pt
t←t+1
end
Particularly in our case, f (x) = χ2 (x). The gradients of χ2 (∇f (x)) with respect to the
38
unknowns µa and D needed in nonlinear CGM are given by
Nd
Ns X
∂χ2 ¯¯ Φm (rs , rd ) ΦR
¯ µ ¶
c (rs , rd ) 1
X
= ln R
G(rd , rk )Φc (rk , rs ) (2.23)
∂µa rk
¯ Φm (rs , rd ) Φc (rs , rd ) Φc (rs , rd )
s=1 d=1
Nd
Ns X
∂χ2 ¯¯ Φm (rs , rd ) ΦR
¯ µ ¶
c (rs , rd ) 1
X
= ln R
∇G(rd , rk ) · ∇Φc (rk , rs ) (2.24)
∂D rk¯ Φm (rs , rd ) Φc (rs , rd ) Φc (rs , rd )
s=1 d=1
depend on the Green’s function G evaluated at each voxel r k for dth detector and Φc evaluated at
each voxel for sth source. In order to calculate the gradients, one needs to solve the Equation 2.3
for Ns times and Equation 2.17 for Nd times. Note that the use of gradients instead of explicit
Jacobian results in significant reduction of computer memory compared to the Jacobian case.
However, a change of variables can reduce the number of calculation and speed up the recon-
³ ´
Φm (rs ,rd ) ΦR
c (rs ,rd )
struction. One can modify the Equation 2.17 by multiplying N 1
P d
d=1 ln ΦR (rs ,r ) Φc (rs ,rd ) Φc (rs ,rd )
m d
Nd
Φm (rs , rd ) ΦR
µ ¶ · µ ¶ ¸
c (rs , rd ) 1 iω
X
ln ∇ · (D(r k )∇G(r ,
k dr )) − µ (r
a k ) − G(r ,
k dr )
ΦR
m (rs , rd ) Φc (rs , rd ) Φc (rs , rd ) v
d=1
Nd
Φm (rs , rd ) ΦR
µ ¶
c (rs , rd ) 1
X
=− ln R
δ(rk , rd ) (2.25)
Φm (rs , rd ) Φc (rs , rd ) Φc (rs , rd )
d=1
39
If one define the following variables which sums over all detectors per source,
Nd
Φm (rs , rd ) ΦR
µ ¶
c (rs , rd ) 1
X
η(rs , rk ) = ln R
G(rk , rd ) (2.26)
Φm (rs , rd ) Φc (rs , rd ) Φc (rs , rd )
d=1
Nd
Φm (rs , rd ) ΦR
µ ¶
c (rs , rd ) 1
X
ν(rs , rk ) = ln R
δ(rk , rd ) (2.27)
Φm (rs , rd ) Φc (rs , rd ) Φc (rs , rd )
d=1
µ ¶
iω
∇ · (D(rk )∇η(rs , rk )) − µa (rk ) − η(rs , rk ) = −ν(rs , rk ) (2.28)
v
The gradients are then expressed in terms of new variable η(r s , rk ) which depends on rs .
Ns
∂χ2 ¯¯
¯ X
= η(rs , rk )Φ(rs , rk ) (2.29)
∂µa ¯rk
s=1
Ns
∂χ2 ¯¯
¯ X
= ∇η(rs , rk ) · ∇Φ(rs , rk ) (2.30)
∂D ¯rk
s=1
One only needs to solve Equation 2.28 Ns times in order to calculate gradients as opposed to
In summary, nonlinear CGM updates ∆x (x = µa , µ0s ) for our application as follows. For
given initial values for x, χ2 (Equation 2.15) and the gradients of χ2 are evaluated (Equation 2.30).
The initial search direction is given by the ∇χ2 (x0 ). Once the initial steps were taken (including
the calculation of gradients), the factor α which minimizes χ 2 along the direction x + αp is found
by inexact line search one-dimensionally. Then unknowns x are updated by the search direction p
scaled by α. For the calculation of the next search direction, ∇χ 2 is evaluated to compute scaling
40
2.4.3 MTOAST : multi-spectral TOAST
Several new techniques have been implemented into TOAST by us: (1) multi-spectral method, (2)
envelope-guided spatially variant regularization, (3) source detector coupling fit, and (4) geometric
In diffuse optical community, quantitative imaging using continuous-wave (CW) method has
been controversial because of nonuniqueness of CW solution to the photon diffuse equation [9].
Since there is no unique solution, crosstalk between absorption and scattering is quite significant,
which in turn affects the accuracy of derived physiological parameters. Yet the CW approach offers
several advantages over frequency-domain or time-domain approaches: good signal to noise, sim-
plicity of technology, low cost per detection channel, and (most importantly for us) being readily
adaptable for lens-coupled CCD data acquisition to yield high spatial information for reliable 3D
reconstruction. Some research groups have devised a scaling approach [208, 287] to circumvent
Corlu et al has shown that a multi-spectral approach helps to overcome the nonuniqueness
Conjugate Gradient inverse algorithm (TOAST) is well-documented in the reference [57] with the
The distinct differences between single-spectral and multi-spectral TOAST are (1) Definition
of unknown, (2) χ2 form, and (3) gradients of χ2 with respect to unknowns. The unknowns are
tion over the wavelength, and the spectral constraint governs the optical properties µ a (λ, r) and
µ0s (λ, r) which changes Φ(λw , rs , rd ). The multi-spectral χ2 is further modified to include the
41
envelope-guided spatial variant regularization term [56],
Nλ X Nd µ µ
Ns X ¶¶2
2 1X Φm (λw , rs , rd ) ΦR
c (λw , rs , rd )
χ = ln +Q (2.31)
2 ΦR
m (λw , rs , rd ) Φc (λw , rs , rd )
w=1 s=1 d=1
PN
where Q = k=1 γ(rk )||µ(rk ) − µ0 (rk )||2 , γ(rk ) is given by Equation 2.22 and µ stands for
∂χ2 ∂χ2
where ∂µa is given by Equation 2.23 and ∂µ0s by
Nd
Ns X
∂χ2 ¯¯ Φm (rs , rd ) ΦR
¯ µ ¶
c (rs , rd ) 1
X
2
= 3D ln ∇G(rd , rk ) · ∇Φc (rk , rs ).
∂µ0s ¯rk ΦR (r
m s d , r ) Φ (r
c s d, r ) Φc s , rd )
(r
s=1 d=1
(2.35)
For breast imaging, a geometric constraint is utilized to give a good initial guess for the re-
construction. This additional technique only involves the changes in the input field to MTOAST,
and not the internal algorithmic changes. The gist of the geometric constraint is to fix the optical
properties of matching fluid based on the frequency-domain measurements and not to allow the
matching fluid region to update its optical properties. Only the breast region is allowed to update
its multi-spectral unknowns such as CHb , CHbO2 , CH2 O , and A. The specific methodology used
for our breast imaging application is described in Chapter 4, Section 4.4.2 in detail.
42
Chapter 3
Experimental Techniques
3.1 Introduction
In this chapter, experimental techniques are outlined. The choice of light sources, detectors and
electronic components depends on the physiological application. First the light sources and detec-
tors used in the DOS/DOT field are introduced. Then, the frequency-domain homodyne system
which is used in all experimental applications in this thesis is described. Specification of the
CCD-based instrument is described in the separate section. In the next section, the methods to
test electronical components and to characterize the whole system for dynamic range are outlined.
Then the specific instrumentation used for each application is described in detail. For breast cancer
imaging, there were two parallel plate instruments: Frequency-domain scanning system and CCD-
DOS instrument was developed. Validation and characterization of the system as a whole was
43
3.2 Components
Three types of light sources are used for DOS and DOT applications : white light, light emitting
White light sources have gained more attention recently, because the importance of multiple
wavelength schemes has been recognized [19, 273]. The spectrum of most white light sources
extends over the visible range and into the near-infrared. Tungsten incandescent lamps radiate light
by heating tungsten filaments. Halogen lamps are essentially efficient high performance tungsten
lamps, with iodine or bromine in the fill gas to return evaporated tungsten to the filament. Arc
discharge lamps operate by ionizing xenon or mercury gas with a short high-voltage pulse and then
allowing the capacitor to discharge through the now-conductive gas. Typically white light sources
In light emitting diodes, free electrons moving across a diode junction combine with holes
from the P layer. In this process, photons are generated. The photon energy is set by the energy
In laser diodes (or diode lasers), a population inversion is induced in the P-N junction region.
Stimulated emission results from this population inversion and a net light amplification and lasing
is achieved by an optical cavity formed with the reflective coatings at opposite ends of the crystal.
Laser diodes are used widely for DOT applications. They are highly monochromatic and respond
44
3.2.2 Detectors
In the following sections, various detectors commonly used in our field are described with the focus
3.2.2.1 Photodiode
P + - P
I + - Depletion + - Depletion
Region Region
+ - + -
N N + -
(a) (b)
Figure 3.1: Photodiode structure of (a) PIN diode and (b) Avalanche Photodiode. Reproduced from
reference [111]
The photodiode is a robust and inexpensive detector for relatively high light levels. Light with
energy greater than band-gap energy hits the photodiode, excites electrons into the conduction
band, and creates a hole in the valence band. An electric field in the depletion layer drives electrons
to the N layer and holes to the P layer. The external circuit collects this photocurrent. This process
is illustrated in Figure 3.1(a) for PIN photodiode. PIN photodiode includes an intrinsic region
between the P and N layers, which results in the expansion of depletion region and thus broader
spectral response. PIN photodiode does not have internal gain mechanism. The need of external
45
amplifier introduces substantial noise. The current is propotional to incident power in normal
operating range.
The avalanche photodiode is a high-speed, high-sensitive photodiode with an internal gain mecha-
nism through a reverse-bias voltage. The electron-hole pairs are generated from exposure to light
with higher photon energy than band gap energy. A reverse voltage applied to the PN junction of
the APD causes the electrons to drift towards N layer and holes towards P layer. An avalanche
effect occurs when the electron-hole pairs acquire sufficient energy to create additional pairs by
colliding with the crystal lattice as shown in Figure 3.1(b). The gain of APD could reach up to
100 [111].
Photomultiplier tube (PMT) is a sensitive detector which amplifies the input light signal 10 5 − 106
times with almost no additional noise. It is usually selected for high speed or low light level
detection. It is suitable for single photon counting applications when the rate photons strike the
Figure 3.2 illustrates how the PMT converts light into a detectable electric signal. The PMT
first converts the photon which strikes a photocathode into a photoelectron by the photoelectric ef-
fect. Amplification is performed through a dynode chain, which consists of 8 to 12 metal plates. A
potential of around 100 volts applied between the photocathode and the first dynode accelerates the
photoelectron into the dynode. Upon striking the dynode, 5 - 6 secondary electrons are produced.
These electrons are accelerated into the next dynode by a potential difference and get multiplied
46
focusing electrode
secondary electron
stem pin
vacuum
LIGHT
stem
faceplate Dynodes
Photocathode
Anode last dynode
Figure 3.2: Schematic of typical photomultiplier with some electron trajectories (in red). Repro-
duced from Hamamatsu manual [120].
by 5 - 6 times each. These cascading effect produces 10 5 − 106 electrons at the anode.
The charged coupled device (CCD) is a solid state sensor with a wafer of silicon crystal. When
the silicon is exposed to light, the photoelectric effect generates electrons from the silicon bonds.
These free electrons are collected by CCD electrodes (as shown in Figure 3.3(a)) at the interface
created by positive surface potential to form an extremely thin but very dense inversion layer. The
amount of charge deposited in the inversion layer is often described by the hypothetical concept
of the potential well. In Figure 3.3(b), the charge transfer mechanism is illustrated using a three-
phase CCD structure as an example [20]. In a three-phase CCD, three sets of electrode strips make
one pixel where the charge accumulates on the electrode biased more positively than the other
two. The charge is transferred by making the adjacent electrode potential raised while the first
lowered. Figure 3.4 shows the CCD chip structure [169] consisted of parallel and serial electrodes
and amplifier on a silicon chip. The channel stops restrain electrons from moving along the length
of electrode, thus defining the extent of the pixel. The charges in potential wells columns are
47
shifted in parallel onto a serial register. Then the charges in the serial register gets shifted out
onto the output node. At this stage, on-chip hardware binning can be incorporated to increase the
signal-to-noise ratio. The charges are then amplified by an on-chip amplifier and digitized.
0V +V 0V 0V
depletion layer
0V 0V +V 0V
p-type semiconductor
(a) (b)
Figure 3.3: (a) Single CCD electrode, (b) Charge transfer mechanism of CCD electrodes (three-
phase CCD system). Reproduced from reference [20].
plate, and the phosphor screen. The electrons due to photoelectric effect are accelerated and multi-
pled through microchannel plate (MCP) via mechanism analogous to photomultiplier, to the phos-
phore screen where the light is released upon striking the coating. The image intensifier can be
combined with CCD to detect in ultra-low-light conditions and resolve extremely short temporal
48
parallel transfer electrodes
1 2 3 1 2 3 1 2 3 1 2 3
1
2
3
1 serial
2 output
3 register
1 electrodes
one
2
pixel
3
1
2
3
silicon chip
output amplifier
Figure 3.4: CCD chip layout for three-phase CCD system. First, charges collected at each electrode
are shifted along parallel electrodes onto serial register electrodes. Then the charges are shifted to
the output amplifier. Reproduced from reference [169].
The dynamic range of the detector is characterized by a low limit set by the noise equivalent power
(NEP) from dark current, a middle linear region composed of various noise sources, and a high
limit set by saturation (Figure 3.5). It is important to identify and characterize the linear region
of system to optimize measurements for given application. The characterization of this region in
the overall electronic-optical system will be described in the following section 3.5. In this section,
When the incident light photons arrive at the detector, not every photon produces electrons.
This factor is commonly characterized by the quantum efficiency η (number of electrons produced
/ incident number of photons), which is given in percentage. Also, for APD and PMT, it is given
49
Detector
Response (V) Saturation
where Ik is the photocurrent [A] and Pin is the incident light power in [Watt]. The quantum
hc
η= Sk , (3.2)
λq
where h is Plank’s constant (6.63×10−34 J×s), c is the speed of light (2.99×108 m/s), q is electron
charge (1.60 × 10−19 coulomb), and λ is the light wavelength (in [m]).
There are three major noise factors in photodiode and photomultiplier: shot noise due to in-
cident photon statistics, dark current induced noise, and thermal noise. The shot noise associated
√
with current flow across a potential barrier is given by ishot = 2qBI [A] where I is the average
50
dc current [A] and B is the bandwidth [Hz] [201]. The shot noise due to the photocurrent I k is
√ √
iphoton = 2qBIk , where the shot noise is in fact proportaional to nphoton , where nphoton is the
equivalent number of photons after degradation caused by imperfect conversion caused by quantum
efficiency η. Since Ik is proportional to the incident light power, the photon shot noise increases
with respect to the increase of Pin (in square root fashion). The lower end of this relation is limited
by the noise from the dark current Idark , which is present even there is no input light. The param-
√
eter is usually given in terms of Noise equivalent power (NEP), where P N EP Sk = 2qBIdark .
These noise factors are given at the point where no amplification has yet taken place, such as at the
cathode of PMT or before the avalanche effect in APD. Thermal noise is due to thermal agitation of
p
electrons within a resistance and given by ithermal = 4kT B/R, where k is Boltzman’s constant
(1.38 × 10−23 J/K), T is absolute temperature [K], and R is the load resistance [Ω]. Thermal noise
usually sets the lower limit on the noise present in an electronic circuit.
To calculate the signal-to-noise ratio (SNR) of a PMT, we consider the signal and noise after
the amplification process of the detector. The amplified current is then I signal = GIk = GSk Pin
p
inoise = 2qBG2 Ik + 2qBG2 Idark . (3.3)
Here, the thermal noise component is ignored since PMT is a current source. SNR is then calcu-
lated via Isignal /inoise . This simplified relationship applies to PMT with the assumption that gain
is quite large and there is minimal additional noise coming from the amplification process (i.e.
dynode chains). In photon counting mode, one could approximate dark noise contribution to be
zero since the discriminator threshold could be set up to reject dark count (i.e. shot-noise limited).
51
However, in the case of the APD, it is more complicated due to its different amplification
mechanism. Dark current in the APD is categorized into surface leakage current I ds which does
not go through the multiplication process and an internal current I dg generated inside the silicon
substrate. The non-uniformity of ionization induces “excess noise” during the multiplification
r
4kT B
iAP D
noise = 2qBG2 F (Ik + Idg ) + 2qBIds + (3.4)
R
where F is excess noise factor related to gain G. In PIN photodiode, usually the thermal noise
plays a major role in the noise contribution since there is no gain mechanism.
The calculation of signal and noise of the CCD takes a different form as follows. There are
four major noise sources associated with the CCD; dark current noise, read noise, shot noise and
fixed-pattern noise. Due to the thermal energy within the silicon lattice of the CCD, dark current
is generated. The statistical fluctuation of dark current (dark current noise) can be reduced by
cooling the CCD with thermoelectric coolers (TEC) or liquid nitrogen. Read noise is the random
noise associated with the process of quantifying the electronic signal on the CCD, especially from
on-chip pre-amplifier. Shot noise arises from the inherent variation of the incident photon flux.
Fixed-pattern noise is a dominant noise at relatively high light level, resulting from differences in
Within the light level below where the saturation begins, the signal-to-noise ratio can be calcu-
lated by
ηNp t
SN RCCD = p (3.5)
ηNp t + (Nd t)2 + Nr2
where η is the quantum efficiency, Np is the photon flux (photons/pixel/second), t is the integration
52
time (seconds), Nd is the dark current (electrons/pixel/second, in short form e − /p/s), and Nr is
the read noise (electrons RMS/pixel, e− /p) due to the on-chip amplifier. Given Pin at one pixel,
Np = λPin /h/c.
For detectors commonly used in our laboratory, SNR for incident power level can be calculated
using the specification information summarized in Table 3.1. SNR of APD, PMT and CCD is
plotted with respect to the incident light power in the horizontal axis in Figure 3.6. The bandwidth
B was assumed to be 1 Hz. The load registance of APD is order of 1 GΩ. The APD inside
the module C5331-04 is S2384. The excess noise index is 0.3, therefore the excess noise factor
F = G0.3 . Also, as a quick comparison, PMT with improved quantum efficiency in NIR region
cathodecurrent = 1 pA.).
√
type model NEP (W/ Hz) sensitivity η gain Idark
APD S2384 4 × 10−13 0.5 A/W 32 % 30 1 nA (after gain)
PMT R928 1.3 × 10−16 0.02 A/W 3% 107 30 nA (anode)
type model Nr (e−1 /p) full well η exposure time Nd (e−1 /p/s)
CCD Versarray 8 200,000 e−1 37 % 500 ms 0.5
Table 3.1: Parameters needed for estimating signal-to-noise ratio for detectors commonly used in
our instrument. Noise equivalent power (NEP), sensitivity and quantum efficiency are given for
λ = 800 nm.
It should be noted that incident light power in Figure 3.6 is per unit detector, i.e. area of
detection was not considered into the calculation yet. For each detector, the incident power range
up to the onset of saturation is shown. The lower limit of APD (S2384) is governed mainly by
the dark current noise component. The cross-over to photon shot noise limited region happens
around 10−10 W. In reality, APD does come with built-in amplifier as a module (C5331-04), whose
significant noise factor was not considered in this calculation. Therefore its SNR in the high light
53
4
10
3
10
Signal/Noise
2
10
1
10
PMT: R928
APD: C5331−04
0 CCD: Versarray
10
PMT: III−V
Figure 3.6: Signal-to-noise ratio (SNR) vs incident light power. SNR of PMT, APD and CCD
commonly used in our instrument are estimated based on the specification given in Table 3.1.
PMT with improved photocathode material is shown as PMT: III-V. When SNR = 1, the detector
performance is limited by the noise. Since the high end of linear range is limited by the saturation
point, only the incident light power range up to the saturation is shown for each detector (i.e.
maximum CCD SNR is 500.).
power range appears to be quite high. PMT R928 in our laboratory is used in analog mode (i.e.
not photon counting mode) and not cooled. Thus, the lower light is limited due to the presence
of dark current noise, but it quickly goes into photon shot noise limited linear region from 10 −15
W to its saturation ∼ 2 × 10−10 W (saturation due to anode linearity limit). This large dynamic
range enables us to employ various combination of multiple source detector separations needed for
imaging geometry. The specific CCD that we are using for breast cancer imaging has high quantum
efficiency. However, the maximum SNR is limited by its linear full well, which is the number of
electrons that the individual CCD pixels can hold before spilling over into adjacent wells.
Figure 3.6 gives an impression that CCD will be definitely the detector of choice. However, one
54
should not use this figure assuming the same amount of incident light falls onto each detector unit
in the real scenario. In the experimental setting, one usually would like to use the same source light
power onto the turbid medium of interest and see how the different detectors respond. The incident
light power Pin for unit detector depends on detector-specific parameters such as the collection
where Adet is the detector area, floss is the loss factor associated with inefficiency from fiber
coupling and factor relating to N.A. of detector [93], and |Φ| is the fluence rate intensity in medium
If we assume a homogeneous medium with typical optical properties measured at source de-
tector separations of 6 cm, |Φ| ∼ S × 10−6 [Watt/cm−2 ] (Equation 2.7). Also, if we assume S= 1
mW and general loss factor floss to be 0.05 for simplicity, then one can see the area of the detector
plays a big role in determining Pin . We usually use 6 mm diameter fiber for PMT R928 and 3
mm diameter fiber for APD C5331-04. For CCD, 0.3 cm pixel size is assumed which is similar to
what we use in our measurement. The bandwidth B was taken as 2 Hz (to match the usual CCD
exposure time of 500 ms). From this detection area differences, incident power and SNR for the
signal level are summarized in Table 3.2. SNR of PMT is larger than that of CCD and much larger
with our experimental observations. In the real experimental set-up, the use of APD for 6 cm
source detector separation is not so practical. This can be seen from the incident power being
very close to its NEP. With more inefficient coupling, Pin can easily become similar to NEP.
On the other hand, PMT has been used extensively for large source detector separations without
55
type model coupling detection area (cm−2 ) Pin SNR
APD S2384 3 mm fiber 0.07 3.5 × 10−12 W 8
PMT R928 6 mm fiber 0.28 1.4 × 10−11 W 660
CCD Versarray 0.3 mm pixel 4.5 × 10−4 9.0 × 10−14 W 180
Table 3.2: Example of SNR calculation of our commonly used detectors in the standard breast
measurement for λ = 800 nm. Depending on the detection area, the incident input power (P in to
the detector varys and thus affect SNR.
difficulties. Note that Pin on a CCD pixel is much lower than that of APD or PMT due to its area
even though the capability of CCD to detect low incident power seems to be better than PMT due
to its excellent quantum efficiency. (Area was not from the physical CCD-chip dimension, but the
magnified imaging area by the lens typically found in the experiment.) However, SNR retrieved is
still reasonable for the geometry that we are currently using. One could argue that by averaging
over many number of pixels to cover larger collection area may vastly improve the SNR. However,
each pixel is limited by its read noise (which is higher than several photons) and averaging over
pixels with no response may not necessarily improve SNR by many-fold as expected. Whereas in
PMT, larger collection area can be utilized to enhance the light detectability since it is not limited
by read noise.
The schematic of the homodyne frequency-domain system is illustrated in Figure 3.7. The local
RF oscillator provides a modulated waveform for the laser diode and the reference signal for the
in-phase and quadrature (I&Q) demodulator. The RF modulated light from the laser diode is
coupled to the turbid medium through optical fibers. Depending on the multiplexing method,
56
RF local
oscillator
Splitter
Laser Diode
Optical
I&Q
switch LPF LPF
Demodulator
Turbid
medium
Amplifiers
Detector Attenuators
Filters
AD board
Computer
Figure 3.7: General Homodyne frequency-domain system (LPF: low pass filter). RF local oscilla-
tor provides the reference signal to I&Q demodulator while modulating the laser diode. The source,
detector positions and wavelengths are multiplexed using a combination of optical switches. Then
the attenuated signal through the turbid medium is detected by the detector, goes through proper
amplification and filtering. I&Q demodulator with low pass filter extracts amplitude and phase
information which are recorded in computer through analog-to-digital (AD) board.
optical switches or optical combiners are used for wavelength and optode position changes. The
attenuated signal after propagating through the turbid medium is detected by either a PMT or an
APD. The signal goes through bandpass filter and amplifirs before reaching the I&Q demodulator.
The I&Q demodulator computes the amplitude and phase of the detected signal with respect to the
reference signal. The details of I&Q demodulator are described in the Section 3.3.3.
In RF modulation of laser diode, one needs to first choose the driving circuit which can drive
the laser diode in continuous wave mode with good stability. The driving circuit depends on the
57
polarity of the laser diode and the monitor photodiode. There are four common types of laser diode
configurations depending on the polarities of the laser diode and photodiode connection as shown
in Figure 3.8. The photodiode embedded within the laser diode circuitry is used for regularizing
the laser diode output by feedback circuit. In naming the style, we are following the conventions
LD LD LD LD
cathode
PD PD PD PD
anode
Figure 3.8: (a) Diode polarity, (b) Laser diode style A, (c) Laser diode style B, (d) Laser diode
style C, (e) Laser diode style D. LD stands for laser diode and PD stands for photodiode.
Most of our laser diodes were modulated using laser diode driver chips from Sharp: IR3C01
for the laser diode style B and IR3C07 for the laser diode style C.
laser style Rr Cr Lr Rc Cc Rv
IR3C01 B 47Ω 45pF (f =70MHz) 330µH 20Ω 22µF 20kΩ
22pF (f =140MHz)
IR3C07 C 47Ω 45pF (f =70MHz) 100µH 10Ω 22µF 100kΩ
22pF (f =140MHz)
Rr, Cr and Lr are the parts essential to RF modulation. If one omits Rr, Cr and Lr from the
circuit in Figure 3.9, the laser diode will operate in CW mode. Rr = 47Ω was chosen to give 50
58
IR3C01 IR3C07
Rr Cr Lr Rc Rr Cr Lr Rc +5 V
1 8 +5 V 1 8
RF in LD RF in LD
PD PD Cc
2 7 2 7
Laser diode Laser diode
Cc
(style B) (style C)
3 6 3 6
Rv Rv
4 5 4 5
-12 V
TTL High = ON TTL High = ON
TTL Low = OFF TTL Low = OFF
(a) (b)
Figure 3.9: RF modulation diagram for laser diodes (a) IR3C01 for laser diode style B, (b) IR3C07
for laser diode style C. Without Rr , Cr and Lr , the laser diodes operate in CW-mode.
1
chosen to match the RF input frequency, i.e. f ∼ 2πRC where R = 50Ω (Cr = 45pF for f = 70
MHz, Cr = 22pF for f = 140 MHz). Lr is added to prevent the RF signal from passing into
the laser diode chip. Rc, Cc and Rv values vary depending on the laser diodes. Rv is a variable
resistor with maximum resistance value given by the Table 3.3. Therefore, the tabulated values are
The heart of the homodyne frequency-domain detection is the I&Q demodulator (Figure 3.10).
Suppose the reference signal is REF = Ar sin(ωt) and the detected signal is DET = Ad sin(ωt+
θ), where θ is the phase shift with respect to the reference. In the I&Q demodulator, the reference
signal is split into two, one of which goes through a 90o phase shifter. The detected signal is
also split into two, but they do not go through the phase-shifter. The multiplication of unshifted
59
Reference signal
o
90 shift
I(t) Q(t)
o
0 shift
Detected signal
Ar Ad
I(t) = sin(ωt) · sin(ωt + θ) + I0
2 2
Ar Ad o
where A = 8 and I0 is the DC offset when the RF is absent. In the quadrature section, 90
Ar Ad
Q(t) = cos(ωt) · sin(ωt + θ) + Q0
2 2
where Q0 is the DC offset when the RF is absent. The low pass filter following at I and Q
output selects out the high frequency signal components. Therefore, I DC = A cos θ + I0 and
60
QDC = A sin θ + Q0 are the final output signal arriving at data acquisition point. The amplitude
p
A= (IDC − I0 )2 + (QDC − Q0 )2 (3.9)
µ ¶
−1 QDC − Q0
θ = tan (3.10)
IDC − I0
It is a standard practice to measure the DC offset I0 and Q0 by blocking the light input to the
detection system.
In this section, specification of CCD camera used in the second generation CCD-based parallel
plate instrument (Section 3.6.2) is discussed. A 16-bit thermoelectically cooled CCD array (Ver-
sArray:1300B, Roper Scientific) is operated at 500 ms exposure time, 2×2 on-chip binning at
high-sensitivity amplifier setting for the region of interest covering an imaging area of 16 cm ×
11 cm onto 570 × 400 binned pixels. The specification is summarized in Table 3.4. Binning of
the image increases SNR at the cost of resolution. On-chip binning which happens at the hardware
level yields higher SNR given exposure time. This feature can also be used to decrease exposure
time for the same level of SNR. The use of ROI in conjunction of binning makes the CCD trans-
fer time shorter than that of the full frame (450 ms for our ROI, 1.8 second for full frame). The
spectral response of this CCD is good for near-infrared range. The quantum efficiency is above
60 % from 380 - 850 nm with its peak around 550 nm (92 %), decreasing towards longer wave-
length (30 % for 950 nm). (Back-illuminated device has better quantum efficiency compared to the
front-illuminated device.)
61
CCD sensor E2V CCD36-40; back-illuminated device (with AR coating)
CCD format 1340 × 1300 imaging pixels; 20 × 20-µm pixels
100% fill factor; 26.8 × 26.0-mm imaging area
single pixel full well 200,000 e− - 300,000 e−
User-selectable gains 0.5×, 1×, 2× (low-noise mode)
Read noise −
8 e rms at 1 MHz (high-sensitivity amplifier)
Dark current 0.1 e− /p/s at -40o C operation
Nonlinearity < 2%
Nonuniformity ≤ 4% over entire CCD area
Dynamic range 16 bits
Frame readout < 1.8 seconds for full frame at 1 MHz
Dark current < 0.5 e− /p/s at -50 o C
Operating temperature -40 o C with TEC (backfilled)
Thermostating precision 0.05o C over entire temperature range
For our CCD, dark current noise is 0.1 e− /p/s (electron/pixel/second) and read noise is 8 e− /p
for the high-sensitivity amplifier. For 500 ms exposure time, the dark current noise is 0.05 e − /p
which is considerably smaller than the read noise. The saturation point is when the potential well
is full (∼200,000 electrons), which corresponds to 1.5 × 10 −13 W. Since the read noise defines the
limitation of the signal, the dynamic range of CCD is obtained by usually dividing linear full well
by read noise. In order to take full advantage of the dynamic range, an appropriate A/D converter
is selected for each camera. The gain is defined in terms of electrons/ADU (analog-to-digital unit).
Suppose the gain is set to 1×, then 1 ADU corresponds to ∼3 electrons (linear full well / 2 16 ).
However, for this particular CCD, the conversion factor was set as 4 e − /ADU, 2 e− /ADU, and
1e− /ADU (which is referred as gain setting). Depending on the gain setting, the saturation due to
the full potential well may not happen before it is limited by digitization limit (2 16 ADU). For the
example presented in Table 3.2 (i.e. Pin = 4.5 × 10−14 Watt), assuming 70% quantum efficiency,
the number of electrons deposited in a pixel is 63,000 e − which corresponds to 31,500 ADU for
medium gain. The order of magnitude of this matches with the experimentally measured value.
62
3.5 System Characterization
Once the instrument is designed and assembled, the characteristic dynamic range needs to be quan-
tified. It is important to first identify the dynamic range of major electronic components, since the
instrument design goal is to build the system limited by the optical components (i.e. detectors),
not by the electronic components. Then the dynamic range of the whole system including the
electronic and optical components is measured to characterize the system. This methodology is
demonstrated in the following sections using a one detector channel frequency-domain homodyne
system.
70 MHz
Oscillator
Power
Splitter
Attenuator
Input
signal
I&Q Amplifiers,
Demodulator etc.
Reference
signal
LPF LPF
AD board
Computer
Figure 3.11: Schematic of electronic dynamic range measurement set-up using electronic attenua-
tor.
63
In determining the dynamic range of electronic components, we started by testing the response
of I&Q demodulator with input signal variation. Then we added amplifiers one by one to see if
The schematic of the test set-up is described in Figure 3.11. A 70 MHz local oscillator signal is
split into two. One provides the reference signal for the I&Q demodulator. The other is attenuated
and serves as the detected signal to the I&Q demodulator. Amplitude response measured for each
RF input signal is plotted for three cases in Figure 3.12: (1) only I&Q demodulator was present, (2)
I&Q demodulator and two amplifiers were present and (3) I&Q demodulator and three amplifiers
were present. The dynamic range defined by I&Q demodulator is RF input of -60 dBm to -5 dBm.
The addition of amplifiers do not limit this range, but merely shifts the range to lower RF input by
its gain factor (i.e. With two amplifier, the dynamic range is -120 dBm to -65 dBm).
Measured
amplitude (dBm)
20
-20
-40
-60 IQ only
2 amplifiers
-80 3 amplifiers
-100
-140 -120 -100 -80 -60 -40 -20 0 20
RF input signal (dBm)
Figure 3.12: Effect of amplifier addition on the electronic dynamic range. 2 amplifiers: test of two
amplifiers and I&Q demodulator, 3 amplifiers: test of three amplifiers and I&Q demodulator
64
The I and Q output contains the DC offset (I0 , Q0 ) as in Equation 3.10. As shown in Fig-
ure 3.13, offset subtraction improves the amplitude dynamic range especially at the low end and
10 250
0 200
Amplitude (dBm)
Phase (degree)
−10 150
−20 100
before offset subtraction
before offset subtraction after offset subtraction
−30 after offset subtraction 50
−40 0
−40 −30 −20 −10 0 10 −40 −30 −20 −10 0 10
RF input (dBm) RF input (dBm)
(a) (b)
Figure 3.13: Effect of offset subtraction on dynamic range. (a) Amplitude vs RF input, (b) Phase
vs RF input. Offset subtraction improves the dynamic range at low RF input and the phase stability.
The dynamic range of the whole system combining the electronic and optical components is as-
sessed using the set-up illustrated in Figure 3.14. The light emitted from light source is reduced by
a mechanical attenuator and then split into two beams by 1×2 fiber splitter. One beam goes to an
optical power meter and the other one goes to the detection part of the optical system. The signal
response of the detection system to the attenuated light signal is plotted against the light signal
measured by the optical power meter in Figure 3.15(a). Figure 3.15(b) shows the deviation of the
measured value from the fitted linear line over the dynamic range. Typically, we choose the range
by the deviation of ± 1%. In this example, the dynamic range is -50 ∼ -10 dBm. Note this range
65
is smaller than the electronic dynamic range of previous section. However, this dynamic range can
Optical
1x2 fiber splitter
Mechanical Powermeter
attenuator
Light Detection
Source electronics
The phase accuracy is tested using the set-up shown in Figure 3.16(a). The optical fiber end
mount and lens are separated by the focal length of the lens so that the light is parallel after exiting
the lens. Then another lens focuses the beam into fiber coupled to the detector. The mount and first
lens are moved as a unit while the distance between the two lenses is varied. The change in the
distance between the two lenses produces a change in phase. Since the modulation frequency is 70
MHz, the expected phase change is 0.84o /cm. Typical linear response in phase at fixed amplitude
66
3
60 Dynamic Range
2
Dynamic Range
Amplitude (dBm)
40
1
Error (%)
20
0 0
−20 −1
measured
−40 fit −2
−60
−3
−70 −60 −50 −40 −30 −20 −10 0 −50 −40 −30 −20 −10
Power (dBm) Power (dBm)
Figure 3.15: Amplitude linearity Test data. (a) Amplitude vs power in linear scale, (b) Amplitude
error (%)
3.6 Instrumentation
The 1st generation parallel plate transmission DOT instrument is utilized for quantification of
bulk properties of healthy female breasts [81] (Chapter 4, Section 4.3). The instrument uses three
wavelengths - 750nm, 786nm and 830nm, and employs a scanning, fiber-coupled PMT detector
(R928, Hamamatsu) for detection. The system is characterized by a noise equivalent power of
√
≈ 0.1pW/ Hz, a linearity in amplitude of less than 1%, and a phase drift of 0.25 o over 80 dB
(with use of attenuators). We calibrated the instrument over a broad range of input powers in order
to extend this range. For transmission measurements the signal variation is typically only ≈30
dB. In clinical measurements we utilize a single source position due at the center of the scanning
region. The lasers are amplitude modulated at 140 MHz to produce diffuse photon density waves
(DPDWs) in the medium. The amplitude and phase of the DPDW is recovered using a homodyne
67
12
8
Mechanical
attenuator movable lens 6
lens unit
4
2
Light Detection
Source electronics 0
0 2 4 6 8 10 12
expected phase shift (degree)
Figure 3.16: (a) Phase linearity test set-up, (b) Phase linearity test data. Measured phase shift (in
degree) vs expected phase shift (in degree) at fixed amplitude within the linear dynamic range.
For the in vivo measurements the volunteer lies in the prone position and her breast is inserted
into a small tank filled with a matching-fluid solution of Intralipid and india-ink mixture (see Figure
3.17). The detector is scanned along the output plate glass surface. The source is attached to the
movable compression plate which applies a gentle compression to the breast. Usually the range of
compression is 4.5 cm to 7 cm. It takes ≈ 15 minutes to acquire data from a 9.6 cm (x) by 4.8 cm
(y) scan region with 153 (17×9) points. We note that the feedback from volunteers was generally
positive; in comparison to the X-ray mammography, they felt that the soft-compression of the DOT
instrument did not cause discomfort. Figure 3.18 shows the detailed electronics of the set-up. We
took two sets of data for each volunteer: 1) From the tank filled with the matching fluid, without
the breast; this data enables us to normalize for instrument response for imaging purposes, 2) From
the tank filled with the matching fluid and the volunteer’s breast. The Intralipid/ink solution helps
68
X−Y scanning
stage Compression Image View
plate
Intralipid
Detector Fiber box
Electronics
Source Fiber
Detector grid
Source on
opposite plane
Figure 3.17: A sketch of the prototype clinical table. The volunteer lies in the prone position with
her breast inserted into the tank through an openning on the bed. Soft compression is applied on
the source plane and detector scans a 2D grid on the opposite plane. Image view shows the source
detector positions as seen in the data.
I-output
Output
Q-output
Diode Lasers:
Amplifiers
750nm, 786nm, 830 nm
Digital Attenuator
3 2x1 Optical Switches PMT
Filters
140 MHz Oscillator
IQ Demodulator
Power Splitter
RF shielded
Reference
NIM Bins
Source Fiber
Intralipid box
Detector Fiber
X-Y stepper
motors
Figure 3.18: Parallel plate scanning instrument electronics consists of three wavelength laser
diodes and 140 MHz homodyne system with PMT.
69
3.6.2 2nd generation parallel plate DOT instrument
Based on the experience gained by the first generation parallel plate instrument, the second gener-
ation instrument was developed with the focus on imaging [62]. A schematic of the DOT instru-
ment is shown in Figure 3.19. The hybrid system combines frequency-domain (FD) remission and
continuous-wave (CW) transmission detection. This parallel-plane DOT system has been exten-
sively characterized for breast cancer imaging using various tissue phantoms and a healthy female
volunteer [62] (Chapter 3, Section 3.7.2) and used extensively for breast cancer imaging (Chapter
The table was designed so that when the female subject lies on it in the prone position, both
her breasts are inside a breast box (60 × 22 × 23 cm) extending underneath the table. A breast
is typically centered and softly compressed between a movable compression plate and a parallel
viewing window with cranio-caudal compression. The compression distance varied from 5.5 - 7.5
cm, depending on breast size. The breast box was coated black and designed to hold the matching
fluid which has optical properties similar to human tissue. The matching fluid was made with
Liposyn III (30 %, Abbott Laboratories, Chicago, IL) as a scattering agent and India ink (Black
India 4415, Sanford, Bellwood, IL) as an absorption agent. The optical properties of the matching
fluid were fixed so that µa (786nm) = 0.05 cm−1 and µ0s (786nm) = 8 cm−1 . The absorbance of
the ink solution was checked after each measurements with a spectrophotometer (Ocean Optics,
USB2000).
Four laser diodes at 690, 750, 786 and 830 nm were sinusoidally intensity modulated at 70
MHz. The modulation depth for each wavelength was 86 %, 99 %, 72 % and 67 % respectively.
Later on, 650 and 905 nm CW laser diodes were added to the instrument (along with optimization
of acquisition time from 12 minutes to 8.4 minites). Combination of optical switches (DiCon Fiber
70
(a)
(b)
Figure 3.19: Schematic of parallel plate diffuse optical tomography instrument. (a) Frequency-
domain (FD) remission and continuous-wave transmission measurements are performed simulta-
neously on a female subject lying in the prone position. (b) The source plate contains 45 source
positions and 9 FD detectors. 984 detection points with 3 mm spacing are selected from CCD data
for image reconstruction.
Optics) making 6 × 2 connection was used for wavelength switching. The light output was relayed
to a 2 × 48 switch (DiCon Fiber Optics, GP700) in order to access the various light source positions
on the compression plate. The compression plate had 45 fibers in 9 × 5 grid with a spacing of
1.6 cm (12.8 cm × 6.4 cm). The optical fibers embedded in the compression plate was of 200
µm diameter (FIS). Time-multiplexing was used for both wavelength and position switching. In
71
Nine 3 mm diameter fiber bundles were connected to the frequency-domain detection module,
technique [288] was utilized to extract the amplitude and phase of the detected remission signal.
The electrical signal from the avalanche photodiode (APD, Hamamatsu C5331-04) is amplified
(Mini-Circuits ZFL-500LN, 24 dB), filtered by a band pass filter (Mini-Circuits BLP-70), and
ZFMIQ-70D) with a series of low pass filters (Mini-Circuits SLP-1.9 and 100 Hz RC circuit)
extracts the amplitude and phase by comparing the signal with the reference signal driving laser
diodes [288]. The frequency-domain measurements are used for accurate quantification of bulk
properties of human tissue and matching fluid, thus improving our initial guess for the image
reconstruction.
A lens-coupled 16-bit CCD camera system was used for collecting CW transmission data with
an anti-reflection coated glass viewing window. A lens (Nikkor AF 50 mm F/1.4D) relayed the
detection plane (inner glass window in contact with breast) onto the CCD chip (2.68 × 2.6 cm). To
reduce ambient light, a long-pass color-glass filter (630 nm, CVI Laser Inc.) was placed in front of
the lens and a light shielding box was placed surrounding the space between the viewing window
and the camera. A thermoelectrically cooled CCD pixel array (Roper Scientific, VersArray:1300F,
1340 × 1300 pixel) was used for light detection, with 1140 × 800 pixel region of interest (ROI)
and 2 × 2 on-chip binning, resulting frame size of 570 × 400 pixels covering a detection area of
72
Figure 3.20: Schematic of (a) the trans-abdominal probe and (b) the NIR frequency domain instru-
ment to which the probe is coupled.
for trans-abdominal NIR spectroscopy [50] (Chapter 5, Section 5.2). Figure 3.20 shows a schematic
of the trans-abdominal probe and the NIR frequency-domain instrument to which the probe is cou-
pled. The probe was designed as a linear array consisting of 2 detector fibers and 6 source fibers.
This probe is capable of performing NIR photon diffusion measurements at a total of 12 source-
detector separations ranging from 1.8 to 9.5 cm. However, in this study, NIR photon diffusion
73
sufficient for retrieval of fetal cerebral blood saturation. The instrument consists of a light source
module and two detection modules. It is described in detail by Yu et al. [293]. Laser diodes at 675
nm, 786 nm and 830 nm are intensity modulated with three different local radio frequency (RF)
oscillators operating at around 70 MHz. The light output from each laser diode are combined using
a fiber combiner into a single optical fiber. In this manner, the single optical fiber can simultane-
ously deliver light output from the three frequency encoded laser diodes [288, 293]. The total light
output from the single optical fiber was 15 mW (3 mW at 675 nm, 9 mW at 786 nm and 3 mW at
830 nm respectively). An optical prism switch was used to direct the light output from the single
optical fiber to the six source fiber positions on the trans-abdominal probe.
The detection module consists of an avalanche photodiode (APD), and two amplifiers, with a
band pass filter between them. The output from the detection module is connected to a demodu-
lation unit, which consists of three In-phase and Quadrature-phase (I&Q) demodulators. The in-
phase and quadrature-phase signals are low-pass filtered, digitized, and then converted into diffuse
wave signal amplitude and phase. Total acquisition time for the NIR photon diffusion measure-
This instrument has been extensively tested in tissue phantom and rat brain studies [293]. The
√
noise equivalent power is less than 10 pW/ Hz and the dynamic range is greater than 70 dB
(amplitude errors < ± 1% and phase error < 1o .). The instrument has very low inter-channel
crosstalk (< -80 dB) and good long-term stability (amplitude errors < ± 1% and phase error < 1 o
in 30 minutes).
74
3.7 Validation with Tissue Phantoms
Liquid phantoms serve as the gold standard to test the instrument and the algorithm using a com-
bination of absorbers and scatterers with known spectra. The titration of absorbers is useful for
testing the µa response of the overall system. Similarly, the titration of scatterers tests the µ 0s re-
sponse. Indocyanine Green (ICG) is helpful for checking the spectral response since its spectra
has a well-defined peak in the NIR spectrum. Ultimately, titration of hemoglobin simulates the
global response from human tissue. The recipe for constructing liquid phantoms is described in
Section 3.8.1. However, solid phantoms are useful in simulating more tissue-like geometry and do
not deteriorate as fast as the liquid phantom. For imaging purposes, a compressible solid phantom
Typical raw data of breast is shown in Figure 3.21, overlayed with the forward data generated
from a fit based on the semi-infinite homogeneous medium analytic solution and a fit based on the
finite difference method (FDM). FDM fit was based on two regions of different optical properties
(breast and matching fluid regions) as shown in Figure 3.22(b). Particularly, NRIM described
in Chapter 2, Section 2.4.1 was modified to calculate bulk optical properties by summing up the
weight matrices only within the breast region and not allowing the matching fluid region to update
optical properties. The dispersion of amplitude and phase from the semi-infinite fit is mainly due
The scanning-based 1st generation instrument has been characterized by extensive phantom
studies [64, 82–85]. In particular, the phantom experiment was devised to best model the actual
breast measurement geometry as described. Balloons filled with different concentrations of ink
75
0.02 1
FD
Meas
0.015 Semi−inf
Phase (rad)
Amp (norm)
0.5
0.01
0 Semi−inf
0.005 FD
Meas
0 −0.5
5 5.5 6 6.5 7 r (cm) 7.5 5 5.5 6 6.5 7 r(cm) 7.5
(a) (b)
Figure 3.21: (a) Amplitude and (b) phase vs source detector separation from a healthy female breast
tissue showing the measured data (Meas) and fits with semi-infinite (Semi-inf) and finite difference
method (FD). FD simulates the spread of measurement from semi-infinite due to boundary effect.
and Intralipid (Figure 3.22(c)) were used as a simulated breast. The balloons were inserted into
the breast box filled with the matching fluid to fill a volume occupied by the average breast tissue
volume found in the clinical measurements. A large fraction of balloon was kept above the match-
ing fluid level to simulate the chestwall. The background properties of the matching fluid were µ a
= 0.05 cm−1 and µ0s = 8 cm−1 at 786 nm. The retrieved bulk optical properties of fifteen balloon
phantoms using FDM fit are fairly linear in comparison with the expected value as shown in Figure
3.23. Even though there are some deviation, it shows a fairly linear relation.
First, the frequency-domain part of the instrument was tested on accuracy of retrieving optical
properties of homogeneous medium. Typical matching fluid optical properties fitted from RF semi-
infinite solution with coupling coefficient as described in Section 2.3.2 is presented in Figure 3.24.
The response of the DOT instrument to overall µa change of the medium was tested by ink titration
76
Background Measurement Volunteer Measurement Baloon Measurement
Geometry Geometry Geometry
air air
x x 6cm x 6cm
10cm
y y y
11cm
11cm
Figure 3.22: Measurement geometry for three cases: (a) background, (b) breast, and (c) tissue
phantom are shown with the corresponding boundaries. The shaded regions show the estimated
breast volume in the segmentation process. The source boundary is diffuse white reflecting and the
detection plane is clear glass.
to the matching fluid (i.e. the tank was filled with matching fluid of varying µ a with the method
described in Appendix Section 3.8.1). Figure 3.25 shows good linear response of DOT instrument
up to µa = 0.08 cm−1 , which is well within the breast optical properties range.
Then, to test the imaging capability of our instrument, various imaging phantoms were devel-
oped for breast cancer imaging application as shown in Figure 3.26. The first two are mostly for
differential type of measurements, each object simulating tumor with higher absorption or scatter-
ing than the rest of the tissue. Figure 3.26(b) was aimed to provide the flexibility to change optical
77
0.08 20
µ recon
µ’ recon
0.07
s
0.06 15
0.05
0.04
10
0.03
0.02
µ real 5 µ’ real
a s
0.01
0.02 0.04 0.06 0.08 5 10 15 20
Figure 3.23: Linear response of Scanning DOT to optical properties variation. µ a variation (left)
and µ0s variation (right)
properties of the object easily by changing liquid solutions and pumping into it. Figure 3.26(c)
We evaluated the resolution performance of our system using point spread function (PSF)
measurements. We used small strongly absorbing point-like objects, specifically 0.4 cm diame-
ter sphers with µa = 2.0 cm−1 and µ0s = 8 cm−1 . Two such objects were arranged at r1 = (-2.5,
3.0, 0.0) and at r1 = (2.3, 3.0, 0.0). Reconstructions were performed for a range of regularization
constants lc from 0.001 to 1000. Typically, NRIM (Chapter 2, Section 2.4.1) was used. The re-
constructed µa (786 nm) images from lc = 0.1 are depicted in Figure 3.27. The cross-sections are
evaluated at the FWHM and they are dependent on the regularization constants. There is gener-
ally a trade off between resolution and image noise. By considering the ratio between the object
contrast and the image noise (CNR), the optimal regularization constant can be chosen. For this
experiment, a 3D point spread function with FWHM of 1.1 × 1.13 × 1.1 cm is obtained.
We also evaluated the ability of the system to image an extended tissue volume and to re-
construct a heterogeneous medium with multiple objects. We dispersed 15 cubic silicone tissue
phantom objects of µa = 0.2 cm−1 and µ0s = 8 cm−1 (0.8 cm edge) throughout the measurement
78
−2 2
10 −1
λ=690 nm µ =0.038, µ ’=8.4 cm
−4 a s
10 1
Normalized Amplitude
−6
10 0
−2
1 2 3 4 5 6 1 2 3 4 5 6
10 2 −1
λ=750 nm µ =0.056, µ ’=7.6 cm
Phase (rad)
−4 a s
10 1
−6
10 0
−2
1 2 3 4 5 6 1 2 3 4 5 6
10 2 −1
λ=786 nm µa=0.049, µs’=7.1 cm
−4
10 1
−6
10 0
−2
1 2 3 4 5 6 1 2 3 4 5 6
10 2
−1
λ=830 nm µ =0.057, µ ’=6.7 cm
−4 a s
10 1
−6
10 0
1 2 3 4 5 6 1 2 3 4 5 6
ρ (cm) ρ (cm)
Figure 3.24: From top to bottom, 690, 750, 786, and 830 nm. Left to right, amplitude vs separations
and phase vs separations.
volume filled with the matching fluid of µa = 0.05 cm−1 and µ0s = 8 cm−1 . The reconstructed
µa (786 nm) images with lc = 0.1 is shown in Figure 3.28 with isosurfaces at µa = 0.092 cm−1 . The
reconstructed objects are well separated in all three dimensions. The spatially variant regulariza-
tion has provided fairly consistent sensitivity and resolution throughout the volume.
phantom, two 3.2 cm3 flow cells (Figure 3.26(b)) were used. One was filled with an ink contrast
(higher µa than the background matching fluid) and the other was filled with an ICG contrast. Mul-
tiple measurements were performed with the titration of ICG while fixing the ink concentration.
The reconstructed µa at all 4 wavelengths were combined to calculate concentration map of ink and
ICG respectively. The concentration images sucessfully distinguish between the ink and the ICG
79
0.14
0.12
Retrieved µ (cm−1)
0.1
a
0.08
0.06
−1
Expected µa (cm )
0.04
0.04 0.06 0.08 0.1 0.12 0.14
Figure 3.25: µa at 786 nm measured and quantified using DOS is plotted with respect to the
expected µa calculated from spectrophotometric measurement.
as shown in Figure 3.29. Moreover, the integrated signals (volume × concentration [cm 3 µM ]) are
accurately retrieved for ICG concentration variations within the physiological region.
Previous tissue phantoms are all small objects (less than 3.2 cm 3 ) suspended inside the match-
ing fluid. However, in vivo breast measurement involves tissue which has different constituents
from the liquid suspended in the matching fluid. This is simulated by using a breast shaped sili-
cone phantom with an embedded object with tubings as shown in Figure 3.26(c). One can change
the optical properties of the embedded object (which simulates a cancer) by flowing different con-
centration of Intralipid/ink solution. Figure 3.30(a) is the reconstructed images with the reference
measurement being the same silicone phantom at different titration stage (smaller ICG concentra-
tion) in the embedded object. The location of the reconstructed absorber matches well with the
expected location. (Confirmed by slicing the silicone phantom in half.) This simulates the case for
extrinsic contrast injection where the breast without the injection serves as the reference. In this
80
(a) (b) (c)
Figure 3.26: Photos of (a) two 1 cm3 silicone cube, (b) hollow cylindrical container with tubings
enabling changes of liquid inside (flow cell), and (c) breast shaped silicone with extension for the
chestwall
case, source coupling coefficients problem diminishes since the signal difference is only coming
from the injection. Figure 3.30(b) is the reconstruction result with the matching fluid as the ref-
erence measurement, with geometric constraint approach. This reconstruction is most similar to
our in vivo breast measurement, where it relies on the intrinsic contrast (Chapter 2, Section 2.4.3
and Chapter 4, Section 4.4.2). The quality of reconstructed image is not as good as the extrinsic
contrast simulation case (Figure 3.30(a)), but the location is fairly accurate. We are in the process
The absorbers typically utilized in constructing liquid tissue phantoms are ink, ICG, Hb, HbO 2 ,
H2 O and lipid. Except the ink spectrum, well established spectra of the absorbers are available in
OMLC homepage (http://omlc.ogi.edu/spectra/, accessed January 2001). In Figure 3.31, the water
and lipid spectras are presented in µa [cm−1 ]. The spectra of Hb and HbO2 are shown in extinction
81
z (cm)
8
6
x (cm)
4 -5 0.09
2
0.08
0 z0 0.07
-2
-4 5 0.06
-6 0.05
-8 -5 0 5 -1
(cm )
-10 0 -8
-5
0 5 -6
5 10 -4
x (cm) 10 y (cm) -2
0 z (cm)
2 2
y (cm) 4 4
6 6
8
-5 0 5 2 4 6
x (cm) y (cm)
Figure 3.27: Reconstructed µa map of two point spread function like objects with high absorption
coeffiencient ² [cm−1 M] (Figure 3.32). ICG spectra changes depending on the suspension medium
(water or plasma) as shown in Figure 3.33. The absorption spectrum of India ink (Figure 3.34(a))
I(λ)
is measured with the spectraphotometer. It is presented in absorbance Ab(λ) = − log I0 (λ) where
I is the transmitted intensity through the absorbing sample of 1 cm thickness and I 0 is the initial
light intensity. India ink is typically utilized for its monotonical spectra makes it easy to produce
expected absorption when testing µa response of the instrument at one wavelength. General con-
figuration for liquid tissue phantom measurement in semi-infinite geometry is shown in Figure
3.34(b).
82
z (cm)
y (cm)
x (cm)
Figure 3.28: Reconstructed µa map of fifteen 1 cm3 objects suspended in the matching fluid
the following
µ0s = µs (1 − g) (3.13)
where λ is in [nm] and µs and µ0s are in [cm−1 ]. µ0s of 0.8 % Intralipid usually used for our
experiments are presented in Figure 3.35. (There is a consistent discrepancy between this expected
value and the measured scattering value. This may be attributed from the difference in scattering
83
Ink & ICG flowable phantom
Figure 3.29: Respective concentration maps of ICG and ink based on reconstructed µ a .
medium we are using (Liposyn III) and van Staveren used (Intralipid).)
First, one needs to dilute the ink concentration to either 1 % or 10 %. If the total liquid volume
is less than 10 l, 1 % dilution is adequate. For the optical mammography, we need 26 l of the
The absorbance of the ink solution to be used should be measured with the spectrophotometer,
84
−3
x 10
13
4
−3
x 10
13
Figure 3.30: Reconstructed µa (cm−1 ) image of a silicone breast shape phantom with an embedded
absorber. (a) Using a measurement on the same silicone phantom with smaller absorption in the
absorber as the reference, (b) Using a measurement on the matching fluid as the reference.
absorbance of 4. Diluting the ink solution to 0.1 % of the original concentration decreases its
absorbance to measurable level. Measure the ink absorbance spectra with spectrophotometer with
If the measured absorbance is Ab(λ), µa (λ) = ln(10) × Ab(λ). The conversion factor ln(10)
stems from the fact that bases of log are different (i.e. Ab = log 10 II0 and µa = ln II0 ). This µa is
that of 0.1 % diluted version of the original concentration. Therefore, one needs to multiply 10 for
From the titration relation Cbef ore × Vbef ore = Caf ter × Vaf ter (C: concentration, V : volume)
can be modified as the following, since µa is proportional to the concentration, µabef ore × V bef ore +
µink ink = µaf ter × V af ter . Especially when calculating the amount of absorber concentration
a ×V a
to be used for desired total µa , one needs to take water absorption into the consideration.
µ010%
s
Intralipid
can be obtained from the Equation 3.11. Usually, 20 or 30 % Intralipid solutions
are readily available. Therefore, one needs to scale µ0s values accordingly. The volume of Intralipid
85
Water Lipid
)
0.5 0.05
−1
absorption coefficient(cm−1)
absorption coefficient(cm 0.4 0.04
0.3 0.03
0.2 0.02
0.1 0.01
0 0
600 700 800 900 1000 600 700 800 900 1000
wavelength (nm) wavelength (nm)
(a) (b)
(c) Example
Suppose that we want to make a solution of µa = 0.05 cm−1 and µ0s = 8 cm−1 at 786 nm for
total volume of 26 l. We will use 10 % diluted ink and 30 % Intralipid. How much volume of
ink and intralipid solution are needed? Suppose the measured absorbance of 0.1 % ink solution is
Ab(786nm) = 1.50. Then in order to get the absorbance of 10 % ink, a factor of 100 should be mul-
tiplied. Also, there is already absorption due to water to be taken into account, i.e. µ aH2 O (786nm)
= 0.021 cm−1 .
For absorption,
µink
a = ln(10) × Ab × 100 = ln(10) × 1.50 × 100 = 345
86
extinction coefficient (cm−1/µM)
Hemoglobins
extinction coefficient (cm−1/µM)
600
Hemoglobins 5
HbO2 HbO2
Hb Hb
500 4
400
3
300
2
200
1
100
wavelength (nm) wavelength (nm)
0 0
200 400 600 800 1000 600 700 800 900 1000
(a) (b)
Figure 3.32: Extinction coefficient of hemoglobins (a) 200 - 1000 nm, (b) 600 - 1000 nm
For scattering,
µ030%
s
intralipid
= 3 × µ010%
s
intralipid
= 3 × 102 cm−1
If one wants to increase the absorption of above solution to µ a = 0.07 cm−1 , then the extra
87
1
5
ICG
extinction coefficient (cm /M)
x 10
2
Normalized absorbance
−1 ICG in water 0.8
ICG in plasma ICG in water
ICG in plasma
1.5 0.6
1 0.4
0.5 0.2
wavelength(nm) 0
0 600 650 700 750 800 850 900
600 700 800 900 1000 wavelength (nm)
(a) (b)
Figure 3.33: (a) Extinction coefficient of ICG (10 µM concentration). (b) Normalized spectra of
ICG in water and plasma.
The ICG preparation used for clinical study is prduced by Akorn, Inc, under the tradename of IC-
solvent. Since the molecular weight of ICG is 775, this combination makes 3.2 mM solution. For
ICG injection study, we use 0.25 mg/kg dose. However, depending on the preparation variation,
this ICG solution into 10 ml water (∼14 µM ) gives the measurable absorbance.
800 ml total volume. There is a significant shift in ICG spectra with increase of concentration.
Figure 3.33(a) is the extinction coefficient for ∼ 10 µM concentration range. For our purpose,
this spectra can be utilized: the extinction coefficient ² at 786 nm is 110210 cm −1 /M . How much
88
3
2.5
2
absorbance
1.5
0.5
0
650 700 750 800 850 900 950
wavelength (nm)
(a) (b)
Figure 3.34: (a) Absorbance spectra of India ink and (b) General configuration for liquid phantom
measurement in semi-infinite geometry
∆µaICG
3.2 × V ICG = × total volume
ln(10) × ²
1% dilution of blood results in good SNR range of spectrophotometer. Use the measurements at
541 nm, 550 nm and 576 nm to determine the total hemoglobin concentrations C t . Assuming
the blood sample is a mixture of oxygenated and deoxygenated hemoglobins, decompose the con-
centrations based on three wavelengths using known extinction coefficient of Hb and HbO 2 from
Figure 3.32(a).
89
12
′
µ of 0.8% Intralipid
s
10
µ′ (cm )
−1
6
s
4
wavelength (nm)
0
600 700 800 900 1000
Then determine the total volume of blood model (Vt ). When diluting the blood, Phosphate
Buffered Saline (PBS) should be used to maintain blood cells at pH=7.4. (The oxygen saturation
of blood diagram shifts with ∆pH = 0.2 difference.) The volume of Intralipid is determined by the
same way as described before. The volume of the blood samples is then determined by V af ter =
For deoxygenation, either yeast or N2 can be used. Yeast induces complete deoxygenation
since it consumes the oxygen. 3-5 grams per liter is adequate. Full oxygenation is obtained by
bubbling O2 into the blood model. NIR probe should be placed sufficiently far away from the
bubbling O2 , as shown in Figure 3.36. Typical StO2 response measured by DOS is presented in
Figure 3.37. StO2 increased fairly fast when the oxygen was turned on and decreased slowly due
to yeast when the oxygen was off. Large downward arrows indicate the time when the blood was
added to the solution. A cycle of oxygen on and off was repeated at each blood titration. Around 0
% and 90% StO2 were achieved at each oxygen off and on period respectively. The last cycle StO 2
90
was noisy since the addition of blood increased absorption significantly as to affect the signal-to-
noise of measurements.
The type of blood (such as whole blood, red blood cells and so on), type of anti-coagulation
agent and expiration date can be found on the cover of the blood bag (from American Red Cross).
For research purposes, we get the blood cells which are not suitable for transfusion. The expiration
date is based on the transfusability. If possible, one needs to get the most recently expired blood.
The blood should be stored in the refrigerator at 1-6 o C. It is usually good up to one month. (Blood
cells can live up to 4 months. However, by the time we get the blood from American Red Cross,
it already consists with portion of dead cells. Live cells make up 10 % of total after 4 months.)
Sampling different sites of the blood bag is possible using sampling site coupler or 3-way stop
cock.
When handling blood, one should wear lab coat, safety glasses and gloves for possible spill or
splash. Also, one handling blood needs to get the Hepatitis shots in advance as well as to take the
training course.
When the experiment is over, dispose needles and pipet tips in the bio-hazard sharps container.
91
Figure 3.37: StO2 response with oxygen supply to blood phantom. StO2 increased fairly fast when
the oxygen was turned on and decreased slowly due to yeast when the oxygen was off at each
blood titration stage. Large downward arrows indicate the time when the blood was added to the
solution.
Dispose other materials in a bio-hazard bag. The bio-hazard bags need to be autoclaved before
pick-up. Add bleach to the containers exposed to blood and blood model itself. Then drain through
the sink. At UPENN, the Principal Investigator should contact the building administrator to have
SMI (infectious waste hauler) come pick up bio-hazard materials. Also, he has to have an exposure
control plan.
RTV 12 silicone product (GE Silicones) consists of two parts : RTV 12A (base compund) and
and Toluene. RTV 12C is 5 % Dibutyl Tin Oxide solution, 20 % Ethyl Silicate 40, 20 % Amino-
92
harmful. Especially one should avoid inhalation and contact with skin and eyes. It is also flamm-
The weight ratio of 20:1 of RTV 12A : RTV 12C is recommended by the manufacturer. How-
ever, increased 10:1 volume ratio is used for making silicone model to accelerate the curing rate.
The density of silicone is similar to water, so weight ratio approximates volume ratio. RTV 12
product makes a bit cloudy silicone when mixed. Other products like RTV 615 makes clear sili-
cone. RTV 12 was chosen because of the price being the lowest. Recommended curing time is 72
As an absorbing agent, carbon black (Raven 5000 Ultra II, Columbian Chemistry Company) is
used. The mean particle size of the carbon black is around 8 nm. Titanium Dioxide (TiO 2 ) is used
as a scattering agent. We used TiO2 from Sigma (T-8141) for most of the solid phantoms.
Silicone adheres to most of the plastic materials. Even if it does not adhere, sides in contact
with the container do not cure as fast as the exposed side to the air. Therefore we built several kinds
of mold that can be taken apart easily after the silicone is cured. Teflon tape (PTFE Teflon tape :
McMaster Carr) is applied on the surface of the mold to easily separate the silicone when cured.
The amount of absorbing and scattering agent to induce the desirable absorption and scattering
was determined empirically. 9 silicone phantoms of the same breast shape (Figure 3.26(c)) with
varying concentration of carbon black and TiO2 were constructed and measured with 1st generation
parallel-plate DOT instruments (frequency-domain). The absorption formula was extracted from
the linear relationship between phantoms under µa = 0.1 cm−1 as shown in Figure 3.38. However,
we did not get the linear relationship with variation of TiO2 amount and µ0s , so one should not
blindly believe the optical properties expected from the recipe. Unlike the liquid phantom, there
are many places the nonlinearity could arise and thus a lot more work is needed. It is provided as a
93
reference point to start and I recommend anyone using this recipe to measure the optical properties
0.2
0.1
0.05
0
0 2 4 6 8 10
carbon black (mg)
Recipe: Silicone
Initially, we have tried making the silicone phantom by mixing absorption and scattering agents
into RTV12A. Since RTV12A is too viscous, the agents settle down to the bottom of the cured
silicone. Then we tried mixing agents into RTV12C. This has resulted in homogeneous mixing.
The recipe is in a complicated form since it was derived from the fixed breast shape of 1650 ml.
total carbon×10
Assuming we use 8 mg carbon in 10 ml solution, volume of carbon solution = 8 .
94
The amount of TiO2 for scattering is
Example
Breast shape phantom with µa = 0.045 cm−1 , µ0s = 7.5 cm−1 , Volume = 1650 ml.
(←(b)).
2. Put 8 mg of carbon black into (a). In this case, ‘zeroing’ with (a) on a scale and add carbon
4. Ice bath sonicate (a) and (b) for 1 hour. Ice bath : put ice cube in the sonicator. (because
RTV 12C has very low flash point) Sonication helps breaking the aggregated particles down.
5. Meanwhile, measure 50 ml of RTV12C into a beaker and measure 1500 ml of RTV12A into
a mixing pale.
6. After the sonication, pour 3.2 ml of (a) and all of (b) into the beaker with RTV12C. Mix
well.
7. Pour the mixture of RTV12C into RTV12A. Mix really well. Using high speed mixer is not
95
9. Place the mold inside the descicator or bigger container connected to vacuum pump to pump
* underlined quantities are to be calculated for different optical property using the formula.
* One needs to use glass bottles for RTV12C. It reacts with plastics.
96
Chapter 4
Breast
4.1 Introduction
The clinical DOT experiments and results on in vivo female breasts are presented in this chapter.
(The clinical motivation for spectroscopical imaging of breast cancer is described in Chapter 1.)
To familiarize the reader with clinical terminology of breast cancer cases, one section outlines
the breast tumor classification, diagnostic and treatment procedure. In the following section, we
quantified the optical properties of healthy in vivo breasts and their dependence on demographic
parameters to assess the inter-patient variation. A tumor contrast index was derived based on
three-dimensional DOT of breast with tumor and was compared with MRI and pathology. Then,
the treatment monitoring capability of DOT was explored for a patient going through neoadjuvant
chemotherapy with comparison to MRI. Lastly, the feasibility of optical blood flow measurements
97
for breast cancer diagnosis and detection was explored and its impact on oxygen metabolism esti-
The classification and the description of breast disease in terms of pathology is summarized from
[164].
There are four categories in benign breast disease which may mimic carcinoma: inflammations,
fibrocystic changes, proliferative breast disease, and benign tumors. Inflammations of the breast
include acute mastitis, periductal mastitis, mammary duct ectasia, fat necrosis and granulomatous
mastitis. Inflammations are rather uncommon. Among these, acute mastitis happens mostly in the
lactating period. Fat necrosis is the liquification of fat and formation of mass related to trauma,
previous surgical intervention, or radiation therapy. Fibrocystic changes include cysts (abnormal
membrane sac containing semitranslucent, turbid fluid), fibrosis (scarring due to rupture of cysts
into the adjacent stroma) and adenosis (abnormal formation or enlargement of glandular tissue).
These changes mimic carcinoma by producing palpable lumps, mammographic densities of calci-
fications, or nipple discharge. Proliferative breast disease such as epithelial hyperplasia, sclerosing
adenosis, and small duct papillomas increase the risk of cancer. Fibroadenoma is a new growth
composed of both fibrous and glandular tissue, which is the most common benign breast tumor.
The malignant breast adenocarcinoma is divided into two groups: In situ (noninvasive) car-
cinoma and invasive (infiltrating) carcinoma. In situ carcinoma includes ductal carcinoma in situ
98
which is the most common type of noninvasive carcinoma and lobular carcinoma in situ. Inva-
sive carcinoma includes invasive ductal carcinoma which is the most common type in all breast
cancers, invasive lobular carcinoma, medulary carcinoma, mucinous (colloid) carcinoma, tubular
carcinoma and invasive papillary carcinoma. Other rare variants of malignant breast cancer are
The frequency of breast disease is represented in the following table composed of the patient
Table 4.1: Frequency of breast disease from patient population seen in a surgical outpatient depart-
ment [164]
Screening mammograms are conducted to separate normal from abnormal findings and identify pa-
tients who need further evaluation. When the abnormalities are detected, a diagnostic mammogram
is performed.
The tissue samples from the lesion are obtained for more definite diagnosis by histopathology.
Fine needle aspiration (FNA) is a sampling of cells for microscopic examination by cytopathologist
for indication for malignancy. Core biopsy takes out a small punched-out tissue using larger needle
than the one used in FNA. Sometimes core biopsy is guided by X-ray mammography, ultrasound,
or MRI. Excisional biopsy is aimed at removal of the abnormal tissue. Samples from biopsy are
frozen and examined by the pathologist for preliminary diagnosis. If a malignancy is found, further
99
4.2.3 Histology grades
A modified Scarff-Bloom-Richardson grading system gives scores ranging from 1 to 3 in three cat-
and mitotic count. The “grade” is calculated by adding the scores. The survival rate decreases at
higher grades.
Mammography is the most common diagnostic imaging modality using X-ray radiation. It is most
effective in detecting abnormalities in fatty tissue. It is not useful for women under the age of 40
due to the high glandularity of their breasts. Imaging views often used for screening mamography
are cranio-caudal (head-to-foot), sagittal and oblique (cranio-caudal view shifted to include axillary
tail) views.
Ultrasound is currently used as an adjuvant for X-ray mammography, particularly for mammo-
graphic findings inconsistent in different views, for lesions not seen on the X-ray mammography
or to differentiate cysts. It is a primary diagnostic tool for women under the age of 30. Clinical
ultrasound gives a two-dimensional image which is perpendicular to the plane where the transducer
is placed.
MRI is still considered a research modality due to expense and low throughput. It has high
sensitivity (true-positive rate), but low specificity (false-negative rate). It may be the most effective
modality for invasive lobular carcinoma which is difficult to detect in mammography, and for
women with breast implants. False-negative interpretations have been reported particularly for
DCIS, due to occasional lack of tumor angiogenesis and poor enhancement. Three-dimensional
MRI images can be seen in either sagittal, axial, or coronal views, depending on the application.
100
(The tomographic views are illustrated in Figure 4.25(a)).
The treatment strategy varies depending on the grade and stage of the cancer. A localized car-
cinoma can be removed with local excision (lumpectomy), which is termed “breast conserving
surgery”. Depending on the extent of breast removal, there are different types of mastectomy. A
simple (total) mastectomy removes only the breast tissue. A modified radical mastectomy removes
the entire breast and axillary lymph nodes. A radical mastectomy refers to the removal of breast
Surgical procedures may be combined with adjuvant therapy such as hormonal therapy, chemother-
apy, and/or radiation therapy. In cases of locally advanced breast cancer, “neoadjuvant chemother-
apy” is performed prior to the surgery to reduce the tumor size and eradicate the cancer cells.
Radiation therapy uses ionizing radiation (X-rays and gamma rays) to affect cells by activation of
radiation-response genes, cell-cycle arrest and radiation-induced cell death. Chemotherapy uses
anti-cancer drugs to stop the growth or multiplication of cancer cells. Since these drugs work
most effectively on cells that divide rapidly, it kills cancer cells more effectively than normal cells.
However, cells in blood, mouth, intestinal tract, nose, nails, vagina and hair are also affected.
The optical and physiological properties of 52 healthy breast tissue [81] are quantified using the
We modeled our parallel plate geometry as an infinite slab, bounded on one side by a glass win-
dow (the detector plane) and on the other side by a white plate (the source plane). Previously [83]
101
we have shown that the chest wall affects the acquired data by effectively extending the diffusive
medium above the tank; in practice this makes our infinite slab approximation even better. Special
care was taken during the data acquisition to center the source and the detector grid on the breast
tissue; however, occasionally some regions from outside the breast tissue were also in the field of
view. In such cases we ignored grid points near the boundaries in our analyses. The solution for an
infinite slab is obtained using image sources and is well known (Equation 2.8). For chromophore
analysis, we employ a multi-spectral approach that utilizes a priori spectral information to reduce
We formulate the inversion problem in terms of CHbO2 , CHb , A and b using a multi-spectral ap-
proach. Thus we globally analyze the whole set of data, instead of analyzing each wavelength inde-
pendently and then combining the results. This approach substantially reduces the inter-parameter
cross-talk [86].
fluence and Φc is the calculated fluence. The sum is over the source detector pairs and all wave-
lengths. We use the Nelder-Mead simplex (direct search) method implemented in MATLAB func-
tion “fminsearch” to fit for the unknowns by minimizing χ2 . We then calculate blood saturation (=
StO2 ) and blood volume (=THC) from the relevant hemoglobin concentrations.
In order to establish a range of optical and physiological properties for the healthy breast tissue,
we now look at histograms of the fitted properties. The results are summarized in Figure 4.1. The
values define the expected physiological range. Mean and standard deviation are shown in Table
4.2.
102
# of Volunteers # of Volunteers
30
830
20 786
750
10
0
0.02 0.04 0.06 0.08 µa (cm−1) 0.1
30
20
10
0 ′ −1
2 4 6 8 10 12 14 16 µs (cm )
# of Volunteers # of Volunteers
40
Blood Saturation
20
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 x100%
30
Blood Volume
20
10
0
10 20 30 40 50 60 70 µM
Figure 4.1: Starting from top left shows µa , µ0s , blood saturation and blood volume histograms.
Table 4.2: Average optical properties and physiological parameters from the histograms.
103
100
90
80
70
B. Saturation (%)
60
50
40
30
20
10
0
10 20 30 40 50 60 70
B. Volume (µ M)
Figure 4.2: Blood saturation vs blood volume with the dashed lines indicating the ranges for normal
tissue from the mean and standard deviation of the healthy breast tissue.
It might be expected that tumors and other diseased tissue are distinguished by the relative
value of their total hemoglobin concentration and blood oxygen saturation. For example, malignant
tumors might be expected to have high blood volume with a low oxygen saturation since both a
higher blood content and higher metabolism are necessary to achieve tumor growth in proliferating
tumor cells [255]. Figure 4.2 shows blood saturation plotted vs blood volume for each breast.
The dashed lines indicate the range of blood volume and blood saturation for normal breasts from
Table 4.2. The error bars for each individual are obtained by the standard deviation of repeated
measurements of the same breast. In order to use endogeneous contrast effectively in DOT, tumor
tissue properties should lie in one of the “other” regions defined on this plot.
The normal state of the tissue covers a wide range of optical properties and it could be expected
that the heterogeneity of the tissues would induce a similar variation within an individual. This
could reduce tumor contrast. The ability to recover all the available physiological information is
crucial in order to relate parameters such as the blood volume and blood saturation to increase
specificity. Ideally an imaging instrument should combine a large number of wavelengths with
104
1.1
Breast
amp/<amp>
1
0.9
0 10 20 30
Minutes
1.1
Intralipid
amp/<amp>
1
0.9
0 10 20 30
Minutes
Figure 4.3: Normalized amplitude measured on breast and Intralipid sample by visiting same 17
points ten times every two minutes.
Apart from the well characterizable noise due to electronics, optics and positioning of the sources
and detectors, there is as additional noise in the measurements due to the changes in the physio-
logical state of the tissue during the measurements. Respiration, movement, heart beat, blood flow
downstream from the hanging breast are some factors that contribute to this “noise”.
In order to estimate the effect of physiological noise on our measurements, the scanning de-
tector was modified to visit the same point ten times every two minutes. This was repeated for
seventeen different points. The results are shown in Figure 4.3 where we plot time series of nor-
malized amplitude (phase not shown) obtained from measurements on an Intralipid sample and on
a breast tissue in vivo. We find that signals from the Intralipid sample are stable within 1 − 2%
whereas on the breast tissue the dispersion is up to 5 − 10%. This provides us with an estimate of
We also performed repeated measurements of the same breast with minimal movement of the
105
100
50
80
B. Saturation (%)
B. Volume (µ M)
40
60
30
40
20 20
10 0
15 20 25 30 35 40 15 20 25 30 35 40
BMI (kg/m2) BMI (kg/m2)
Figure 4.4: Left: Blood Volume vs BMI with a decaying exponential fit (correlation coefficient
0.42), Right: Blood Saturation (correlation coefficient 0.03) vs BMI
breast. By comparing repeated measurements, we find that the average standard deviation is 11%
for µa , 4% for µ0s , 4% for blood saturation and 5% for blood volume (see for example error bars in
Figure 4.2). These values are consistent with the variations in the amplitude and phase observed in
Figure 4.3.
As mentioned above, it might be expected that optical properties would show a variation with
demographics. Pogue et al [219, 220] used a similar system geared towards imaging and reported
that blood volume had a correlation with body mass index (BMI) which is related to the weight and
height of an individual. They did not report any strong correlation between any other quantities
and BMI or age. Cerussi et al [38, 39] reported weak correlation between blood volume and b with
age (as well as some other quantities that are not available to us).
Our findings are shown in Figure 4.4 for correlations with BMI. We see a similar correlation
of blood volume to BMI as reported by Pogue et al [219, 220]. A higher BMI indicates more
tissue fat content. In compositional studies a higher fat content correlates with a lower blood
106
14
12
µs ′ (cm−1)
10
4
15 20 25 30 35 40
BMI (kg/m2)
Figure 4.5: µ0s at 830nm vs BMI with a decaying exponential fit. Other wavelengths show similar
trends. The correlation coefficient is 0.46.
100
50
80
B. Saturation (%)
B. volume (µ M)
40
60
30
40
20
20
10 0
10 20 30 40 50 60 70 10 20 30 40 50 60 70
age age
Figure 4.6: Left: Blood Volume vs age, Right: Blood Saturation vs age
content [75, 105, 163, 255, 279, 286]. The correlation coefficient is significanly higher for blood
We also observed a similar correlation of BMI and µ0s as shown in Figure 4.5. Cerussi et
al [38, 39] showed that the scattering power and µ0s change with the fat content. BMI is also a
measure of the tissue fat content hence this result is in agreement with the their reasoning.
Figure 4.6 shows the correlation with age. Our results again indicate an agreement with Pogue
Our study had a different sensitivity than Cerussi et al who showed that there is considerable
107
change in breast properties with age. Their main result is that older breast tissue has a different
water and lipid content which effects the scattering and absorption properties of the tissue. Their
instrument was particularly sensitive to this aspect because it measured mainly the outer ∼1 cm of
the breast tissue and had many wavelengths which allowed accurate derivation of the wavelength
dependence of the scattering. Our results, by contrast, sample a larger tissue volume in transmis-
sion geometry and has a vast spatial information rather than spectral information. Therefore, we
sample the fatty tissue as well as the nodules and vasculature extensively. This is a weakness for
our measured properties in terms of correlating them with demographics, age and hormonal sta-
tus, however, it is imperative to establish these properties in our geometries since most imaging
systems rely on sampling a large volume of breast tissue. It has previously been reported that the
differences in the acquisition geometry changes the computed bulk properties in extensive studies
For three-dimensional reconstruction of optical and physiological parameters for breast cancer
patients, the second generation parallel-plate CCD-based hybrid instrument was utilized (Chapter
3, Section 3.6.2). The sequential nature of data acquisition produces a set of 2D continuous wave
CCD intensity maps. An example of a set of reference and in vivo breast raw data with a 786 nm
light source is shown in Figure 4.7. The two-dimensional CW intensity images taken with CCD
for three central sources of source plate (Figure 3.19(b)) is shown. There is a significant difference
in the source power between source positions and wavelengths due to non-uniform response of our
108
switch and fiber coupling. Each intensity image shown in Figure 4.7 is scaled by its own minimum
and maximum range. The absolute values between images are therefore not comparable.
The intensity images of the matching fluid in lower row of Figure 4.7 are characterized by its
concentric distribution of intensity with the maximum intensity occuring at the source position.
The intensity images of typical in vivo breast in upper row of Figure 4.7 show significant deviation
from concentric distribution observed from homogeneous matching fluid medium. This deviation
arises from the boundary effect between the breast and the matching fluid as well as the breast
heterogeneity (especially at the boundary where the breast is touching the detection glass).
breast measurements
11
cm
16 cm
Figure 4.7: Measured CW intensity images of breast (upper row) and matching fluid (lower row)
at source position 22, 23, and 24 (three central sources in the middle row of source plane shown in
Figure 3.19(b)).
109
A quick composite look at the attenuation level of the raw breast data with respect to the homo-
geneous reference data can be offered by constructing a transillumination picture. This approach
where widebeam or plane-wave illumination were used. We define the detected intensity at detec-
sillumination picture by summing the contribution of all sources, and normalizing with reference
ÃP !
Ns
I(s, rd )
T (rd ) = −log PNs s . (4.1)
s I 0 (s, rd )
Transillumination may reveal the signal contrast due to the presence of the tumor (Figure 4.8(a),
compare with subject # 68 results in Figure 4.13), but most of time it shows sensitivity to the
presence of surface blood vessels near detector plane (Figure 4.8(b)). In the latter case, the transil-
lumination picture can aid in identifying image artifacts due to surface structures. Black ellipsoidal
lines show the ellipsoid-approximated breast boundaries; outer line is the largest extent of breast
and inner line is the breast touching the detection plane. The attenuation level is higher in breast
than in the matching fluid for T > 0 and vice versa. In Figure 4.8(a), the tumor is located in
middle inner quadrant of the right breast, which correponds to the high attenuating enhancement
in this view. Note the attenuation level is relatively low inside the breast and T ∼ 0 far outside
of the breast, since the region is occupied by the matching fluid. The effect of perturbation due to
presence of breast is seen outside of the breast boundary due to its diffusive nature (i.e. T 6= 0 at
the boundary). In Figure 4.8(b), the tumor is located in middle outer quadrant of the left breast,
which correponds to the left outer side of the breast outline. The nipple is shown at the middle
110
lower and the blood vessel nearby the detection plane is clearly seen, which was also observed in
an outline photograph.
−2 0.3 −2 1.2
0.2 1
0 0
0.1 0.8
2 2
0
0.6
4 −0.1 4
0.4
6 −0.2
6
−0.3 0.2
8
−8 −4 0 4 8 −8 −4 0 4 8
(a) (b)
Figure 4.8: Transillumination pictures of two breasts. Black ellipsoidal lines show the ellipsoid-
approximated breast boundaries; outer line is the largest extent of breast and inner line is the breast
touching the detection plane. All dimensions are in centimeter.
Three-dimensional (3D) images are reconstructed from the data using the multi-spectral recon-
Section 2.4.3.
CHb , CHbO2 , CH2 O , Clipid and A. We fix the b value according to bulk values obtained from the
frequency-domain measurements. Since our particular imaging geometry involves space occupied
by breast and matching fluid, image segmentation of breast and matching fluid was used through-
out the calculation. The breast region was approximated as a 3-dimensional half-ellipsoid based on
its outline in the photo taken prior to measurement scan. To assign background and initial values to
the two regions, the bulk optical properties of matching fluid were derived from frequency-domain
111
reference measurements made on the box when it was completely filled with matching fluid. The
breast optical properties were derived from frequency-domain measurements in contact with breast
tion with multi-spectral approach [81] was utilized to fit directly for the bulk C Hb , CHbO2 , CH2 O ,
A and b in each region. The scattering power, b, was allowed to take on different values in the breast
nation of 31% water or bulk CH2 O estimated from frequency-domain measurement and 57% lipid
lipid absorption (from literature [163, 279, 286]).For the initial guess, C Hb , CHbO2 and A were as-
signed to the breast and the matching fluid based on the bulk measurements, e.g. zero hemoglobin
concentration in the matching fluid region. After the reconstruction of C Hb (r), CHbO2 (r) and
A(r), 3D images of total hemoglobin concentration (THC(r) = C Hb (r) + CHbO2 (r)), blood oxy-
genation saturation (StO2 (r) = CHbO2 (r)/THC(r)) and scattering µ0s (r, λ) = = A(r)λ−b(r) were
constructed.
Figure 4.9 shows the orientation of the three-dimensional reconstructed DOT image. A series
of slices along the y axis are arranged from left to right, from the source plane to the detector
plane, respectively. Each slice represents a 16 × 11 cm image in x-z plane, with the caudal-
cranial view (i.e. from feet to head, same as the CCD camera view). The orientation of each
image is such that the left side of the image slice is lateral (towards outer side of breast) and
112
right side is medial (towards middle of the breasts) for the left breast, and vice versa for the right
breast. For convenience of presentation, slices at selected spatial intervals are presented. Since the
reconstructed data on FEM nodes is interpolated to regular grid of 0.2 cm spacing, each slice has
Representative three-dimensional reconstructed images are shown for three malignant cancer cases
and one benign tumor case. For these cases, 31 % water and 57 % lipid concentration were assumed
and b distribution was fixed according to fitted value from FD measurement. We then reconstructed
CHbO2 , CHb and A. The first malignant cancer case shows a cancer near nipple area and the second
malignant case shows a cancer far away from the nipple. The third malignant cancer case shows
two cancers, of which one was subject to core biopsy. In the benign fibroadenoma case, DOT did
not detect the mass in the quandrant where mass was present. For each case, summary of radiology
113
and histopathology report is presented to give information about the tumor. Radiologists who have
experiences in breast cancer MRI compared DOT and MRI and confirmed the position of lesion.
Subject #103 was a 53 year old postmenopausal female with a 2 × 2 × 2 cm subareolar mass.
The cancer was located near nipple of the right breast as shown in Figure 4.10. The mammogram
measured a 1.5 cm mass, ultrasound measured the mass to be 1.1 × 1.3 × 1.8 cm, and MRI
measured it to be 2.2 × 2.1 cm. A representative sagittal MRI containing the nipple is shown
in right subfigure of Figure 4.10. (The enhancement near the chestwall was deemed to be DCIS,
but histology did not find DCIS.) The histopathology analysis after mastectomy revealed invasive
ductal carcinoma behind the nipple of 2 × 2 × 2 cm size and benign proliferative breast within the
5cm
Figure 4.10: Tumor location of subject #103 (near nipple). Left: location of tumor in frontal view,
Center: location of tumor in caudal-cranial breast outline photograph by CCD, Right : sagittal
MRI showing the enhancement of tumor due to high uptake of gadolinium.
Figure 4.11 shows the reconstructed DOT images of total hemoglobin concentration (THC),
blood oxygen saturation (StO2 ) and reduced scattering coefficient (µ0s ) at 786 nm. There is an
enhancement of THC and µ0s near nipple area which correponds to radiologist-confirmed cancer
position. StO2 does not show any enhancement corresponding to the mass, but the StO 2 value near
114
Total hemoglobin concentration (µM) 56
50
11 cm
source detector 5
plane 16 cm ∆ y = 1 cm plane
Figure 4.11: 3D reconstructed images of THC, StO2 and µ0s at 786nm of subject #103, right breast
with ductal carcinoma
Subject #68 was a 41 year old premenopausal female with an adenocarcinoma in her right breast.
According to X-ray mammography, ultrasound and MRI, the cancer was 4 cm in size and located
in the middle inner quadrant (3 o’clock). In our CCD camera view, the cancer is located at the left
High THC and µ0s enhancement of ∼2 cm size spans from y = 1 cm to 5 cm, corresponding to 3
o’clock position in Figure 4.13. There are slight shift between radiologist-confirmed mass position
and DOT-enhanced position. However, this could rise from the difference in the compression
scheme between the MRI and DOT as well as the variation in breast positioning.
Subject #111 was a 59 year old female with invasive ductal carcinoma in her right breast. She had
115
(a) (b)
Figure 4.12: Tumor location of subject #68. (a) In frontal view, the tumor is located in middle
inner quadrant about 3-4 o’clock of the right breast. (b) In caudal-cranial CCD view, it correponds
to the left side of the breast. All dimensions are in centimeter.
two masses at 2 o’clock (2.0 cm) and at 10 o’clock detected by X-ray mammography, ultrasound
and MRI. Ultrasound-guided core biopsy with 15 gauge needle on 10 o’clock mass was performed
8 days before optical measurement, which found invasive ductal carcinoma with high grade nuclei.
In the optical measurement, bruise due to recent core biopsy was observed in lower outer quadrant
In THC and µ0s images (Figure 4.15), two distinct enhancements corresponding to 2 and 10
o’clock are observed. The extension of 10 o’clock enhancement through the middle sections in y
direction could be attributed to the bruise effect induced by core biopsy. An interesting distribution
of StO2 is noticeable where the biopsied area near the detection plane has much lower oxygen
saturation.
Subject #95 was a 34 year old premenopausal female with a fibroadenoma in her right breast. X-
ray mammography detected a mass at 9 o’clock, 4 cm from the nipple of size (0.6 × 1.2 × 2.0
cm) as shown in Figure 4.16. The histopathology analysis done after the lumpectomy found 1.5
116
Total hemoglobin concentration (µM) 13
Adenocarcinoma
6
Blood oxygen saturation (%) 100
50
−1
µ ′ at 786nm (cm )
s
25
Figure 4.13: 3D reconstructed images of THC, StO2 and µ0s at 786nm of subject #68, right breast
with adenocarcinoma
cm fibroadenoma.
The reconstructed images in Figure 4.17 do not show noticeable contrast near the mass. This
may be due to lack of vasculature around the fibroadenoma. However, some fibroadenoma may
develop a vasculature. We need to measure more benign cases to assess the feasibility of using
optical contrast to distinguish between benign and malignant cases. There is an image artifact
appearing near source plane, which warrants the improvement of reconstruction scheme in the
future.
Subject #69 was a 56 year old postmenopausal female with invasive and in situ ductal carcinoma
in her left breast (Figure 4.18). X-ray mammography detected a mass behind the nipple of 1.5 cm
117
Figure 4.14: Tumor location of subject #111. (a) In frontal view, the tumors are located at 2 and 10
o’clock of the right breast. (b) In caudal-cranial CCD view, 2 o’clock mass correponds to the left
side and 10 o’clock mass corresponds to the right side of the breast. The biopsied 10 o’clock mass
is associated with core-biopsy induced bruise extending through lower outer quadrant, as shown in
dark color in this outline photograph. All dimensions are in centimeter.
spiculated mass associated with pleomorphic calcifications (category 5). MRI also found spicu-
lated mass of 2 cm in size in the subareolar position. The histopathology analysis done after the
mastectomy reported firm 2.1 cm area with adjacent but separate area of 0.4 cm, which are invasive
constructed CH2 O as well as CHbO2 , CHb and A for subject #69. (Descriptions of single-spectral
and multi-spectral approaches are in Chapter 2.) The contrast arising from the water concen-
tration has been observed by other groups and its physiological implication has been empha-
sized [140, 259]. Our previous analysis was concentrated in reconstructing C HbO2 , CHb and A
due to limited number of wavelengths. However, it is important to consider the water concentra-
tion in the reconstruction and it is our future direction to increase light sources sensitive to water
region. For the existing data, the reconstruction of additional parameter C H2 O was explored in the
context of comparing single-spectral and multi-spectral approaches. Lipid concentration was fixed
as 57%. The reconstructed images of THC, StO2 , µ0s at 786nm and CHbO2 using single-spectral
approach are presented in Figure 4.19 and those using multi-spectral approach are presented in
118
Total hemoglobin concentration (µM) 30
50
−1
µ ′ at 786nm (cm )
s
30
11 cm
source detector 5
plane 16 cm ∆ y = 1 cm plane
Figure 4.15: 3D reconstructed images of THC, StO2 and µ0s at 786nm of subject #111, right breast
with multiple ductal carcinomas.
Figure 4.20. In both cases, invasive ductal carcinoma is detected by the enhancement of THC and
µ0s . However, the reconstructed water concentration using the single-spectral approach show nega-
tive concentration, which was probably to compensate a false increase in THC. The multi-spectral
method provides a more robust water concentration within the physiological range.
In order to quantify tumor contrast, an average (p̄) and a standard deviation (σ p ) was calculated
T HC, StO2 , µ0s in each image. The tumor region was defined by p > p̄ + 2 × σ p , since values
greater than 2 × σp have a 95% chance to be different from the average with the assumption of a
gaussian distribution. For StO2 , threshold of p < p̄−2×σp was considered based on tumor hypoxia
hypothesis; this did not yield any tumor region. Tumor regions estimated from THC and µ 0s were
119
Figure 4.16: Tumor location of subject #95. (a) In frontal view, the tumor is located at 9 o’clock
of the right breast. (b) In caudal-cranial CCD view, 9 o’clock mass correponds to the right side of
the breast. All dimensions are in centimeter.
averaged to define the average tumor region. Relative THC (rT HC = T HC tumor /T HCnormal )
was calculated by averaging THC in the tumor region and outside the tumor region as defined
above. Relative µ0s (rµ0s ) and relative StO2 (rStO2 ) were defined in the same way. rT HC error
bars were estimated based on standard deviation of T HCtumor and T HCnormal . Relative µ0s (rµ0s )
and relative StO2 (rStO2 ) error bars were defined in the same way. We define an optical index
rT HC·rµ0s
(OI) which is a combination of the tumor contrast by OI = rStO2 to explore the possibility
of maximizing optical contrast. The concept of the optical index was first introduced in our com-
munity by Tromberg group [40]. In other imaging modality, it is a common practice to devise an
index to maximize the contrast. Individual values of rT HC, rStO 2 , rµ0s and OI for breast cancer
Table 4.3: Tumor contrast rT HC, rStO2 , rµ0s and Optical index OI for individual subjects are
presented.
120
Total hemoglobin concentration (µM)
36
O
Fibroadenoma
18
Blood oxygen saturation (%)
100
50
−1
µ ′ at 786nm (cm )
s
15
Figure 4.17: 3D reconstructed images of THC, StO2 and µ0s at 786nm of subject #95, right breast
with fibroadenoma.
The tumor contrasts (rT HC, rµ0s , rStO2 ) from twenty-one subjects with carcinomas are plot-
ted in a bar graph in Figure 4.21. The standard errors for total of 22 carcinomas are shown as error
bars in Figure 4.21. Note that rStO2 is around 1 and does not vary much. Combined contrast
index OI is 2.63±0.29.
Using three-dimensional DOT, we observed high tumor contrast compared to adjacent normal
tissue for 21 subjects with breast carcinomas. High THC contrast of tumor has also been reported
by other groups [70, 140, 144, 145, 221, 222, 297, 300], and has been supported by histopathologic
analysis of microvessel density counts [222]. The scattering contrast is still illusive since no careful
comparable histopathologic analysis has not been done for scattering contrast and also due to the
absorption and scattering crosstalk issues. Nevertheless, the hypothesis that the increased number
of scattering organells due to proliferation of cells and the increased fibrosis would increase the
scattering support our finding. Some groups also see some scattering contrast [70], but not to
121
Figure 4.18: Tumor location of subject #69. (a) In frontal view, the tumor is located behind the
nipple of the left breast. (b) In caudal-cranial CCD view, it correponds to slightly above the nipple.
the extent we found. Since our CW wavelengths were not optimal, we do expect the presence
In the next section, the dynamic changes in the DOT-derived physiological tumor contrast with
4.5.1 Introduction
Jakubowski et al [140] have recently demonstrated the capability of diffuse optical spectroscopy
(i.e. as opposed to diffuse optical imaging) for monitoring neoadjuvant chemotherapy in a breast
cancer patient. (The clinical motivation for neoadjuvant chemotherapy monitoring is described in
Chapter 1, Section 1.2.) This important paper introduced a new clinical application to the field.
However, quantification of breast cancer properties from spectroscopic data alone requires assump-
tions about tissues (e.g. homogeneous media, etc.) which reduce the fidelity of their results [26].
In addition, the remission measurement geometry used in their experiments is primarily useful for
122
Total hemoglobin concentration (µM)
50
50
µ ′ at 786nm (cm−1)
s
20
5
Water concentration (%)
100
Figure 4.19: 3D reconstructed images of THC, StO2 , µ0s at 786nm and CHbO2 of subject #69, left
breast with invasive carcinoma using single-spectral method. Lipid concentration was fixed as 57
%. Note the negative water concentration at the cancer site.
The potential of diffuse optical imaging as a chemotherapy monitoring tool has not yet been
explored. We present a case study which demonstrates the feasibility of this approach. We have
utilized a 4-wavelength near-infrared hybrid DOT instrument with continuous wave transmission
and frequency-domain remission detection for breast imaging [62]. Our subject had a locally
advanced invasive ductal carcinoma, and underwent 4 cycles of Adriamycin plus Cytoxan and 4
cycles of Taxotere prior to surgery. DCE-MRI were performed at 3 time points throughout the
therapy. After completion of Adriamycin cycles, we tracked the subject with DOT at 3 time points.
123
Total hemoglobin concentration (µM)
26
50
µ ′ at 786nm (cm−1)
s
20
5
Water concentration (%)
50
25
Figure 4.20: 3D reconstructed images of THC, StO2 , µ0s at 786nm and CHbO2 of subject #69, left
breast with invasive carcinoma using multi-spectral method. Lipid concentration was fixed as 57
%.
were reconstructed. We found that tumor volume and total hemoglobin tumor-to-normal contrast
decreased over the course of neoadjuvant chemotherapy. Furthermore, tumor volume changes
measured by DOT showed good correlation with DCE-MRI measurements of the same subject.
4.5.2 Methods
ductal carcinoma in her left breast at the Hospital of the University of Pennsylvania. The therapy
124
3
2.5
Tumor/Normal
2
1.5
0.5
0
rTHC rStO2 rµ ’ OI
s
Figure 4.21: Tumor contrast of 22 carcinomas. rT HC, rStO 2 and rµ0s are the relative ratio
HC·rµ0s
between tumor and normal. OI is an optical index defined as rTrStO 2
.
600 mg/m2 , regimen denoted as AC) followed by docetaxel (Taxotere, 100 mg/m 2 , regimen de-
noted as T ). Herein we will refer to the treatment as “AC” followed by “T ”. Each cycle was
taken at 3 week intervals. She participated in the MRI research study CALGB150007/150012 :
“Contrast Enhanced Breast MRI and Correlative Science Studies to Characterize Tumor Response
in Patients Undergoing Neoadjuvant Chemotherapy for Locally Advanced Breast Cancer (I-SPY)”
(PI: L. Esserman, MD, MBA). The timing diagram for chemotherapy and imaging measurements
DCE-MRI measurements were performed at the following time points: one week before chemother-
apy (pre-chemotherapy), week 12 following completion of AC, but prior to initiation of Taxotere
(T ) therapy, and week 23 following the completion of Taxotere therapy, but prior to surgical tu-
mor removal (mastectomy). MRI of the breast was performed at 1.5T (General Electric, Signa,
Milwaukee, WI) using in-house sagittal compression receiver coils [135]. At each time point,
125
Surgery
Adriamycin+Cytoxan Taxotere
Chemotherapy
-0 3 6 9 12 15 18 21 24 weeks
Imaging
MRI DOT
T1-weighted spoiled gradient echo imaging of the affected breast performed before, and twice
The patient underwent simple mastectomy following the 8 th chemotherapy cycle. The surgical
sample was inspected grossly and margins were inked in a standard fashion. The sample was then
oriented anatomically, and sagittally sectioned to correspond to the MRI imaging plane. Gross
inspection revealed an indurated area of abnormal, firm tissue in the superior breast measuring 1.5
cm in anterior-posterior dimension. This area was evaluated histologically with Hematoxylin and
Eosin staining.
DOT instrumentation used for measurements is described in Chapter 3, Section 3.6.2. Before each
DOT measurement, informed consent was obtained from the patient in accordance with University
of Pennsylvania Institutional Review Board. Based on the tumor location identified by palpation
(Figure 4.23(a)), the breast position with respect to the viewing window was optimized (Figure
126
4.23(b)). Then a soft compression was applied to hold the breast in a stable position. The com-
pression distance was fixed at 7.5 cm. A snapshot of the breast outline was taken by the CCD
camera before filling the box with the matching fluid (Figure 4.23(b)). After filling, the diffuse
optical image scan was conducted for 12 minutes. After human subject measurements, reference
optical measurements were performed on the box filled completely with matching fluid. For the
reference measurements, a silicone (RTV12 with carbon black and TiO 2 ) block was placed on top
of the box to extend the diffuse medium in a manner analogous to the subject’s chestwall, and
thus avoid signal saturation due to the air boundary. DOT measurements were performed at 10,
14, and 19 weeks after the first AC cycle (see Figure 4.22). These time points correspond to after
the fourth AC (4th chemotherapy), the first Taxotere (5th chemotherapy), and the third Taxotere
2 cm
5 cm
11 cm
16 cm
(a) (b)
Figure 4.23: Tumor location. (a) According to X-ray mammogram, ultrasound and MRI, the
primary tumor was located at the 12 o’clock position in the left breast. (b) Photo of breast outline in
caudal-cranial view (feet to head view). The location of the tumor was approximated by palpation
prior to DOT measurement and is indicated by a circle.
127
4.5.2.3 MRI Data Analysis
Tumor measurements by MRI were performed through analysis of enhancing tumor pixels. Sub-
intensity projections in the sagittal and axial planes were obtained to facilitate accurate tumor mea-
surement in three orthogonal dimensions. A radiologist with breast MRI experience measured the
tumor in three orthogonal planes. Tumor volume was then computed assuming an ellipsoid shape.
The transillumination pictures of breast at different chemotherapy time points are presented in
Figure 4.24. The transillumination picture offers a quick composite look at the attenuation level of
the raw breast data with respect to the reference data as described in Section 4.4.1. It also enabled
Transillumination at 830 nm
After 4th
Chemotherapy
After 5th
Chemotherapy
After 7th
11 cm
Chemotherapy
16 cm
Figure 4.24: Transillumination of breast at 830 nm normalized with respect to reference measure-
ment, offering a composite view of attenuation level.
128
4.5.2.5 DOT Data Analysis: 3D Reconstruction
The details of the data analysis scheme is described in previous Section 4.4.2. The reconstruction
volume was a 16 cm × 7.5 cm × 16 cm region, extending into the chestwall area. In this volume, a
finite element mesh with 58087 nodes was used by the finite element method based forward solver
grid (total 984 detection points, Figure 3.19(b)) with 3 mm spacing for each source. Hemoglobin
concentrations CHb (r), CHbO2 (r), and the scattering prefactor A(r) were chosen as unknowns to
be reconstructed while other variables such as water concentration, C H2 O (r) and lipid concentra-
tion, Clipid (r) and scattering power, b(r) were fixed as described below. For image segmentaion
using a geometric constraint, the breast region was approximated as 3D ellipsoid based on Figure
4.23(b). µbackground
a inside the breast was then fixed as a combination of 15% water (estimated
from frequency-domain measurement of bulk CH2 O ) and 57% lipid absorption. µbackground
a for
the matching fluid region was fixed at the fitted bulk µa of the matching fluid. The unknowns to be
reconstructed were CHb , CHbO2 and A. After the reconstruction of unknowns at each voxel r, 3D
images of THC(r), StO2 (r) and µ0s (r) at 786 nm were constructed.
As shown in Figure 4.25(a), there are three standard orientation of views in tomography. Since
the compression schemes of MRI (sagittal) and DOT (axial) are different, a true one-to-one image
In order to compare the tumor positions obtained by MRI with those obtained by DOT, one
must derive a transformation relating coordinates in the MRI axial image to coordinates in the DOT
axial image. We developed a simple scaling transformation for this purpose, which by its nature
129
Sagittal view
Coronal
tumor max
rMRI Y MRI
Axial
x
max
z Z MRI
y
Figure 4.25: (a) Orientation of MRI view. (b) A sagittal view of MRI (top) and DOT (bottom)
showing differences in compression. A linear scaling transformation scheme was utilized to find
the tumor center with respect to the nipple in DOT image (rtumor tumor
DOT ) from MRI rM RI .
cannot account for all of the deformations arising from breast compression in orthogonal directions.
The schematic of the transformation is illustrated in Figure 4.25(b). First, the breast dimensions
of corresponding ‘central’ image slices (i.e. slices containing the nipple) were compared. We
derived linear scale factors from the ratio of breast length and breast width in these corresponding
max /Z max , where X max , Y max , Z max is the longest linear dimension of the breast in the
γ = ZDOT M RI i i i
130
We located the tumor center and tumor boundaries in the MRI image by finding the region with
high intensity due to large gadolinium uptake. We then rescaled the tumor center coordinate r tumor
M RI
to rtumor
DOT by multiplying by scaling factors (i.e. α, β and γ). The tumor boundaries are defined
in this rescaled fashion as well. Importantly, the tumor in the DOT image lies approximately
within the volume defined by the rescaling based on pre-chemotherapy MRI. We note this scaling
approach assumes linear deformation and does not account for elastic differences between tumor
and surrounding tissue. Future models will be developed to account for elastic variations and
To define a tumor volume in the DOT images, a threshold based on the standard deviations of the
reconstructed parameters was introduced. First, DOT parameters were decomposed into µ a and µ0s
at 690, 750, 786 and 830 nm. Then for each image (T HC, StO 2 , CHb , CHbO2 , and µa , µ0s at the
4 wavelengths) an average (p̄) and a standard deviation (σ p ) was calculated. The tumor region was
defined by p > p̄ + 2 × σp , since values greater than 2 × σp have a 95% chance to be different from
the average with the assumption of a gaussian distribution. The average tumor volume (and stan-
dard deviation) was calculated by averaging all tumor regions defined by all images except StO 2
(where the variation did not exceed 2 × σp ). Relative THC (rT HC = T HCtumor /T HCnormal )
was calculated by averaging THC in the average tumor region and outside the tumor region as
defined above. rT HC error bars were estimated based on standard deviation of T HC tumor and
T HCnormal .
131
4.5.3 Results
According to the X-ray mammography, ultrasound, and MRI, the primary cancer was located at the
12 o’clock position in the left breast as shown in Figure 4.23. The mammography and ultrasound
reported adjacent multiple masses, the largest being 2.1 × 2.2 × 2.1 cm in size at pre-chemotherapy
time points. Pre-chemotherapy DCE-MRI measured the tumor size to be 5.3 × 2.2 × 2.7 cm.
MRI has been shown to be more accurate in depicting the size and overall extent of tumor than
mammography or ultrasound [30, 92, 186]. Tumor size estimation by DCE-MRI at different time
points is summarized in Table 4.4. The approximate tumor location in prone position assessed by
palpation was ∼ 8.5 cm from the nipple along the breast contour. In our camera view (caudal-
cranial view), this tumor location corresponds to a distant upper central position from the nipple as
Table 4.4: Tumor size measured with DCE-MRI at different time points during neoadjuvant
chemotherapy (AC : Adriamycin + Cytoxan, T : Taxotere). Tumor volume was estimated by
assuming ellipsoidal tumor shape.
The transillumination images in Figure 4.24 show high signal attenuation at the surface blood
vessel and around the upper central region particularly for data taken after the 4 th chemotherapy
cycle. This surface blood vessel was first identified through observation while positioning and was
confirmed with DCE-MRI images. The transillumination picture is quite sensitive to such surface
features.
132
coefficient (µ0s ) at 786 nm and blood oxygen saturation (StO2 ) are presented in Figures 4.26, 4.27,
4.28 respectively. In each figure, images from left to right correspond to 3D DOT image slices
taken from source to detector planes. The DOT images corresponding to different time points
within the chemotherapy cycle, are arranged from top to bottom. Colorbar scales are fixed from 5
to 15 cm−1 for µ0s at 786 nm, and from 50 to 100% for StO2 . However, colorbar scales for THC are
not fixed and are adjusted at each time point to maximize the THC color contrast between tumor
and normal region. The THC scale was adjusted so its maximum value was twice its minimum
In the THC images (Figure 4.26) after the 4th chemotherapy cycle, a high THC region is
found in slices near the source plane (1 - 3 cm deep from surface) and near the upper central
part of the breast. The lesion position corresponds to the tumor location estimated initially by
palpation at 12 o’clock and 8.5 cm away from nipple (as shown in Figure 4.23). After the 5 th
chemotherapy cycle, the tumor region is still identified by THC contrast near the source plane and
in the upper central region. However, the contrasted region appears smaller than the corresponding
region after the 4th chemotherapy cycle. Also, the average THC decreased significantly (i.e. from
21.4 ± 1.4 µM to 9.1 ± 0.5 µM). The THC distribution after the 7 th chemotherapy cycle is more
homogeneous throughout the slices compared to previous chemotherapy cycles. Within the original
tumor margins, the high THC region shifts towards outside of the tumor, leaving a relatively low
THC region occupying most of the tumor extent. The average THC increased slightly between 5 th
Figure 4.27 exhibits a similar trend. It shows higher µ0s values in the tumor region, and this
region with high µ0s shrinks over the course of treatment. The µ0s range does not vary as dramati-
cally as the THC, and gives a higher contrast ratio between tumor and normal tissue. Slices near
133
After 4th Chemotherapy Total Hemoglobin Concentration ( µM )
30
6
After 7th Chemotherapy Total Hemoglobin Concentration ( µM )
16
11 cm
source
16 cm
detector 8
plane ∆ y = 1 cm plane
(head) (feet)
the detection plane are affected by artifacts related to large vessels. These artifacts bear close re-
semblance to the transillumination picture in Figure 4.24, which is sensitive to large blood vessels
on the surface near the detection plane. This effect is also apparent in the THC images, but to a
much lesser degree. Therefore, slices near the detection plane (i.e. within 1 cm) were excluded in
StO2 images in Figure 4.28 were relatively homogeneous and do not show contrast in the tumor
region (i.e. not exceeding 2 × σp ). Note, however, the overall StO2 value decreased significantly
after the 5th chemotherapy cycle and remained constant thereafter (i.e. from 81.2 ± 1.4% to 59.9
MRI images projected in sagittal and axial views are shown in Figure 4.29 for the pre-chemotherapy
point, after AC therapy (week 12) and after Taxotere therapy (week 23). The intensity ranges were
134
After 4th Chemotherapy µ ’ at 786 nm (cm−1)
s
15
5
−1
After 7th Chemotherapy µ ’ at 786 nm (cm )
s
15
source detector 5
plane plane
(head) (feet)
Figure 4.27: Three-dimensional reconstructed images of µ 0s at 786 nm. Image slices from source
to detection plane are presented at 1 cm intervals in caudal-cranial view, from left to right. DOT
images corresponding to after 4th, 5th and 7th chemotherapy were arranged from top to bottom.
fixed among images at different time points. The intensity of the DCE-MRI image is higher in the
tumor due to increased tumor vascularity and gadolinium contrast uptake. Before chemotherapy,
the tumor is clearly seen around 12 o’clock to 1 o’clock (upper quadrant in sagittal view, near center
in axial view). After completion of chemotherapy cycles, the size and intensity of the enhancing re-
5 - 6 cm non-enhancing soft tissue region with only a scattered punctate form of enhancement. The
majority of the visible mass at this time point was poorly enhancing and was deemed to represent
fibrosis. The tumor position and volume measured by DCE-MRI is summarized in Table 4.4.
Histology revealed extensive fibrosis with focal areas of inflammation, but with few vessels.
However, viable tumor cells, morphologically identical the original core biopsy specimen (obtained
prior to chemotherapy) were identified. They were diffusely scattered throughout the fibrotic region
135
After 4th Chemotherapy Blood Oxygen Saturation (%)
100
50
After 5th Chemotherapy Blood Oxygen Saturation (%)
100
50
After 7th Chemotherapy Blood Oxygen Saturation (%)
100
source detector 50
plane plane
(head) (feet)
Figure 4.28: Three-dimensional reconstructed blood oxygen saturation images. Image slices from
source to detection plane are presented at 1 cm intervals in caudal-cranial view, from left to right.
DOT images corresponding to after 4th, 5th and 7th chemotherapy were arranged from top to
bottom.
as individual cells and small groups. No macroscopical viable tumor mass was identified.
In both MRI and DOT, the tumor was found in 12 o’clock, many centimeters away from the
nipple. In an attempt to assess the correlation between MRI and DOT images, the simple scaling
transformation scheme described earlier was performed with the assumption that nipple and tumor
center are good common reference points. The DOT axial image slice corresponding to rescaled
axial image slice in Figure 4.30(a). The tumor center found in MRI (X) and DOT (square) agrees
well with employment of this transformation. To show the extent of the tumor in detail, the axial
DOT images with smaller y intervals (0.4 cm) are shown in Figure 4.30(b). Higher THC contrast
in DOT are distributed in three-dimensions approximately within the tumor margins defined from
136
Sagittal view Axial view
Pre-chemotherapy
After completion of
AC cycles
18 cm
After completion of
Taxotere cycles 7.5 cm
x
z z
y
9 cm
7.5 cm
Figure 4.29: Dynamic-contrast enhanced MRI images of left breast. From top to bottom, rep-
resentative sagittal and axial slices along highest tumor contrast are shown for each time point :
pre-chemotherapy, after completion of AC cycles, and after completion of Taxotere cycles.
The tumor volume was defined independently for MRI and DOT. MRI tumor volume was de-
termined by a radiologist, whereas DOT tumor volume was determined by thresholding from the
distribution of DOT parameters. Figure 4.31(a) shows the decrease of tumor volume with progres-
sion of chemotherapy. MRI results show the tumor size decreased significantly after completion
of AC cycles. There is a general trend of decreasing tumor volume in MRI and DOT respectively.
The degree of change is different between MRI and DOT, but direct comparison is not possible
because the measurement time-point mismatch. The coinciding time-point is just after the 4 th
137
x
max
Z
max X MRI
MRI
(a)
Source y=3.2 cm 15
plane (y=0) ∆ y = 0.4 cm
(b)
Figure 4.30: Correlation of tumor position in three-dimensions. (a) Tumor center position from
MRI image (a X in the axial slice) is transformed by a linear scaling scheme to the DOT coordinates
(a square in the corresponding DOT axial slice). The scaled tumor center (a square) lies within a
high total hemoglobin concentration region. (b) The scaled tumor boundaries around the tumor
center is superimposed in three-dimensional DOT images. Axial images shown with smaller y
intervals (0.4 cm) show the extent of the tumor in detail.
chemotherapy cycle (completion of AC), where tumor volume measured by DOT is greater than
tumor volume measured by MRI. We note that DOT tumor volume can be overestimated due to
the ill-posed nature of DOT, which introduces blurring [62]. In DOT, there was a significant tumor
Significant decreases of rT HC were observed after the 5th chemotherapy cycle, which is the first
Taxotere chemotherapy. The rT HC after 5th and 7th chemotherapy cycles are equivalent, but
138
accompanied with significant tumor volume decrease, implying tumor neo-vasculature regression
with chemotherapy.
Tumor Volume
35 1st 2nd 3rd 4th 5th 6th 7th 8th chemotherapy
30 AC cycles T cycles
25 MRI
Volume (cm3)
DOT
20
15
10
DOT
5 0.21 ± 0.10 cm3
0
0 3 6 9 12 15 18 21 24
Weeks after 1st Chemotherapy Cycle
(a)
THCtumor/THCnormal
1.4
DOT
1.3
rTHC
1.2
1.1
1
0 3 6 9 12 15 18 21 24
Weeks after 1st Chemotherapy Cycle
(b)
Figure 4.31: (a) Decrease of tumor volume quantified by DOT and MRI, (b) Change of rT HC in
the tumor volume. Significant decrease in rT HC is noted before and after the 5 th chemotherapy.
139
4.5.4 Discussion
tissue blood oxygenation (StO2 ), and scattering coefficient are useful for localizing, quantifying
and tracking breast cancer based on vascularity during neoadjuvant chemotherapy. THC and scat-
tering showed localized contrast between the cancer and the normal region. The THC increase in
tumor is expected due to angiogenesis accompanying tumor growth [255]. The enhancement of
scattering might be expected based on changes in nuclear size and on the increase in concentration
of organelles such as mitochondria due to the high metabolism in cancer cells [255]. Generally,
StO2 might have been expected to be lower in the tumor region due to high oxygen demand, how-
The localized changes of these physiological parameters over time clearly demonstrate the
dynamic imaging capability of the DOT method. The DOT measurements were carried out at time
points just after completion of AC (Adriamycin+Cytoxan) cycles and then just after the first and
third Taxotere cycles. The anti-vascular and apoptotic effects of taxane [104] are consistent with
the significant decrease in tumor vasculature as measured in the tumor THC contrast, the tumor
volume decrease, and the intensity decrease from DCE-MRI. Interestingly, the variations in average
THC appear to be qualitatively correlated with measurements of patient hematocrit over the same
time period, which varied from 38% to 33% to 36%. The trends observed by DOT and MRI are
consistent with pathologic findings about the effectiveness of chemotherapy. The carcinoma cells
in the post-chemotherapy stage were finely and diffusely dispersed in fibrous connective tissue
which represented the bulk of the residual mass. This remaining viable tumor was detected by MRI
as focal enhancements and by DOT as small but positive THC contrast. Even though there was
no significant spatial contrast in StO2 for this subject, a significant decrease of average StO2 was
140
observed over time; this type of information might also be useful for chemotherapy monitoring,
since tumor response can depend on the oxygenation of the tumor and surrounding tissues [265,
266].
Jakubowski et al [140] reported THC and H2 O as major contrast factors, but did not report the
behavior of scattering. In our case, water concentration was estimated by the frequency-domain
measurements and was then fixed for continuous-wave image reconstruction. In addition to fix-
ing the water and lipid concentration, we have tried fitting for the water concentration, but the
reconstructed water concentration did not change much from its initial value. This may be because
our measurements used only 4 wavelengths (< 900nm), all of which were outside of the water-
sensitive range. As for the scattering contrast, it is possible that some portion of this signal may
arise from absorption-scatter image cross-talk, since the wavelengths used to discern chromophores
and scattering were not optimized [58]. The scattering artifact near the detection plane resembled
the blood vessel in transillumination, and thus raises concerns about scattering-absorption cross-
talk. However, our 3D simulations with noise in the same geometry find less than 20% cross-talk
between THC and scattering, whereas the observed scattering contrast (averaged over tumor vol-
ume) amounts for more than a 60% increase over the normal tissues. We therefore suspect that the
observed scattering contrast may indeed originate, at least partially, from tumor physiology.
To improve quantification of tumor optical contrast and validate the DOT method in clinical
settings, several improvements are underway. Laser sources at optimal wavelengths for separation
of chromophore contributions including water and scattering should further improve quantification
accuracy. Currently, the imperfect tumor region correlation between MRI and DOT data arises
because the linear approximations we have made for rescaling image do not account for the shift
141
of tumor due to compression and deformation of the breast. Better modeling of the elastic defor-
mation of breast in different compression geometries is under investigation. For better correlation
between DOT and MRI, it would be ideal to synchronize measurement time and employ the same
compression schemes. In this present study, it was not possible to synchronize DOT and MRI
due to logistics of two separate research protocols and the constraints of patient availability. The
improved coordination of DOT and MRI is now being performed for neoadjuvant chemotherapy
patients. The measurement of additional pathologic parameters such as microvessel density or nu-
clei size will be invaluable in correlating with tumor vasculature and scattering factors and thus
confirming DOT findings; these parameters are not routinely measured in the clinic. Finally, with
respect to monitoring, more frequent DOT measurements should improve the therapeutic value of
this technique.
4.5.5 Conclusion
We have demonstrated 3D diffuse optical tomography for monitoring physiological tumor re-
sponses during neoadjuvant chemotherapy in a single patient with a locally advanced breast cancer.
We have also compared our results to dynamic contrast enhanced magnetic resonance imaging and
pathologic analysis of the same patient. Three-dimensional reconstructed total hemoglobin con-
centration and scattering images successfully localized tumors and quantified the tumor volume
decrease and the THC contrast decrease over the course of chemotherapy. These physiological
parameters, measurable by DOT, may help to improve our understanding of chemotherapy mech-
142
4.6 Optical measurement of Blood flow in breast cancer
Until now, blood flow has been a quantity that is inaccessible to deep tissue optical methods. Our
laboratory has pioneered the development of diffuse correlation spectroscopy (DCS) for measure-
ment of blood flow in deep tissues (for a recent review see Durduran [80]) carrying out extensive
validation and demonstrating its utility. In the following, we explore the blood flow contrast of
breast tumors following an approach to DOS advanced by Tromberg’s group [139,140] which uses
hand-held spectroscopic probes. They have demonstrated its utility in measuring both the static
tumor contrast as well as changes in contrast due to chemo-therapy, hormonal status and age. Such
hand-held probes are attractive since they allow development of portable instruments and therefore,
frequent measurements with little patient disconfort. Our approach in this work is similar and uses
a hand-held probe that is scanned over the normal tissue and the palpable tumor regions, except we
Blood flow in breast cancer is an important quantity to monitor which also provides a novel
contrast over methods that measure essentially tumor morphology. Some amount of differentiation
capability was demonstrated between malignant and benign tumors [170] as well as the ability to
monitor various therapies [71, 79, 172]. Therefore, we expect that diffuse correlation spectroscopy
to be of value since the instruments are inexpensive, portable and the signals are robust. It is further
known that the metabolic changes can preceed dimensionally measurable changes [31] which have
been accessible to traditional imaging or clinical palpation methods. DCS can also be readily
combined with DOS measuring oxygenation and therefore, calculating the oxygen metabolism of
There have been previous attempts to measure blood flow in the breast using PET [17,283,284],
color and power doppler ultrasound [59,149,170] and MRI [71]. PET studies were limited in extent
143
but have shown that blood flow tends to increase in malignant tumors. Ultrasound techniques
on the other hand were used from late 80s to end of 90s and have been inconclusive as to the
clincial utility of the technique. PET was limited because of its cost and availability, whereas
ultrasound techniques had poor signal-to-noise and low contrast. Ultrasound techniques are also
biased towards large vessels and therefore to issues such as arterio-venous shunting. MRI studies
In this investigation, we have recruited three subjects with palpable tumors, two subjects with
mammographically identified calcification and two healthy subjects. The measurements were car-
ried at Hospital of University of Pennsylvania and were approved by the Internal Review Board.
Briefly, subjects were asked to lay back in the supine position, thus flattening the breast and in-
creasing tumor accesibility. An experienced researcher marked the position and the extent of the
tumor on a transparency paper with a grid for the record of tumor position. Then she used the
hand-held probe shown in Figure 4.32 to scan in horizontal and vertical directions in 2 cm incre-
ments across the tumor. Two scan directions were used to check the repeability of the signal and
ensure that variations were not due to changes in the probe pressure. In the case of healthy volun-
teers, an arbitrary region was drawn as the tumor site and a measurement was obtained by scanning
across that region. This measurement provided information about heterogeneity of blood flow in
breast tissue. Average optical properties necessary for analysis were obtained from separate DOS
Details of the instrument are described elsewhere [80]. Briefly, a long coherence laser (Crysta
Laser, NV) operating in continous wave mode at 785nm is coupled to a 1×4 optical switch (Di-
con, CA) and used to serially switch between four source positions. Four fast, photon-counting
avalanche photodiodes (Perkin Elmer, Canada) coupled to four single mode fibers were used to
144
Hand-held
Probe
Tumor
2.5 cm
DCS Dets.
Sources
Figure 4.32: Hand-held probe with four source-detector pairs is scanned horizontally and vertically
in 2 cm increments spanning the estimated tumor region as well as the surrounding healthy tissue.
detect the intensity fluctuations of the surface speckles. The TTL output is fed to a four-channel
custom build correlator board (Correlator.Com, NJ) and the resulting intensity auto-correlation
functions are recorded by a computer. A complete set of data is acquired every 6 seconds and five
such sets are acquired at each position. For this study, we disregard the crossed source-detector
pairs and record the position of each of the four source-detector positions directly across from each
other (separation of 2.5 cm) from each scanned position. The recorded correlation functions are
then fit to a solution of the diffuse photon correlation correlation equation [80] to obtain an index
proportional to the blood flow. The results are normalized to the mean value of the measurements
of the healthy tissue and the standard deviation is reported as the error bar. We, therefore, report
Figure 4.33 shows four correlation curves from two patients. When blood flow increases, the
temporal auto-correlation function decays more rapidly. It is evident that the blood flow is larger
in the tumor region (compare dark and light curves) in both cases. Figure 4.34 shows horizontal
and vertical profiles from one malignant tumor and one healthy breast. There is very little variation
observable in the healthy breast, where as the blood flow increased in both directions over the
145
1.4 Tumor 1.4 Tumor
Normal Normal
1.3 1.3
g2 (τ)
g2 (τ)
1.2 1.2
1.1 1.1
1 1
−6 −5 −4 −3 −2 −6 −5 −4 −3 −2
log(τ) (sec) log(τ) (sec)
(a) (b)
Figure 4.33: Temporal auto-correlation curves measured in tumor (dark) and healthy (light) tissue
from two patients. Faster decay corresponds to increased blood flow.
tumor indicating that the observed contrast is due to the tumor and not because of the natural
In order to quantify the blood flow change in the tumors, we have used the estimated tumor
outline and tabulated the mean (± standard deviation) rBF in that region. Table 4.5 shows the dis-
tribution of the values for all subjects. Three groups are visible; (1) there is very little heterogeneity
in the healthy breast (2.7 % variation), (2) the blood flow of malignant tumors is increased to 230
% of healthy tissue, whereas, (3) there is only a moderate increase in benign tumors (to 153 %).
Although, the power of the statistics of this study is not enough to conclusively claim differentia-
tion, we note that these results are in qualitative agreement with previous Doppler ultrasound and
PET results [17, 59, 149, 170, 283, 284] where ∼ 470-550 % increases in blood flow were reported
in malignant tumors with smaller contrast in benign cases. In studies with larger populations, blood
flow indices were used to differentiate upto nine different types of breast diseases [170].
These findings clearly demonstrate that we are able to optically detect robust changes in blood
flow in palpable tumors. Futher studies with more source-detector pairs are now being undertaken
to analyze the potential partial volume effects that may influence our results. We note that these
146
350
Healthy
300 Patient
250
rBF
200
150
100
50
−6 −4 −2 0 2 4 6
Position
(a)
350
Healthy
300 Patient
250
rBF
200
150
100
50
−6 −4 −2 0 2 4 6
Position
(b)
Figure 4.34: Relative blood flow (rBF) scans from one patient with a malignant tumor and a healthy
volunteer are shown for both (a) horizontal and (b) vertical scans.
147
ID rBF (± std) (%) type
1 100.5 ± 13.4 healthy
2 105 ± 86 healthy (large std)
3 144 ± 21 benign, calcification
4 163 ± 26 benign, calcification
5 184 ± NA malignant
6 212 ± 98 malignant
7 298 ± 51 malignant
Table 4.5: Tabulation of relative blood flow (rBF) measured at the estimated tumor regions from
all subjects grouped as healthy, benign and malignant diseases.
palpable tumors are relatively superficial and previous optical studies [139, 140] have shown that
source detector separations around 2.5 cm can probe them in a repeatable manner. Additionally,
if we assume that the palpable region corresponds to roughly the same depth from the skin, the
partial volume effects are further divided out by normalizing to the healthy tissue blood flow. In
the future, we will acquire data with a hybrid instrument [80] in order to measure the oxygenation
and total hemoglobin concentrations changes simultanously and estimate the changes in oxygen
metabolism of the tumors. The instruments are built on small clinical carts and the study time is
relatively short (∼ 10 minutes). Therefore, it is feasible to acquire data at each patient visit and in
the triage area. We anticipate these methods will be clinically useful for therapy monitoring, dose-
adjustment and potentially for assesing the efficacy of the therapy from the first week, therefore,
Using DOT/DOS technique, we have quantified in vivo healthy breast properties and demonstrated
the detection and quantification of breast tumor contrast against the surrounding nondiseased tis-
sue. Also, treatment monitoring capability of three-dimensional DOT was demonstrated for a case
148
of locally advanced breast cancer going through neoadjuvant chemotherapy. Blood flow measure-
ments of breast shed light as an additional optical parameter to aid in classification of breast cancer.
The theoretical and experimental techniques involving DOT/DOS are constantly evolving to
better quantify the breast tumor properties. For our current CCD-based instrument, our reconstruc-
tion is largely based on CW component of data, where we need to consider crosstalk between ab-
sorption and scattering. Our current wavelengths are not optimal for separating this crosstalk [57].
However, the simulation using current wavelength only yielded 20 - 30 % crosstalk between THC
and scattering, which implies our scattering contrast is not wholly due to the crosstalk. Care-
ful tissue phantom studies using in vitro hemoglobin will help identify this issue, but the origin
of scattering contrast in in vivo measurement still is obscure. It is shown that microvessel den-
sity correlates well with total hemoglobin concentration measured by DOT [222]. Retrospective
histopathology correlation study with emphasis on the microvessle density and other parameter
highlighting the source of scattering would be beneficial linking the microscopic histology and
Another technical difficulty we have faced during clinical measurements is the current and
intrinsic limitation of parallel plate design to accommodate lesions near axillary tail and chestwall.
Also, near the chestwall, the side of the breast is exposed to the boundary of air and the matching
fluid, which needs careful modeling. We are currently in the process of adding optimal wavelengths
and optimizing the table design to enable more breast tissue to be inside our imaging volume
without being exposed to the air boundary. In the extreme case of breast cancer near axillary tail
In some in vivo measurements, the reconstructed images could be contaminated with source
and detector artifacts. These may be coming from source detector coupling mismatch between
149
the breast and the matching fluid. In addition to the use of source detector coupling fit, various
approaches are devised to reduce these effects which may degrade the quantification of breast
cancer. The quantification of breast cancer could be substantially improved using a priori spatial
information from other imaging modalities [33, 34, 165, 294]. In our laboratory, initial phantom
measurements incorporating the scanning ultrasound has shown good quantification capability.
We are currently in the beginning stage of new breast cancer imaging technology where we
are validating our results against the gold standard or the existing imaging modalities such as X-
ray mammography, ultrasound and/or MRI. In section 4.5, the correlation study between MRI
and DOT on a locally advanced breast cancer showed reasonable agreement even with use of
simple scaling approach. More elaborate modeling considering the elastic deformation of breast
is needed, which may require the hardware modification and careful coregistration scheme using
markers. Of course, the measurements of MRI and DOT in the same table would be ideal for
coregistering purpose. With careful design, CCD-based DOT system could be also incorporated
We have briefly mentioned the construction of the optical index based on the relative tumor
contrast of optical properties for carcinoma cases. Additional measurement such as blood flow
could be added into the index. Or the metabolism derived from the blood flow and saturation can
be added as another parameter. This concept of optical index could lead to identification of optical
properties which discern the benign and malign tumors. More in vivo data on various kinds of
150
Neoadjuvant chemotherapy setting has emerged as an ideal platform to demonstrate the moni-
toring capability of DOT. With recently established collaboration, more coordinated measurements
of DOT with MRI are being performed in our laboratory. However, more frequent DOT measure-
ments are desirable since the significant physiological changes relating to treatment efficacy may
happen within a week [140]. Then the treatment efficacy could be assessed in the similar ap-
proach taken by Yu et al [292], Wang et al [273] and Sunar et al [250]. We have recently acquired
DOT measurements with Indocyanine Green (ICG) injection for some neoadjuvant chemotherapy
patients. In addition of enhancement of optical contrast, the kinetics of ICG could reveal the infor-
mation on the blood flow characteristics in the breast tissue. With current instrument, we can only
follow the kinetics on fixed source position. More rapid instrument enabling switching between
multiple source position during 5 minutes interval would be ideal for ICG kinetics monitoring
purpose. Comparing the blood flow information extracted from ICG kinetics and the blood flow
151
Chapter 5
5.1 Introduction
The ultimate goal of fetal brain project is to develop an accurate non-invasive trans-abdominal
monitoring device for the fetal brain oxygenation state, while the fetus is still in the uterus. The
clinical motivation and brief history regarding the fetal oximetry project is described in the Chapter
1, Section 1.3. In the following sections, the feasibility of in utero fetal brain oxygenation moni-
toring is demonstrated using a pregnant ewe model with induced fetal hypoxia. Then the prospect
of translating this technology to clinical setting is discussed in the following section based on the
152
5.2 Trans-abdominal Near Infrared Oximetry of Hypoxic Stress in
5.2.1 Introduction
The goal of this study was to demonstrate the feasibility of trans-abdominal NIR spectroscopy
for detecting and quantifying fetal hypoxia in utero in a pregnant ewe model (n = 5). We have
built a multi-wavelength NIR frequency-domain instrument with the capability to perform NIR
photon diffusion measurements through tissue over a wide range of source-detector separations.
We also developed a two-layer numerical diffusion model (for the maternal and fetal layers) to
quantify fetal cerebral blood saturation in utero. Good agreement was found between fetal blood
saturation determined by the trans-abdominal NIR method, and arterial and venous fetal blood
saturation quantified from fetal blood samples using a hemoximeter (gold standard). We conclude
trans-abdominal NIR oximetry has the capability to quantify different degrees of hypoxia in the
Five pregnant ewes (132-144 days gestation) were evaluated in this study. The animals were han-
dled according to the National Institutes of Health guidelines of the Institutional Animal Care and
Use Committee. The protocol consisted of the following steps: (1) anesthesia and catheteriza-
tion of the pregnant ewe, (2) catheterization of the fetus, (3) catheterization of the ewe for aortic
occlusion, and (4) NIR photon diffusion measurements and fetal blood sampling during baseline,
153
Transabdominal probe
aortic occlusion
through catheter
with balloon
probe
detector fiber
maternal
source fiber
skin
d : distance between
probe and brain
amniotic sac
First, the pregnant ewe was anesthetized with halothane [187]. The carotid artery was catheter-
ized for maternal arterial blood sampling and blood pressure monitoring. Second, the uterus was
exposed by a mid-line abdominal incision, and a small hysterotomy was performed to expose the
fetus for catheterization. The left brachial artery was catheterized for fetal arterial blood sampling
and the right brachial artery was catheterized for fetal blood pressure monitoring. The jugular vein
was catheterized for fetal venous blood sampling. The fetal body was then placed back in the
uterus. The uterus was tied around the fetal neck (purse-string method) to expose the fetal head
for the trans-abdominal NIR measurements. The exposed head was placed directly underneath the
maternal skin and secured by suturing its ears to skin. The purse-string approach was employed
to minimize the effect of the uterus in this pilot investigation. Third, a catheter with an inflatable
balloon was inserted through the femoral artery of the pregnant ewe for aortic occlusion. Aortic
occlusion through the femoral artery of the pregnant ewe is expected to directly reduce uterine
154
blood flow to the fetus, but have a minimal effect on maternal oxygenation. Finally, the probe was
placed on the maternal abdomen directly above the fetal head as shown in Figure 5.1. The details
of the instrument are described in Chapter 3, Section 3.6.3. The NIR measurements commenced
and were performed continuously during the entire baseline-hypoxia-recovery cycle. Fetal arterial
and venous blood samples were drawn during the baseline NIR measurement. Then the balloon
was inflated to block uterine blood flow and induce fetal hypoxia. The balloon was inflated until
the fetal blood pressure dropped rapidly. The inflation was maintained for 209 ± 38 seconds. Then
the balloon was deflated and the fetus was allowed to recover. Blood samples were drawn from
fetus every 30 seconds during hypoxia and once after recovery. Maternal arterial blood samples
were sampled and checked periodically to ensure the maternal arterial saturation was not perturbed
by aortic occlusion.
The thickness of the maternal layer was measured with a caliper to be 4.0 ± 0.4 mm. The
thickness of the fetal skull was obtained postmortem and it was 5.0 ± 0.5 mm. Maternal arte-
rial oxygen saturation, fetal arterial oxygen saturation (SaO 2 ) and fetal venous oxygen saturation
(SvO2 ) were quantified from the fetal blood samples with an OSM3 Hemoximeter (Radiometer
Medical, Coppenhagen).
The general steps for analysis of the NIR photon diffusion measurements are as follows. The
reflected photon fluence Φ(µa (r, λ), µ0s (r, λ)) at the surface of the medium is measured. It depends
on absorption coefficient (µa ) and reduced scattering coefficient (µ0s ) and the separation between
the source and detector, r, at a given wavelength λ. µ a and µ0s are determined by minimizing χ2 =
|Φ Φc 2
P
Φb
m
− Φbc
| , where Φm is the measured fluence and Φc is the fluence calculated using a photon
m
155
diffusion model. Φbm and Φbc are the measured and calculated baseline fluences, respectively in the
initial normoxic state (i.e. before the hypoxic perturbation). After µ a is determined, concentrations
of oxy-hemoglobin (CHbO2 ) and deoxy-hemoglobin (CHb ) are calculated [291]. Total hemoglobin
concentration (THC = CHb + CHbO2 ) and tissue blood oxygen saturation (StO2 = (CHbO2 /THC)
The data analysis procedure employed in this study consisted of three steps: (1) construction of
Φm from NIR measurements, (2) formulation of Φc from two-layer diffusion model using a priori
spectral and spatial information, and (3) retrieval of fetal blood saturation from χ 2 minimization
Sliding window (4 second) averaging was used to smooth the amplitude (A m ) and phase (θm )
data measured over the time course of the normoxia-hypoxia-recovery cycle in each study. The
measured fluence Φm (r, λ) = Am exp(iθm ). The baseline fluence measurement Φbm was deter-
mined by averaging the first 100 data points from the amplitude- and phase-time curves during the
A numerical solution to the diffusion equation for a two-layer model was used to generate Φ c .
Two-layer diffusion models have been investigated by a number of researchers using analytical
[3, 29, 153, 211, 231] or numerical solutions [68, 295]. Here, the finite difference method was used
to numerically solve the diffusion equation for a three-dimensional, two-layer model of the in utero
system. In this model, the tissues unperturbed by aortic occlusion were approximated as the top
lay er whose thickness is d; this layer includes the maternal skin, fetal head skin and the f etal
skull. The fetal brain was incorporated as the bottom layer. Strictly speaking the fetal skin is also
influenced by hypoxic perturbations, but since thi s layer is quite thin (≈1 mm) it was considered
156
Incorporation of a priori spectral information about the absorbers and scatterers in the tis-
sue reduces the number of unknowns and the inter-parameter cross-talk between µ a and µ0s [81].
Mie scattering formula in this wavelength range, i.e. µ0s (λ) = Aλ−b where A and b are determined
by scatterer size and density. Instead of treating µa and µ0s at each wavelength as unknowns, we use
the absorber concentrations and scattering parameters A and b as unknowns in the Φ c calculation,
Table 5.1: The values of fixed parameters in data analysis. A was determined by b and µ 0786nm
s .
Since CHb and CHbO2 of fetal brain are the major parameters of interest and make the largest
contribution to the signal variation, we fixed all other parameters. Table 5.1 shows the fixed pa-
rameters for each layer of the two-layer model. In the top layer, d was fixed based on thickness
measurements of the maternal skin, fetal skin and fetal skull postmortem. The background µ bg
a (λ),
the scattering properties A and b, the baseline THC of both layers, and the baseline StO 2 of the top
layer were assumed based on values reported in the literature [179, 254, 286]. The StO 2 values of
the bottom layer were obtained from the hemoximeter measurements during the normoxic baseline
157
in each cycle. Specifically, the baseline StO2 value for the fetus (bottom layer) was determined
using a compartmental model [63] where StO2 is made up of 43% SaO2 and 57% SvO2 . Since
the aortic occlusion protocol is expected to perturb fetal hemodynamics only, the top layer THC
and StO2 were assumed to be the same as the baseline throughout the cycle. Once all the fixed
parameters were defined for the calculation of Φc , the MATLAB function, fminsearch, utilizing
the Nelder-Mead Simplex method (iterative method) was used for χ 2 minimization in order to ex-
tract fetal brain CHb and CHbO2 . In order to test the effect of the fixed parameters on the error
propagation in the two-layer diffusion algorithm, each parameter was varied within a range (≈ ±
25%) reported in the literature. An error analysis indicated that the two-layer diffusion algorithm
was relatively insensitive to variations in the majority of fixed parameters. Variation in these fixed
parameters resulted in at most 2% variation in the fetal blood saturation values. The assumptions
relating the arterial and venous contribution to the fetal baseline StO 2 were limiting parameters
in our calculations. Varying the arterial/venous contribution from 43/57% [63] to 30/70% [124]
resulted in 0.5-5% variation in calculated fetal StO2 , depending on the level of hypoxia. This
variation was the basis for the fetal StO2 error bars.
5.2.3 Results
at wavelengths 690, 786 and 830 nm. The amplitude and phase were normalized to the baseline
amplitude and phase. The effects of hypoxia are particularly evident in the amplitude vs. time
trace. The decrease in amplitude at 675 nm and increase at 830 nm indicates a relative increase in
158
(a)
1.4
1.2
Am / Abm
1
0.8
0.6
0.4
0 200 400 600 800 1000 1200 1400
Inflation Deflation (b) Time (sec)
0.06
θm − θm (rad)
675 nm
0.04 786 nm
0.02 830 nm
b
0
−0.02
−0.04
600 800 1000 1200 1400 1600 1800 2000
Time (sec)
Figure 5.2: Normalized measurements of amplitude (Am /Abm ) and the phase-shift (θm − θm b ) vs.
which is the expected trend for hypoxia. The amplitude at 786 nm, which is close to the isos-
bestic point of hemoglobin, exhibits an intermediate response. Time traces at other source-detector
separations showed similar trends, but with different magnitudes. No obvious physiological inter-
In Figure 5.3 two examples of the fetal blood saturation obtained from NIR trans-abdominal
spectroscopy and from fetal blood samples quantified with the hemoximeter are presented. There
is a decrease in blood saturation with inflation and an increase in blood saturation with deflation of
the balloon. The blood saturation determined from the two-layer model fit shows good agreement
with the hemoximeter results. Herein the two-layer diffusion model fits of blood saturation will be
159
(a) (b)
100 100
Inflation Deflation Inflation Deflation
Blood saturation (%)
80 80
40 40
Two−layer fit
Arterial
Venous
20 20
0 0
0 200 400 600 800 1000 1200 1400 100 200 300 400 500
time (sec) time (sec)
Figure 5.3: Two examples of the fetal blood saturation obtained from NIR trans-abdominal spec-
troscopy and from fetal blood samples quantified with the hemoximeter. ‘Two-layer fit’ is the
blood saturation quantified by NIRS using two-layer model fit, ‘Arterial’ is arterial fetal blood and
‘Venous’ is venous fetal blood measured by the hemoximeter.
The correlation between fetal blood saturation measured by trans-abdominal NIR spectroscopy
(NIRS) and hemoximeter was also examined. NIR blood saturations StO 2 were selected from
points in the time traces where fetal arterial and blood samples were withdrawn. Blood saturation
values were calculated using the previously described compartmental model [48, 63, 106] from
the fetal arterial and venous blood saturations (measured from the fetal blood samples using the
A linear relationship between the NIR and hemoximeter fetal blood saturation over a wide
range of blood saturation values is observed in Figure 5.4, with a correlation coefficient R equal to
The blood saturation data were obtained from normoxic, hypoxic and intermediate data points
(n = 47), collected from seven hypoxic cycles (n = 7). Notice that the variance in blood saturation
increases as the fetus progresses from a normoxic to a hypoxic state. Furthermore, there is a
poorer correlation between the NIRS and hemoximeter measured blood saturation in the lower
160
100
60
40
20
0
0 20 40 60 80 100
Blood saturation by Hemoximeter(%)
Figure 5.4: Linear relationship between the fetal blood saturation measured by NIR instrument and
hemoximeter over a wide range of blood saturation values with a correlation coefficient R equal to
0.76 (p < 0.01).
range of blood saturations. This may be because the blood saturation of the fetus in this range was
transient, so it was difficult to perfectly synchronize the NIR measurements and the blood sample
withdrawals.
The baseline state (before inflation of the balloon) and stable hypoxic state (while the balloon is
in the fully inflated state) was considered for further assessment of comparison. The difference be-
tween the baseline state and stable hypoxic state for the NIRS (∆StO 2 ) and hemoximeter (∆ShO2 )
blood saturations are compared in Figure 5.5. The data is shown for two groups of hypoxia. Mod-
erate hypoxic cycles (n = 3) are grouped with ∆StO2 = 30 ± 7% and severe hypoxic cycles (n =
4) are grouped with ∆StO2 = 60 ± 5%. A paired t-test analysis shows ∆StO2 and ∆ShO2 are
statistically similar within the moderate and the severe group (p < 0.05 for null hypothesis). An
unpaired t-test shows that the difference between moderate and severe ∆ShO 2 is significant (p <
161
70
∆ShO2 (Hemoximeter)
50
40
30
20
10
0
Moderate Hypoxic Severe Hypoxic
Group (N=3) Group (N=4)
Figure 5.5: Difference in blood saturation between the baseline state and stable hypoxic state for
the NIRS (∆StO2 ) and hemoximeter (∆ShO2 ) measured blood saturations. The data is shown for
two groups of hypoxia.
5.2.4 Discussion
This study demonstrates for the first time, the feasibility of using trans-abdominal NIR spec-
troscopy for detecting and quantifying fetal hypoxia in utero. The pregnant ewe model is a
widely used model for studying fetal physiology and was ideally suited for this proof-of-principle
study. Perturbations of fetal blood saturation were performed in a controlled manner and fetal
arterial and venous blood saturations determined from fetal blood samples served as a reliable
gold standard to which the NIR measured blood saturations could be compared. The multi-
wavelength, multi-separation NIR frequency domain instrument coupled with the numerical two-
layer diffusion model proved capable of retrieving fetal blood saturation in utero, accurately and
162
non-invasively. The frequency-domain technique is more effective than the previous CW tech-
niques [227, 228, 270]; the former is better able to decouple absorption from scattering, which
is important for quantifying oxy- and deoxy-hemoglobin concentrations in tissue. The two-layer
numerical diffusion model also represents a significant improvement over the widely used homo-
geneous model; the former model was clearly able to deconvolve fetal from maternal blood satu-
rations, rather than volume-averaging them. When the homogeneous model was used on this fetal
data under the same condition, it underestimated the change in blood saturation. Multi-wavelength
NIR photon diffusion measurements enabled the use of a priori spectral information to reduce
number of unknowns (absorbers) in the two-layer diffusion model. Finally, NIR reflectance mea-
surements at multiple source-detector separations probe different tissue depths, thus optimizing the
deconvolution of fetal from maternal signals using the two-layer diffusion model.
In this study, several modifications were made to the pregnant ewe model to simplify the fetal
hypoxia protocol. First, hypoxia in the fetus was indirectly induced through aortic occlusion of
the maternal femoral artery. Alternative approaches for inducing fetal hypoxia are to lower the
maternal blood saturation by lowering the fraction of inspired oxygen (FiO 2 ) or through umbilical
cord occlusion. These alternative approaches were evaluated in preliminary studies. The problem
with lowering maternal FiO2 was that both maternal and fetal blood saturations were affected
and therefore the perturbation was not unique to the fetus. With the umbilical cord occlusion
approach, fetal morbidity and mortality were significant. Another modification made in this study
was the uterine layer was removed from the field during trans-abdominal NIR spectroscopy of the
fetus. The protocol was simplified in this manner to first establish the accuracy of quantifying fetal
cerebral blood saturation, for the case in which the overlying layers are not affected by the hypoxic
perturbation. The success of this pilot investigation sets the precedent for future animal model
163
studies. In future studies, the complexity of the additional uterine layer and different perturbation
approaches will be investigated. Additionally, a larger number of animals will be studied to obtain
We now discuss the future outlook of the fetal brain project with focus on the traslation to human
case based on the experience gained from the preliminary C-section clinical data.
Preliminary clinical data were collected using a simple dual wavelength, dual light source frequency-
domain spectroscopy with one detector channel (Figure 5.6). Two 750 nm laser diodes (Sharp,
LT031MD and LT030MD) and two 780 nm laser diodes (Sharp, LT024MD) were amplitude mod-
ulated by a 70 MHz local oscillator (Wilmanco, VSA-70+13dBm). LT031MD was used for larger
source detector separation because of its higher power. The laser diodes were time-shared at a
frequency of 1.25 Hz. Two pairs of a 750 nm and a 780 nm laser diodes were coupled to an optical
fiber (diameter = 1mm, N.A. = 0.37) by a laser diode power combiner (Oz optics) respectively.
The laser outputs were balanced to give similar level by adjusting the attenuator embedded in the
power combiner. The laser outputs after an optical fiber were about 1 mW. The optical fibers
were mounted on a probe at two different distance from a detector. A photomultiplier tube (PMT,
Hamamatsu, H5783-01) was used for the detection. The detected signal is amplified by a series
dyne scheme based on a I&Q demodulator (Mini-circuits, MIQY-70D) with lowpass filters was
used to detect the amplitude and phase of the diffuse photon density wave (DPDW) [288].
164
Time share
750 nm Turbid Medium
(10 Hz) (10 mW)
( H.V. = 1000 V )
LD
LD amplifier
PMT
LD
Bandpass
combiner filter
Optical Fiber
LD ( 1 mm dia., 0.37 N.A. ) (60~90 MHz)
780 nm
(20 mW)
RF I
source I&Q
LPF
(70 MHz) Demodulator
A/D computer D/A
Q LPF
Figure 5.6: Schematic diagram of dual wavelength dual source homodyne frequency-domain spec-
troscopy
The coupling coefficients from detectors and light sources were corrected by reference mea-
surements on a solid homogeneous tissue phantom (made of lesin) with known optical properties
to correlate with the pre-partum transabdominal measurements on a pregnant woman (Figure 5.7).
An elective Cesarean section was chosen particularly for this purpose since the time between the
transabdominal measurements and delivery is more controllable than in a labor. When available,
the cord gas was analyzed for oxygen saturation. The protocol was approved by the Internal Review
When a patient consented to participate in the study, the location of fetal head was found by
hand-held ultrasound tranducer which was in regular use in the hospital. A pre-partum measure-
ment was done on the patient’s abdomen right after anesthesia was administered to her and before
the Cesarean section operation began. When the neonate was delievered, a post-partum measure-
ment on neonate’s forehead was performed. For rare occasions, we had access to the fetal forehead
165
right before the umbilical cord was cut by using sterilized field over the probe.
In the pre-partum measurement, a soft black rubber probe with two optical fibers for light
sources placed at 3 cm and 7 cm from a PMT respectively was used. TO-8 PMT was mounted
directly on the probe and was operated at high voltage of 1000 V to maximize the signal to noise
ratio at separation of 7 cm. The probe was placed directly over the fetal head after it was located
by ultrasound transducer. The ultrasound also provided information on the distance between the
maternal abdomen and fetal skull. In the post-partum measurement, a soft black rubber probe with
one optical fiber placed at 3 cm from an optical bundle coupled to a PMT was used on a neonate
forehead.
(a) (b)
(c) (d)
Figure 5.7: Illustration of clinical in utero measurements. (a) Transabdominal measurement on the
maternal abdomen, (b) Inside view of transabdominal measurement, (c) Direct fetal brain measure-
ments before detechment of the umbilical cord during Cesarean section, (d) Direct neonate brain
measurements after birth
It was feasible to use the instrument to follow this protocol in the Cesarean section environment
166
without causing much extra delay for the operation. However, the improvement in the probe design
would be much anticipated for the future study. Especially PMT was difficult to use in the operation
environment due to its sensitivity to ambient stray light. Fast shutter mechanism to protect PMT
from being exposed to the ambient light directly should be implemented as well as the general
The optical properties and blood oxygenation obtained from single-spectral semi-infinite ho-
mogeneous medium fit (as described in Section 2.3) and Beer-Lambert law are summarized in the
following. (A) is from the transabdominal measurements at source detector separation of 3 cm and
(B) is from that of 7 cm. (C) is from the neonatal forehead measurement after the umbilical cord
was detached. Typically, measurement (C) was carried out 17 ± 6 minutes after the birth.
0 0
N µ750nm
a µ780nm
a µs750nm µs780nm THC (mM) StO2 (%)
(A) 8 0.059±0.018 0.060±0.023 9.7±1.7 8.5 ± 1.9 0.085 ± 0.050 69 ± 17
(B) 20 0.060±0.018 0.059±0.019 7.0±1.3 6.8 ± 1.2 0.080±0.041 59±31
(C) 13 0.174±0.074 0.142±0.046 7.7 ± 3.9 8.0±3.4 0.24±0.083 44±20
measurements were not available from the beginning of the study ((A) and (B)). For some cases
where the neonate’s state needs care, we did not have access to the neonate measurements in (C).
Since the fetal head has higher absorption coefficient as shown in (C), µ a at separation 7 cm (B) is
The averaged µa s agree with those found in the literature. Fishkin et al [99] reported the normal
abdomen µa = 0.0626 ± 0.003 cm−1 and µ0s = 9.11 ± 0.15 cm−1 for source-detector separation
of 2.7 cm at 811 nm which is congruent with values found in (a). Gratton et al [112] reports human
forehead µa = 0.16 cm−1 and µ0s = 7.3 cm−1 at 715 nm. Matcher et al [179] reports human
167
forehead µa = 0.16 ± 0.01 cm−1 and µ0s = 9.4 ± 0.7 cm−1 at 800 nm for 14 adult subjects. The
averaged absoprtion coefficients at separation 3 and 7 cm (Table 5.2 (A) and (B)) do not exhibit
the difference. For individual patients, when the fetal head depth was less than 2.3 cm, µ a at 7 cm
was greater than µa at 3 cm as expected. However, µa contrast between the separations were quite
small. µ0s at 7 cm is consistantly lower than µ0s at 3 cm. Whether this is truly physiological result
or this is due to µa and µ0s crosstalk or coupling coefficient offset, it is rather ambiguous at this
point since there were only marginal number of measurements to satisfy the number of unknowns
and the coupling coefficients were assumed to be the same in the solid phantom and the tissue.
The effects of coupling coefficient have recently gained considerable interest. The magnitude of
coupling coefficient would be such that not to affect homogeneous fit too much but would be large
Neonate blood saturation is lower than expected. This has been observed in the animal stud-
ies on piglets and lambs [51] and also observed by Hueber et al [133]. They attributed this to
wavelength dependent effective background absorption coefficient whose origin could be model
mismatch.
For three cases, measurements on the fetal forehead right before the umbilical cord was de-
tached were available. In these cases, the blood saturation from the fetus correlated linearly in
positive direction with maternal blood saturation whereas the neonate saturation (C) was not. How-
ever, fetal blood saturation was much lower than maternal blood saturation in this case. Also, for
the case blood saturation was monitored within a few minutes after umbilical cord detachment,
blood saturation of neonate increased with time which was also monitored with pulse oximeter on
neonate’s finger. Same phenomena has been reported by Isobe et al [136]. From these observa-
tions, one can draw a hypothesis that the fetal blood saturation may be lower and increases after the
168
umbilical cord is detached. This suggests the difficulties in the original study design in correlating
the neonate blood saturation to transabdominal blood saturation unless the neonate is accessible
Figure 5.8 summarizes above observation by showing a bar graph of µ a and µ0s at 750 and
780 nm, THC and StO2 of transabdominal measurement (N=15) at separation 7 cm, fetal head
measurement (N=3), and neonatal head measurement (N=10). Transabdominal THC is much lower
than that of fetal head and neonate head. µ0s does not seem to be significantly different from each
Low neonate blood saturation indicates the need of incorporation of more wavelength to the
instrument. Lack of contrast in absorption coefficients at short and long separations indicates either
the measurements at source detector separation of 7 cm is not sensitive enough to the fetal signal or
the over-simplified homogeneous model of in-utero system is underestimating the effect of fetus.
Even though the comparison between averaged µa at source detector separation 3 cm and 7
cm did not yield much difference, this can be easily attributed to model mismatch (i.e. not using
two-layer model). The clinical data still reveal the encouraging trend in Figure 5.9, where the µ a
at source detector separation of 7 cm decreases with increase of the fetal brain depth measured by
the ultrasound. Since the absorption of fetal brain is higher than that of maternal layer, as the fetal
brain depth is smaller, the expected absorption from homogeneous fit would be higher due to more
In order to translate the technology developed in Section 5.2 to clinical settings, further improve-
ments must be considered. Implementation of the two-layer diffusion model required information
169
12
750nm 750nm
786nm
0.25 786nm
10
µ (cm ) 0.2 8
−1
µ ′ (cm )
−1
0.15 6
a
s
0.1 4
0.05 2
0 0
Transabdominal Fetal head Neonate head Transabdominal Fetal head Neonate head
100
Total hemoglobin concentration (mM)
0.15 40
0.1
20
0.05
0 0
Transabdominal Fetal head Neonate head Transabdominal Fetal head Neonate head
Figure 5.8: Summary of physiological parameters (µa , µ0s , THC, and StO2 ) at prior (transabdomi-
nal, N=15), during (fetal head, N=3) and after birth (neonate head, N=10).
about several variables: top layer thickness which served as a priori spatial information, and fetal
baseline blood saturation from the hemoximeter measurements. Other parameters were assumed
according to the literature values, and their variation resulted in minimal influence on the fetal
blood saturation calculation. In clinical studies, the top layer thickness can be measured by ul-
trasound. Other available parameters such as maternal arterial blood saturation can be utilized to
further constrain the variables. However, the baseline fetal blood saturation is not available in the
clinical environment. If the baseline fetal blood saturation had been overestimated, the algorithm
would have underestimated the severity of hypoxia in the fetus and vice versa. Absolute quan-
tification of baseline fetal blood saturation is possible with measurements at more source-detector
170
0.16
0.14
µa (cm−1) at 750 nm
0.12
0.1
0.08
0.06
0.04
0.02
0
1.5 2 2.5 3 3.5 4
fetal head depth (cm)
Figure 5.9: µa measured at source detector separation of 7 cm using transabdominal DOS vs fetal
brain depth measured by ultrasound
separations and more wavelengths [58] (with extensive calibration [25]). The other major difficulty
arises because the distance from the maternal abdomen to the fetal brain is 2-4 cm [227] in humans.
For the clinical situation, the two-layer approximation is especially crucial since the optical
properties of maternal layer and fetal head are distinctively different as well as the sheer dimension.
To explore the degree of influence in human case, two-layer system consisting of maternal layer
(mostly fat) and fetal head as shown in Figure 5.10 is simulated with optical properties assigned
171
Most of the signal gets diluted due to the presence of meternal fat layer, and it is a function
of the layer thickness. This is illustrated in Figure 5.11, where the two-layer data from the optical
properties similar to clinical in utero system are generated using finite difference method. Then
two-layer data at source detector separation of 3 and 7 cm were fitted to the semi-infinite homoge-
neous solution (Equation 2.7). Also, fit using two-layer solution is presented. The homogeneous
fit underestimates the optical properties, resulting in somewhat in-between values between the fe-
tal head and maternal layer properties. The degree of degradation is more pronounced for the
shorter source detector separation as expected. Given the underestimation of optical properties by
homogeneous solution, it is clear that one needs to fully utilize the two-layer approximation.
0.18 12
0.16 11
0.14 10
two−layer fit
µa (cm−1)
µ′ (cm−1)
0.1 8
0.08 7
0.06 6
2 2.5 3 3.5 4 4.5 2 2.5 3 3.5 4 4.5
Thickness of the top layer (cm) Thickness of the top layer (cm)
(a) (b)
Figure 5.11: Effect of homogeneous fit on two-layer system. (a) Fitted absorption coefficient using
homogeneous solution and two-layer solution from the simulated data based on a two-layer model,
(b) Fitted reduced scattering coefficient.
This requirement of two-layer in turn needs optimization of the instrumentation towards mea-
suring at larger source detector separations than used in the animal study. The method to determine
172
the optimal source detector separation and the choice of detectors is illustrated in Appendix (Sec-
tion 5.4).
Even with the optimized instrument, the validation of this technique in the clinical setting is
still problematic. Additional animal study may be necessary before taking to the clinic. Especially
the animal model should be designed to reflect the similar thickness of maternal layer as the hu-
man case. Also, the effect of uterine layer should be assessed as well as exploring different type
of oxygen perturbation to the fetus. In the clinical C-section setting, correlating neonatal blood
oxygen saturation with umbilical cord is not effective since the neonate’s oxygen level rises rather
rapidly after birth. The best correlation would be through measurement on fetal head right be-
fore the umbilical cord is detached. For this measurement, better probe design which can readily
accommodate sterile requirement of operation is needed. The access to the hypoxic fetus would
be necessary in the end. Ideal setting for validation of this technique may lie in the fetal surgery
environment where the fetus is taken out of womb for surgical intervention and put back.
This appendix section is included to illustrate the methodology of estimating the optimal source
The two-layer model used is a three dimensional rectangular block (21 cm × 14 cm × 7 cm)
consisting of two different layers of different optical properties: a top layer simulating the maternal
layer and a bottom layer simulating the fetal head. Typically µ a1 = 0.06 cm−1 and µ0s1 = 8.0 cm−1
for the top layer and µa2 = 0.15 cm−1 and µ0s2 = 6.0 cm −1 for the bottom layer was used. The
173
120
2.0 cm
2.5 cm
100 3.0 cm
60
40
20
0
0 2 4 6 8 10 12 14
source detector separation (cm)
Figure 5.12: Noise model is incorporated into the two-layer signal to give overall signal to noise.
This reflects the signal to noise decrease with increase of source detector separations. Layer thick-
ness of 2 cm (solid line), 2.5 cm (dashed line), 3 cm (dotted line) are shown.
µa1 and µ0s1 were from the average clinical data from measurements at 3 cm separation (Table 5.2
(B),(C)). The boundaries were set as absorbing (n = 1.0) in all sides except one side (maternal
abdomen) with air to tissue index mismatched boundary (n = 1.4). After the optical properties of
each layer and the thickness of the top layer was assigned for each voxel, finite difference method
was used to solve the diffusion equation for this geometry [129].
A noise model described in Chapter 3, Section 3.2.2.6 is utilized to generate the noise which
depends on the signal level. In the following the case where P source = 20 mW, floss = 0.05, Adet
= 0.1 cm2 was considered to convert the fluence from forward model shown in Figure 5.12. The
overall signal to noise starts decreasing after source detector separation of 8 cm in both cases. By
increasing the factors Psource and ∆A, and decreasing floss shift the curve in Figure 5.12 to right
174
100 30
2.0 cm 2.0 cm
25
20 5
0 0
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
source detector separation (cm) source detector separation (cm)
(a) (b)
Figure 5.13: Fractional (a) amplitude difference in percentage and (b) phase difference in degree
are obtained between the amplitude from two-layer system and amplitude from homogneous sys-
tem with top layer optical properties at different source detector separations. It is a measure of
sensitivity to the bottom layer optical properites. Layer thickness of 2 cm (solid line), 2.5 cm
(dashed line), 3 cm (dotted line) are shown.
The sensitivity to the signal from the bottom layer is obtained by treating the contribution of fetal
signal on the overall signal is a perturbation caused due to its presence as contrasted to the case
where the whole region is homogeneous with optical properties of maternal layer. In the Figure
5.13, the fractional amplitude difference between two-layer case at top layer thickness of 2 - 4
cm and the homogeneous case of µa1 = 0.06 cm−1 and µ0s1 = 8.0 cm−1 are plotted against the
source detector separations. It shows that the fractional differences increases with increase of
separations. The fractional differences decrease with increase of thickness. As expected before,
the sensitivity to fetal signal increases with source detector separations and decrease with maternal
layer thickness.
175
500
2.0 cm
300
200
100
0
0 2 4 6 8 10 12 14
source detector separation (cm)
Figure 5.14: Differential signal to noise corresponding to fetal signal is plotted. Optimal source
detector separation exist at the maximum differential signal to noise. Layer thickness of 2 cm (solid
line), 2.5 cm (dashed line), 3 cm (dotted line) are shown.
Decrease of overall signal (Figure 5.12) and increase of sensitivity (Figure 5.13) with increase of
source detector separation competes with each other in detection of fetal signal. Multiplying the
overall signal and sensitivity gives the differential signal over noise which is the SNR from fetus.
In Figure 5.14, this effective SNR from fetus peaks at certain optimal source detector separation
where the effects of overall signal and sensitivity are balanced. The optimal separations increase
with increase of layer thickness. The optimal separations range around 8 - 10 cm. However,
the overall differential signal to noise ratio decreases with increase of layer thickness. Therefore
for thickness of 4 cm, even if the optimal separation is selected the measurement will give poor
Ultimately, the detectability of fetal signal relies on the ability of the model based inversion algo-
rithm to extract fetal information accurately. Therefore, two-layer diffusion model based inversion
176
0.16 10
homogeneous fit at 3cm
homogeneous fit at 7cm
0.14 9 two layer fit of top layer
two layer fit of bottom layer
0.12 8
µ′ (cm )
µa (cm )
−1
−1
0.1 7
s
0.08 6
0.06 5
0.04 4
1 1.5 2 2.5 3 3.5 4 1 1.5 2 2.5 3 3.5 4
Thickness of the top layer (cm) Thickness of the top layer (cm)
(a) (b)
Figure 5.15: Comparison between the homogeneous model fit and the two-layer model fit. In the
forward data, amplitude noise based on (a) and random phase noise of 0.1 o were added. Then it is
fitted to semi-infinite homogeneous analytic solution at source detector separation of 3 cm (cross)
and 7 cm (filled triangle). The optical properties of top layer (open circle) and bottom layer (filled
circle) are give by fitting to a two-layer numerical solution.
algorithm was developed using the finite difference method based diffusion equation solver as for-
ward model. In the clinical situation, we can take advantage of the fact that the thickness of the
maternal layer can be measured with the ultrasound technique. In order to retrieve the optical prop-
erties (i.e. µa1 , µ0s1 , µa2 and µ0s2 ), Nelder-Mead Simplex algorithm was used to update the optical
measured baseline fluence, Φc is calculated sample fluence, Φ0c is calculated baseline fluence, is
minimized.
The detectability of fetal signal, that is the ability of the inverse algorithm to retrieve accu-
rate fetal information in the presence of realistic noise factors is demonstrated for noise-model
incorporated simulated data and compared with fitting result from homogeneous model.
Random amplitude noise based on the realistic noise model was added on the forward data
177
generated for two-layer and homogeneous case using finite difference method for two-layer mea-
surement and baseline measurement. This simulates the experimental situation where the signal
deteriorates with increase of source detector separation. However, constant random noise of 0.1 o
was added to the phase. Differential signal between two-layer and homogeneous baseline was then
used as inputs for the inverse procedures. Crosses and filled triangles in Figure 5.15 are given by
Open circles and filled circles are top layer and bottom layer optical properties obtained by fitting
to two-layer model based numerical solution at two source detector separations. Typically, the
short separation is fixed at 3 cm and longer source separation is determined by optimal source de-
tector separations found in Figure 5.14. The semi-infinite fit is stable with respect to the amount of
noise added. However, the fitted values approaches to the values of top layer as thickness of layer
increases. Even at thickness of 1 cm, optical properties do not reach those of bottom layer. Also,
the contrast between 3 and 7 cm result is inherently small which is consistent with the findings
in the clinical data analysis. The two-layer model fit, on the other hand, is sensitive to the level
of noise in the data. In this case, if the layer thickness exceeds 3 cm, the fit becomes unreliable.
However, the overall value reflects much better on the values of bottom layer than homogeneous
fit. The effective signal to noise should be at least higher than 10 to give accuracy up to 10 % in
absorption coefficient estimation of bottom layer. Estimation of top layer optical properties were
always much better (less than 5 %) and it seems to be less affected by the layer thickness.
The methodology to predict the performance of the instrumentation for given system specification
described above can be applied to any sets of specification. It is shown for the given example that it
178
√
is feasible to detect fetal signal for thickness of 2 - 3 cm for NEP of system around 10 −13 W/ Hz
√
and thickness of 2 - 4 cm for NEP of system around 10−16 W/ Hz at source detector separations
So far, only a special case of two-layer was considered. However, the same methodology can
be applied for various optical properties and thickness variation. The optical properties and layer
thickness variation gives a range of signal. From a series of simulation of varying parameters, top
layer optical properties are found to be the main factors. Variation of µ a1 = 0.02 - 0.12 cm−1 at
µ0s1 = 8 cm−1 causes variation of amplitude signal in order of 107 across the optode separation of
3 - 10 cm. However, if the optode separation is chosen, the variation range is about 10 3 which can
be easily achieved experimentally. Variation of µ0s1 = 5 - 12 cm−1 at µa1 = 0.04 cm−1 is similar in
magnitude for fixed optode separation. The variation induced by the bottom layer optical properties
is at most order of 102 when the optical properties of top layer and thickness is fixed.
Therefore, when selecting the detector, NEP of the detector should be as small as possible,
detector area should be large and the dynamic range should be as large as possible. PMT is more
desirable to APD in terms of low NEP. However, dynamic range of PMT is smaller compared with
APD. When choosing the detectors, a detector with high gain and high anode sensitivity is favored
in terms of signal amplification. The chains of amplifiers can be selected such that the signal is
amplified not to be limited by A/D board resolution. Efforts to reduce the overall NEP should be
carried out for the detection system. Phase noise was not discussed in detail in this investigation.
However, phase noise contributes in raising the overall NEP. Heterodyne scheme may work better
than homodyne scheme in terms of phase noise. The comparison of two schemes are for future
investigation. Increasing modulated light source power increases the sensitivity to the bottom layer
detection. If the optode separations were fixed, modest dynamic range 10 3 − 104 per detection
179
position to give effective signal to noise much better than 10 is required to give reliable fit to two-
layer model. Coupling coefficients were not considered in much detail. However, their magnitude
may be large enough to obscure the sensitivity to fetal signal. Source detector combination should
implement scheme to fit for coupling coefficients reliably. For this at least 2 sources and 2 detectors
are needed. Increasing wavelength of light is expected to improve the fidelity in blood volume and
oxygenation evaluation.
180
Chapter 6
Summary
In this thesis, the motivation, theoretical background, experimental techniques and clinical results
on in vivo non-invasive breast cancer imaging and fetal brain oxygen monitoring using diffuse op-
tical tomography and spectroscopy were presented. For the breast cancer imaging application, the
quantification of tumor contrast based on total hemoglobin concentration and scattering through
lowing neoadjuvant chemotherapy. Also, a pilot study of blood flow measurement on breast cancer
cases is presented. The breast cancer imaging project is at an exploratory stage of finding the opti-
cal cancer contrast, constantly improving techniques for better quantification. The future direction
lies in the (1) improvement of the instrument focusing on multiple source detector positions, multi-
ple wavelength and speed, (2) incorporation of various techniques into reconstruction algorithm to
use a priori information and to reduce imaging artifacts, (3) active exploration of additional optical
parameters such as blood flow as well as establishing links with histopathology, (4) coregistration
with other imaging modality to complement each imaging technique, (5) threrapy monitoring for
181
treatment efficacy, and (6) active exploration of optical contrast agent usage which may make tran-
For the fetal brain oxygen monitoring application, accurate quantification of fetal hypoxia using
two-layer model based diffuse optical spectroscopy was demonstrated using a pregnant ewe model
for aortic occlusion induced fetal hypoxia. The translation of this technique to the clinical setting
involves (1) construction of an optimized instrument for deep tissue detection, (2) a validation study
on the animal model with thickness of maternal layer similar to the human case, and (3) a carefully
182
Glossary
—— SYMBOLS ——
² Extinction coefficient.
µa Absorption Coefficient.
ΦR
m Measured Photon-Fluence Rate of the reference.
ΦR
c Calculated Photon-Fluence Rate of the reference.
—— A ——
183
—— C ——
CH 2 O Water concentration.
CW Continuous Wave.
—— D ——
—— F ——
FD Frequency Domain.
—— H ——
H2 O Water.
184
HbO2 Oxygenated Hemoglobin.
Hb Deoxygenated Hemoglobin.
—— I ——
—— L ——
—— M ——
—— N ——
NIR Near-infrared.
—— P ——
—— R ——
rµ0s Relative reduced scattering coefficient (ratio between tumor and normal).
rStO2 Relative Tissue Oxygen Saturation (ratio between tumor and normal).
185
rT HC Relative Total Hemoglobin Concentration (ratio between tumor and normal).
—— S ——
—— T ——
TD Time Domain.
186
Bibliography
[1] Breast cancer. Early detection and prompt treatment are critical. Mayo. Clin. Health Lett.
[2] American cancer society, statistics for 2004 : Cancer facts & figures 2004.
[3] G. Alexandrakis, T. J. Farrell, and M. S. Patterson. Monte carlo diffusion hybrid model
for photon migration in a two-layer turbid medium in the frequency domain. Appl. Opt.
U. Glas, O. Jarlman, S. A. Larsson, and et al. Lightscanning versus mammography for the
detection of breast cancer in screening and clinical practice. a Swedish multicenter study.
[5] R. Aronson. Boundary conditions for diffusion of light. J. Opt. Soc. Am. A 12(11), 2532–
2539 (1995).
[6] S. R. Arridge. Photon-measurement density functions. Part 1: Analytical forms. Appl. Opt.
187
[7] S. R. Arridge. Optical tomography in medical imaging. Inverse Problems 15, R41–R93
(1999).
[8] S. R. Arridge and J. C. Hebden. Optical imaging in medicine: II. Modelling and reconstruc-
[10] S. R. Arridge and M. Schweiger. Inverse methods for optical tomography. In Information
[12] S. R. Arridge and M. Schweiger. Image reconstruction in optical tomography. Phil. Trans.
[13] S. R. Arridge and M. Schweiger. A gradient-based optimisation scheme for optical tomog-
[14] C. J. Baines and R. Dayan. A tangled web: factors likely to affect the efficacy of screening
images. In Optical Tomography and Spectroscopy of Tissue V volume 4955 pages 84–92.
SPIE (2003).
188
[16] R. J. J. Bartrum and H. C. Crow. Transillumination lightscanning to diagnose breast cancer:
emission tomography for in-vivo measurement of regional blood flow, oxygen utilisation,
and blood volume in patients with breast carcinoma. Lancet 1, 131–134 (1984).
[20] J. D. E. Beynon and D. R. Lamb. Charge-coupled devices and their applications. McGraw-
characteristics of 115 missed cancers later detected with screening mammography and the
[22] D. A. Boas. Diffuse photon probes of structural and dynamical properties of turbid media:
Q. Zhang. Imaging the body with diffuse optical tomography. IEEE Signal Processing
189
[24] D. A. Boas, L. E. Campbell, and A. G. Yodh. Scattering and imaging with diffusing temporal
[25] D. A. Boas, T. Gaudette, and S. R. Arridge. Simultaneous imaging and optode calibration
The accuracy of near infrared spectroscopy and imaging during focal changes in cerebral
[27] D. A. Boas, M. A. O’Leary, B. Chance, and A. G. Yodh. Scattering and wavelength trans-
duction of diffuse photon density waves. Phys. Rev. E 47(5), R2999–R3002 (1993).
[29] D. A. Boas and A. G. Yodh. Spatially varying dynamical properties of turbid media probed
with diffusing temporal light correlation. J. Opt. Soc. Am. A 14(1), 192–215 (1997).
[31] E. Bombardieri and F. Crippa. PET imaging in breast cancer. Q. J. Nucl. Med. 45(3), 245–
55 (2001).
[32] C. J. Bradley, C. W. Given, and C. Roberts. Disparities in cancer diagnosis and survival.
190
[33] B. Brooksby, S. D. Jiang, H. Dehghani, B. W. Pogue, K. D. Paulsen, C. Kogel, M. Doyley,
raphy breast image reconstruction with a priori structural information from MRI: Algorithm
(2003).
[37] M. C. Case and P. F. Zweifel. Linear Transport Theory. Addison-Wesley New York (1967).
combe, and B. J. Tromberg. Sources of absorption and scattering contrast for near-infrared
[40] A. E. Cerussi, N. Shah, D. Hsiang, T. Du, M. Compton, A. Fedyk, and B. J. Tromberg. Sta-
tistical analysis of over 60 malignant tumors measured in vivo by broaband diffuse optical
191
[41] B. Chance. Near-infrared images using continuous, phase-modulated, and pulsed light with
quantitation of blood and blood oxygenation. Ann. N. Y. Acad. Sci. 838, 19–45 (1998).
[42] B. Chance and R. Alfano. Optical tomography, photon migration, and spectroscopy of tissue
and model media: Theory, human studies, and instrumentation. volume 2389. SPIE (1995).
[43] B. Chance, S. Nioka, J. Kent, K. McCully, M. Fountain, R. Greenfeld, and G. Holtom. Time-
diffusive light and ultrasound imaging. Appl. Opt. 40, 6367–6380 (2001).
[46] X. Cheng and D. A. Boas. Diffuse optical reflectance tomography with continuous-wave
192
[49] R. Choe, A. Corlu, K. Lee, T. Durduran, S. D. Konecky, M. Grosicka-Koptyra, S. R. Arridge,
Diffuse optical tomography of breast cancer during neoadjuvant chemotherapy: a case study
Transabdominal near infrared oximetry of hypoxic stress in fetal sheep brain in utero. Proc.
parison between homogeneous and two-layer model for spectroscopic measurement of fetal
cerebral blood oxygenation in-utero. In Biomedical Topical Meetings (The Optical Society
evaluating response of primary breast cancer to systemic therapy. Breast Cancer Res. Treat.
4, 309–313 (1984).
from optical projections in light-diffusing media. Appl. Opt. 36(1), 180–213 (1997).
[54] S. B. Colak, M. B. van der Mark, G. W. Hooft, J. H. Hoogenraad, E. S. van der Linden,
and F. A. Kuijpers. Clinical optical tomography and NIR spectroscopy for breast cancer
193
[55] D. Contini, F. Martelli, and G. Zaccanti. Photon migration through a turbid slab described
by a model based on diffusion approximation. I. Theory. Appl. Opt. 36, 4587–4599 (1997).
Topical Meetings (The Optical Society of America, Washington, DC) ThD4 (2004).
Yodh. Diffuse optical tomography with spectral constraints and wavelength optimization.
signals from breast tumors. work in progress. Radiology 176(1), 175–180 (1990).
of the menstrual cycle on the red and near-infrared optical properties of the human breast.
changes in the absorption spectrum of the female breast. In Biomedical Topicals Meeting
194
plane transmission geometry: Evaluation of a hybrid frequency domain/continuous wave
clinical system for breast imaging. Med. Phys. 30, 235–247 (2003).
optical tomography of cerebral blood flow, oxygenation, and metabolism in rat during focal
arrangements for diffuse optical tomography: A singular-value analysis. Opt. Lett. 26, 701–
703 (2001).
[66] M. Cutler. Transillumination of the breast. Surg. Gynecol. Obstet. 48, 721–727 (1929).
[67] B. Davison and J. B. Sykes. Neutron Transport Theory. Oxford University Press London
(1957).
optical tomography: resolution in small-object imaging. Appl. Opt. 42, 3117–3128 (2003).
dimensional near-infrared tomography of the breast: initial simulation, phantom, and clini-
195
[71] J. P. Delille, P. J. Slanetz, E. D. Yeh, E. F. Halpern, D. B. Kopans, and L. Garrido. Invasive
[72] D. T. Delpy, M. Cope, P. van der Zee, S. Arridge, S. Wray, and J. Wyatt. Estimation of
optical pathlength through tissue from direct time of flight measurement. Phys. Med. Biol.
[73] C. Dildy. Intrapartum fetal pulse oximetry: Past, present, and future. Am. J. Obstet. Gynecol.
cancer with dynamic contrast-enhanced MRI of the breast. Eur. J. Surg. Oncol. 27, 617–620
(2001).
[75] F. A. Duck. Physical properties of tissue pages 320–328. Academic Press (1990).
[76] J. J. Duderstadt and L. J. Hamilton. Nuclear Reactor Analysis. Wiley New York (1976).
D. T. Delpy. Measurement of cranial optical path length as a function of age using phase
intensity modulated optical spectrometer for near infrared spectroscopy and imaging. In
Photon Migration and Imaging in Random Media and Tissues volume 1888. SPIE (1993).
196
[79] R. E. Durand. Intermittent blood flow in solid tumors - an under-appreciated source of ‘drug
[80] T. Durduran. Non-invasive measurements of tissue hemodynamics with hybrid diffuse optical
and A. G. Yodh. Bulk optical properties of healthy female breast tissue. Phys. Med. Biol.
and A. G. Yodh. Bulk optical properties of normal breasts and tissue phantoms obtained
with clinical optical imager. In Biomedical Topicals Meeting pages 386–388 (The Optical
and A. G. Yodh. Bulk optical properties of normal breasts and tissue phantoms obtained
with clinical optical imager. In Optical Imaging Workshop Bethesda, Maryland September
(1999).
Bulk optical properties of normal female breasts measured with a frequency domain clinical
197
noise. In Optical Tomography and Spectroscopy of Tissue III volume 3597 San Jose, Cali-
and B. Chance. Explicit inclusion of chromophore absorption and scattering spectra for dif-
fuse optical imaging and spectroscopy. In Optical Tomography and Spectroscopy of Tissue
[87] T. Durduran, A. G. Yodh, B. Chance, and D. A. Boas. Does the photon-diffusion coefficient
Yodh. Diffuse optical measurement of blood flow, blood oxygenation, and metabolism in a
human brain during sensorimotor cortex activation. Opt. Lett. 29, 1766–1768 (2004).
bayesian image reconstruction from sparse and noisy data sets: Near-infrared fluorescence
198
[92] L. Esserman, N. Hylton, L. Yassa, J. Barclay, S. Frankel, and E. Sickles. Utility of magnetic
resonance imaging in the management of breast cancer: evidence for improved preoperative
[93] S. Fantini, M. A. Franceschini, and E. Gratton. Effective source term in the diffusion equa-
tion for photon transport in turbid media. Appl. Opt. 36(1), 156–163 (1997).
[94] T. J. Farrel, M. S. Patterson, and B. Wilson. A diffusion theory model of spatially resolved,
steady-state diffuse reflectance for the noninvasive determination of tissue optical properties
[95] S. A. Feig. Decreased breast cancer mortality through mammographic screening: results of
199
[99] J. B. Fishkin, O. Coquoz, E. R. Anderson, M. Brenner, and B. J. Tromberg. Frequency-
domain photon migration measurements of normal and malignant tissue optical properties
[100] J. B. Fishkin and E. Gratton. Propagation of photon density waves in strongly scattering
media containing an absorbing ‘semi-infinite’ plane bounded by a straight edge. J. Opt. Soc.
[101] S. W. Fletcher. Breast cancer screening among women in their forties: an overview of the
mammography: Initial clinical results. Proc. Natl. Acad. Sci. U. S. A. 94, 6468–6473
(1997).
nonreassuring fetal heart rate patterns. Am. J. Obstet. Gynecol. 183(5), 1049–1058 (2000).
giogenesis sustains tumor dormancy in patients with breast cancer treated with adjuvant
Colditz, and D. J. Hunter. Association of age and reproductive factors with benign breast
200
tissue composition. Cancer Epidemiol. Biomarkers Prev. 8(10), 873–879 (1999).
crovascular systems and cerebral atrophy in spontaneously hypertensive rats. Brain Res.
[107] A. P. Gibson, J. C. Hebden, and S. R. Arridge. Recent advances in diffuse optical imaging.
[111] J. G. Graeme. Photodiode amplifiers: OP amp solutions. McGraw-Hill New York (1996).
scattering and absorption changes in muscle and brain. Phil. Trans. R. Soc. Lond., B, Biol.
[113] E. Gratton, W. Mantulin, M. J. Ven, M. van de Ven, J. Fishkin, M. Maris, and B. Chance.
201
[114] R. A. J. Groenhuis, H. A. Ferwerda, and J. J. Ten Bosch. Scattering and absorption of turbid
materials determined from reflection measurements. I. Theory. Appl. Opt. 22, 2456–2462
(1983).
detection and characterization of breast tumors. Appl. Opt. 42, 3170–3186 (2003).
of a time-domain optical mammograph and first in vivo applications. Appl. Opt. 38, 2927–
2943 (1999).
of cysts from solid tumors in the breast with diffuse optical tomography. Acad. Radiol.
202
[121] R. C. Haskell, L. O. Svaasand, T. Tsay, T. Feng, M. S. McAddams, and B. J. Tromberg.
Boundary conditions for the diffusion equation in radiative transfer. J. Opt. Soc. Am. A
the premature infant brain. Phys. Med. Biol. 47(23), 4155–4166 (2002).
[123] E. Heffer, V. Pera, O. Schutz, H. Siebold, and S. Fantini. Near-infrared imaging of the
(1998).
Photon Migration and Imaging in Random Media and Tissues volume 1988 pages 275–288.
SPIE (1993).
[126] W. H. Hindle, L. Davis, and D. Wright. Clinical value of mammography for symptomatic
women 35 years of age and younger. Am. J. Obstet. Gynecol. 180(6), 1484–90 (1999).
Bedside functional imaging of the premature infant brain during passive motor activation.
203
[128] M. J. Holboke. Manual: 3-D numerical reconstruction of optical properties using diffuse
[129] M. J. Holboke. Manual: Numerical solution to the diffuse photon density wave equation in
[131] M. J. Homer. Breast imaging: Pitfalls, controversies, and some practical thoughts. Radiol
[132] J. H. Hoogeraad, M. B. van der Mark, S. B. Colak, G. W. Hooft, and E. S. van der Linden.
First results from the phillips optical mammoscope. In Photon Propagation of Tissues III
haemoglobin spectrometry in the piglet brain during hypoxic stress, using a frequency-
[134] N. Iftimia, X. J. Gu, Y. Xu, and H. B. Jiang. A compact, parallel-detection diffuse optical
[135] E. K. Insko, T. J. Connick, M. D. Schnall, and S. G. Orel. Multicoil array for high resolution
204
[136] K. Isobe, T. Kusaka, Y. Fujikawa, M. Kondo, K. Kawada, S. Yasuda, S. Itoh, K. Hirao, and
ately after birth in the human neonate using full-spectrum near infrared spectroscopy. J.
[137] S. L. Jacques. Time resolved propagation of ultrashort laser pulses within turbid tissues.
[142] H. Jiang, K. D. Paulsen, and U. L. Osterberg. Optical image reconstruction using DC data:
205
[143] H. Jiang, K. D. Paulsen, U. L. Osterberg, and M. S. Patterson. Improved continuous light dif-
fusion imaging in single- and multi-target tissue-like phantoms. Phys Med Biol 43(3), 675–
693 (1998).
[144] H. Jiang, Y. Xu, N. Ifitimia, J. Eggert, K. Klove, L. Baron, and L. Fajardo. Three-
dimensional optical tomographic imaging of breast in a human subject. IEEE Trans. Med.
optical imaging of the breast with model-based reconstruction. Acad. Radiol. 9, 186–194
(2002).
spectral detection of pressure-induced changes in breast tissue. Opt. Lett. 28, 1212–1214
(2003).
[147] A. C. Kak and M. Slaney. Principles of computerized tomographic imaging. IEEE Press
[148] J. R. Kappa, H. D. Berkowitz, R. Seestedt, and B. Chance. Evaluation of calf muscle oxygen
content. using NIRS in patients with peripheral vascular disease. In Medical intelligence
noma: measurement of tumor response to primary medical therapy with color doppler flow
206
[150] M. Keijzer, W. M. Star, and P. Storch. Optical diffusion in layered media. Appl. Opt.
[151] K. Kerlikowske, J. Barclay, D. Grady, E. A. Sickles, and V. Ernster. Comparison of risk fac-
tors for ductal carcinoma in situ and invasive breast cancer. J. Natl. Cancer Inst. 89(1), 76–
82 (1997).
[152] K. Kerlikowske, D. Grady, J. Barclay, E. A. Sickles, A. Eaton, and V. Ernster. Risk of breast
cancer based on age and mammographic interpretation [ssee comments]. JAMA 276(1), 39–
43 (1996).
[153] A. Kienle and T. Glanzmann. In vivo determination of the optical properties of muscle with
time-resolved reflectance using a layered model. Phys. Med. Biol. 44, 2689–2702 (1999).
ing of absorption perturbation for diffuse optical tomography. Appl. Opt. 42, 3129–3144
(2003).
[155] M. V. Klibanov, T. R. Lucas, and R. M. Frank. A fast and accurate imaging algorithm in
ping for monitoring of breast cancer treatment during neoadjuvant therapy. Magn. Reson.
[157] D. B. Kopans. Breast Imaging. Lippincott-Raven Philadelphia and New York (1998).
207
[158] C. D. Kurth, J. L. Steven, L. M. Montenegro, H. M. Watzman, J. W. Gaynor, T. L. Spray, and
S. C. Nicolson. Cerebral oxygen saturation before congenital heart surgery. Ann. Thorac.
[160] L. Lapidus and G. F. Pinder. Numerical solution of partial differential equations in science
[161] J. Lavery. Nonstress fetal heart rate testing. Clin. Obstet. Gynecol. 25, 689 (1990).
[162] K. Lee, R. Choe, A. Corlu, S. D. Konecky, T. Durduran, and A. G. Yodh. Artifact reduction
and fibroglandular tissue volumes in the breasts of women 20-83 years old: comparison of
(1997).
[164] S. C. Lester and R. S. Cotran. The breast. In Pathologic Basis of Disease pages 1093–1119.
W. B. Saunders (1999).
208
[166] X. D. Li, B. Chance, and A. G. Yodh. Fluorescent heterogeneities in turbid media: limits
for detection, characterization, and comparison with absorption. Appl. Opt. 37, 6833–6844
(1998).
phy for biomedical imaging with diffuse photon density waves. Opt. Lett. 22, 1198–1198
(1997).
photon density waves in homogenous and heterogeneous turbid media: analytic solutions
[169] C. D. Mackay. Charge-coupled devices in astronomy. Ann. Rev. Astron. Astrophys. 24, 255–
283 (1986).
[170] H. Madjar, W. Sauerbrei, H. J. Prömpeler, R. Wolfarth, and H. Gufler. Color doppler and
duplex flow analysis for classification of breast lesions. Gynecol. Oncol. 64, 392–403
(1997).
in locally advanced breast cancer treated with neoadjuvant chemotherapy. J. Nucl. Med.
209
[173] F. A. Manning, C. R. Harman, and I. Morrison. Fetal assessment based on fetal biophysical
[174] V. A. Markel, V. Mital, and J. C. Schotland. Inverse problem in optical diffusion tomography.
III. inversion formulas and singular-value decomposition. J. Opt. Soc. Am. A 20(5), 890–
902 (2003).
tomography. IV. nonlinear inversion formulas. J. Opt. Soc. Am. A 20(5), 903–912 (2003).
[177] V. A. Markel and J. C. Schotland. Inverse problem in optical diffusion tomography. II. role
[179] S. J. Matcher, M. Cope, and D. T. Delpy. In vivo measurements of the wavelength de-
pendence of tissue- scattering coefficients between 760 and 900 nm measured with time-
[180] C. L. Matson and H. L. Liu. Analysis of the forward problem with diffuse photon density
waves in turbid media by use of a diffraction tomography model. J. Opt. Soc. Am A 16, 455–
466 (1999).
210
[181] T. O. McBride, B. W. Pogue, S. Jiang, U. L. Osterberg, and K. D. Paulsen. A parallel-
tumor oxygen status using human melanoma xenografts as a model. Cancer Res. 63, 7232–
7240 (2003).
[183] B. Monsees, J. M. Destouet, and W. G. Totty. Light scanning versus mammography in breast
anisms of light scattering from biological cells relevant to noninvasive optical-tissue diag-
[185] J. R. Mourant, T. Fuselier, J. Boyer, T. M. Johnson, and I. J. Bigio. Predictions and measure-
ments of scattering and absorption over broad wavelength ranges in tissue phantoms. Appl.
and I. Taylor. Staging of symptomatic primary breast cancer with MR imaging. Am. J.
[187] M. J. M. Nijland, U. Shankar, V. Iyer, and M. G. Ross. Assessment of fetal scalp oxygen sat-
uration determination in the sheep by transmission pulse oximetry. Am. J. Obstet. Gynecol.
211
[188] A. M. K. Nilsson, C. Sturesson, D. L. Liu, and S. Andersson-Engels. Changes in spectral
[189] J. Nocedal and S. J. Wright. Numerical Optimization. Springer New York (1999).
[191] V. Ntziachristos and B. Chance. Probing physiology and molecular function using optical
[192] V. Ntziachristos, B. Chance, and A. G. Yodh. Differential diffuse optical tomography. Opt.
of highly heterogeneous media. IEEE Trans. Med. Imaging 20, 470–478 (2001).
[194] V. Ntziachristos, A. G. Yodh, M. Schnall, and B. Chance. Concurrent MRI and diffuse
optical tomography of breast after indocyanine green enhancement. Proc. Natl. Acad. Sci.
[196] M. A. O’Leary. Imaging with diffuse photon density waves. Ph.D. disseration University of
Pennsylvania (1996).
212
[198] M. A. O’Leary, D. A. Boas, B. Chance, and A. G. Yodh. Experimental images of het-
diffusive waves with other imaging modalities to enhance optical specificity. In Advances
[201] H. W. Ott. Noise reduction techniques in electronic systems. John Wiley & Sons New York
(1988).
and P. Beaudry. Multi-port time-domain laser mammography: results on solid phantom and
[203] M. S. Patterson, B. Chance, and B. C. Wilson. Time resolved reflectance and transmittance
for the non-invasive measurement of tissue optical properties. Appl. Opt. 28, 2331–2336
(1989).
domain reflectance for the determination of the scattering and absorption properties of tis-
[205] K. D. Paulsen and H. Jiang. Spatially varying optical property reconstruction using a finite
element diffusion equation approximation. Am. Assoc. Med. Phys. 22(6), 691–701 (1995).
213
[206] K. D. Paulsen and H. Jiang. Enhanced frequency-domain optical image reconstruction in
tissues through total variation minimization. Appl. Opt. 35, 3447–3458 (1996).
[207] Y. Pei, H. L. Graber, and R. L. Barbour. Influence of systematic error in reference states on
image quiality and on stability of derived information for DC optical imaging. Appl. Opt.
[209] A. Pera and A. K. Freimanis. The choice of radiologic procedures in the diagnosis of breast
mography to enhance the detection of breast tumors. J. Biomed. Opt. 8, 517–524 (2003).
[211] T. H. Pham, T. Spott, L. O. Svaasand, and B. J. Tromberg. Quantifying the properties of two-
layer turbid media with frequency-domain diffuse reflectance. Appl. Opt. 39(25), 4733–
4745 (2000).
surements of the female breast at different interfiber distances. J. Biomed. Opt. 9(6), 1143–
1151 (2004).
214
[213] A. Pifferi, P. Taroni, A. Torricelli, F. Messina, R. Cubeddu, and G. Danesini. Four-
wavelength time-resolved optical mammography in the 680 980-nm range. Opt. Lett.
scattering in breast tissue: Analysis of intersubject variability and menstrual cycle changes.
regularization improves diffuse optical tomography. Appl. Opt. 38, 2950–2961 (1999).
finite tissue volumes using diffusion theory. Phys. Med. Biol. 39, 1157–1180 (1994).
domain system for noninvasive tissue spectroscopy. J. Biomed. Opt. 1(3), 311–323 (1996).
http://www.dartmouth.edu/ biolaser/.
215
[220] B. W. Pogue, S. D. Poplack, T. O. McBride, S. Jiang, U. L. Osterberg, and K. D. Paulsen.
Breast tissue and tumor hemoglobin and oxygen saturation imaging with multi-spectral near-
infrared computed tomography. In Adv. Exp. Med. Biol. International Society on Oxygen
[223] A. Prentice and T. Lind. Fetal heart rate monitoring during labour - too frequent intervention,
[224] W. H. Press. Numerical recipes in C : the art of scientific computing. Cambridge University
[225] A. E. Profio and G. A. Navarro. Scientific basis of breast diaphanography. Med. Phys.
tribution of vastus lateralis blood flow and oxyhemoglobin saturation measured at the end
216
[227] N. Ramanujam, H. Long, M. Rode, I. Forouzan, M. Morgan, and B. Chance. Antepar-
through the fetal head in utero. J. Matern. Fetal Med. 8, 275–288 (1999).
migration through fetal head in utero using continuous wave, near infrared spectroscopy:
clinical and experimental model studies. J. Biomed. Opt. 5(2), 173–184 (2000).
[230] J. Ripoll and V. Ntziachristos. Iterative boundary method for diffuse optical tomography. J.
[232] J. Ripoll, R. B. Schulz, and V. Ntziachristos. Free-space propagation of diffuse light: Theory
[233] R. Roy and E. M. Sevick-Muraca. Truncated newton’s optimization scheme for absorption
and fluorescence optical tomography: Part I Theory and formulation. Opt. Express 4, 353–
371 (1999).
217
[234] R. Roy and E. M. Sevick-Muraca. Truncated newton’s optimization scheme for absorption
[235] R. Roy and E. M. Sevick-Muraca. Active constrained truncated newton method for simple-
time-resolved instrument for medical optical tomography. Rev. Sci. Instrum. 71, 256–265
(2000).
[238] J. M. Schmitt, A. Knuttel, and J. R. Knutson. Interference of diffusive light waves. J. Opt.
for fast functional optical tomography. Rev. Sci. Instrum. 73, 429–439 (2002).
[240] J. C. Schotland. Continous-wave diffusion imaging. J. Opt. Soc. Am. A 14(1), 275–279
(1997).
[241] J. C. Schotland, J. C. Haselgrove, and J. S. Leigh. Photon hitting density. Appl. Opt.
218
[243] M. C. Segel, D. D. Paulus, and G. N. Hortobagyi. Advanced primary breast cancer: as-
(1988).
[244] E. M. Sevick, B. Chance, J. Leigh, S. Nioka, and M. Maris. Quantitation of time- and
185 (1992).
variations in optical and physiological properties of healthy breast tissue. J. Biomed. Opt.
[247] E. A. Sickles. Breast cancer detection with transillumination and mammography. Am. J.
[248] L. Spinelli, A. Torricelli, A. Pifferi, P. Taroni, G. M. Danesini, and R. Cubeddu. Bulk optical
properties and tissue components in the female breast from multiwavelength time-resolved
tion, oxygen saturation, and scattering measured in vivo by near-infrared breast tomography.
219
[250] U. Sunar, J. Zhang, J. Du, T. Durduran, C. Zhou, G. Yu, A. Kilger, H. Quon, R. Lustig,
L. Loevner, S. Nioka, A. G. Yodh, and B. Chance. Clinical responses of head & neck
terization and imaging using photon density waves. Opt. Eng. 32, 258–266 (1993).
[253] P. Taroni, G. Danesini, A. Torricelli, A. Pifferi, L. Spinelli, and R. Cubeddu. Clinical trial of
time-resolved scanning optical mammography at 4 wavelengths between 683 and 975 nm.
[254] P. Taroni, A. Pifferi, A. Torricelli, D. Comelli, and R. Cubeddu. In vivo absorption and
scattering spectroscopy of biological tissues. Photoch. Photobio. Sci. 2(2), 124–129 (2003).
[255] S. Thomsen and D. Tatman. Physiological and pathological factors of human breast disease
that can influence optical diagnosis. Ann. N. Y. Acad. Sci. 838, 171–193 (1998).
[257] P. B. Toft. Prenatal and perinatal striatal injury: A hypothetical cause of attention-deficit-
220
[258] A. Torricelli, L. Spinelli, A. Pifferi, P. Taroni, R. Cubeddu, and G. M. Danesini. Use
Changes in the findings of dynamic MRI by preoperative CAF chemotherapy for patients
with breast cancer of stage II and III: pathologic correlation. Oncol. Rep. 6(4), 727–732
(1999).
[262] H. G. van Staveren, C. J. M. Moes, J. van Marle, and v. M. J. C. Prahl, S A. Light scattering
in Intralipid-10% in the wavelength range of 400-1100 nanometers. App. Opt. 30, 4507–
4514 (1991).
[265] P. Vaupel and M. Hockel. Blood supply, oxygenation status and metabolic micromilieu of
breast cancers: characterization and therapeutic relevance. Int. J. Oncol. 17(5), 869–879
(2000).
221
[266] P. Vaupel, A. Mayer, S. Briest, and M. Hockel. Oxygenation gain factor: a novel parame-
ter characterizing the association between hemoglobin level and the oxygenation status of
[267] P. Vaupel, K. Schlenger, C. Knoop, and M. Hockel. Oxygenation of human tumors: evalu-
[268] P. Vaupel, O. Thews, D. K. Kelleher, and M. Hoeckel. Current status of knowledge and crit-
ical issues in tumor oxygenation. results from 25 years research in tumor pathophysiology.
fetal head in utero using continuous wave, near-infrared spectroscopy: development and
evaluation of experimental and numerical models. J. Biomed. Opt. 5(2), 163–172 (2000).
frequency-domain optical measurements in highly scattering media. Appl. Opt. 36(1), 170–
179 (1997).
222
[273] H. Wang, M. E. Putt, M. J. Emanuele, D. B. Shin, E. Glatstein, A. G. Yodh, and T. M. Busch.
[274] J. Wang, T. T. Shih, J. C. Hsu, and Y. W. Li. The evaluation of false negative mammography
from malignant and benign breast lesions. Clin. Imaging 24(2), 96–103 (2000).
[277] R. Weissleder and V. Ntziachristos. Shredding light onto live molecular targets. Nat Med
and K. D. Paulsen. Analysis of the microvasculature and tissue type ratios in normal vs.
benign and malignant breast tissue. Anal. Quant. Cytol. Histol. 26(3), 166–174 (2004).
[279] D. R. White, H. Q. Woodard, and S. M. Hammond. Average soft-tissue and bone models
Yasui, Y amd Taplin. Variation in mammographic breast density by time in menstrual cycle
among women aged 40-49 years. J. Natl. Cancer Inst. 90(12), 906–910 (1998).
223
[281] B. C. Wilson, M. S. Patterson, and S. T. Flock. Indirect versus direct techniques for the
(1987).
[282] B. C. Wilson, M. S. Patterson, and L. Lilge. Implicit and explicit dosimetry in photodynamic
ments of blood flow and exchanging water space in breast tumors using positron emission
tomography: A rapid and noninvasive dynamic method. Cancer Res. 52, 1592–97 (1992).
monoclonal antibody distribution and blood flow using positron emission tomography and
124 iodine in patients with breast cancer. Int. J. Cancer 47, 344–347 (1991).
ton. Mapping of hemodynamics on the human calf with near infrared spectroscopy and the
influence of the adipose tissue thickness. Adv. Exp. Med. Biol. 510, 225–230 (2003).
[286] H. Q. Woodard and D. R. White. The composition of body tissues. Br. J. Radiol. 59, 1209–
1219 (1986).
[287] Y. Xu, X. Gu, T. Khan, and H. Jiang. Absorption and scattering images of heteroge-
neous scattering media can be simultaneously reconstructed by use of dc data. Appl. Opt.
224
[288] Y. S. Yang, H. L. Liu, X. D. Li, and B. Chance. Low-cost frequency-domain photon migra-
tion instrument for tissue spectroscopy, oximetry, and imaging. Opt. Eng. 36(5), 1562–1569
(1997).
imaging of absorption and scattering distributions by a born iterative method. J. Opt. Soc.
[290] A. Yodh and B. Chance. Spectroscopy and imaging with diffusing light. Phys. Today
[291] A. G. Yodh and D. A. Boas. Functional imaging with diffusing light. In Biomedical Pho-
multiplexing system for in vivo diffuse light measurements of rapid cerebral hemodynamics.
[294] H. Zhao, X. Gu, and H. Jiang. Imaging small absorbing and scattering objects in turbid
225
[295] H. J. Zhao, Y. Tanikawa, F. Gao, T. Onodera, A. Sassaroli, K. Tanaka, and Y. Yamada. Maps
of optical differential pathlength factor of human adult forehead, somatosensory motor and
occipital regions at multi-wavelengths in nir. Phys. Med. Biol. 47(12), 2075–2093 (2002).
[296] J. X. Zhu, D. J. Pine, and D. A. Weitz. Internal reflection of diffusive light in random media.
[297] Q. Zhu, N. Chen, and S. H. Kurtzman. Imaging tumor angiogenesis by use of combined
near-infrared diffusive light and ultrasound. Opt. Lett. 28, 337–339 (2003).
[298] Q. Zhu, E. Conant, and B. Chance. Optical imaging as an adjunct to sonograph in differen-
tiating benign from malignant breast lesions. J. Biomed. Opt. 5, 229–236 (2000).
[299] Q. Zhu, T. Durduran, V. Ntziachristos, M. Holboke, and A. G. Yodh. Imager that combines
near-infrared diffusive light and ultrasound. Opt. Lett. 24, 1050–1052 (1999).
abdominal monitoring of fetal arterial blood oxygenation using pulse oximetry. J. Biomed.
226