Sunteți pe pagina 1din 5

Bonfring International Journal of Software Engineering and Soft Computing, Vol. 9, No.

2, April 2019 26

Detection of Polycystic Ovary Syndrome from


Ultrasound Images Using SIFT Descriptors
C. Gopalakrishnan and M. Iyapparaja

Abstract--- Polycystic Ovary Syndrome (PCOS) is a ultrasound images the error of PCOS is conceivable. Number
reproductive metabolic complexity described by the quantity of follicles isn't explicit for the polycystic ovaries; in some
of follicles present in the ovary. Ultrasound imaging of the cases the typical ovary likewise contains increasingly number
ovary contains basic data about the size, number of follicles of follicles [3]. To separate polycystic ovary from numerous
and its position. Continuously, the discovery of PCOS is a follicles of typical ovary, a standard criteria has been
troublesome undertaking for radiologists because of the advanced by European Society of Human Reproduction and
different sizes of follicles and is very associated with veins and Embryology, the American Society for Reproductive
tissues. In this paper, ultrasound images of ovary are analyzed Prescription depict, the polycystic ovaries contain at least
using various image processing techniques. For preprocessing
12 follicles with 2 – 9 mm in breadth. Amid the image
different standard methodologies are applied on ovary images securing stage, commotion is presented in the obtained image
to enhance the quality of the image. Canny edge detection in a great deal of conceivable way. The air hole between
method is used to detect the edges of the follicles from the
human body and the transducer test presents the commotion in
ultrasound image. Feature Extraction is an important phase in the image which will influence the determination [4]. At the
detection of objects. A robust feature descriptor identification point when contrast with CT and MRI modalities ultrasound
algorithm called Scale-Invariant Feature Transform is
image contains abundance of clamor especially Speckle Noise.
adopted in this work to detect the presence of the syndrome Dot clamor is multiplicative in nature which will make
from the image. Support Vector Machine is deployed for analysis troublesome, by debases edge and littlest
training and classification of the data. The proposed system
recognizable detail in a image. The goals and the nature of a
outperforms other state-of-art methods and proved its efficacy. image are exceedingly founded on the recurrence extend and
Keywords--- Canny, Follicles, Polycystic Ovary Syndrome, the transducers utilized in the test. These days at the season of
Scale Invariant Feature Transform, Support Vector Machines. procurement to improve the nature of an ultrasound images
higher end ultrasound machines are presented with amazing
transducers in the test, likewise the gel is utilized to decrease
I. INTRODUCTION
the air prattle between the test and the human body. After

P OLYCYSTIC ovary syndrome is a complex predominant


endocrine issue which genuinely impacts ladies'
wellbeing. The confusion is portrayed by the development of
procurement, to improve the division result and for exact
finding, ultrasound ovary images are sifted utilizing different
image de-noising strategies. To precisely find the quantity of
numerous follicular pimples in the ovary. PCOS discourage follicles in the ultrasound ovary image, it is important to
the equalization of follicle invigorating hormone and decrease the spot clamor. A large portion of the ground-
luteinizing hormone which requirements for an egg to develop breaking separating strategies accessible today contain various
appropriately inside the ovaries [1]. The collection of not control parameters which must be changed in accordance with
completely created follicles prompts PCOS which have the get ideal execution. The radiologist utilizing regular
basic side effects of fruitlessness, unpredictable menstrual ultrasound machine to examine the ovary and they are turns
periods or no menstrual period, undesired hair development on the test dynamically to filter ovaries albeit first perceive every
the face and body, weight gain, stoutness, skin inflammation follicle and afterward estimating its size and measurement.
and hyperinsulinemia [2]. Restorative imaging, for example, This framework is manual, tedious additionally excruciating to
CT, MRI and Ultrasound methods are familiar with review the the patient. To overcome those issues, a robust computational
human body so as to finding of sicknesses and its treatment. framework is proposed for the detection of follicles.
Ultrasound is for the most part utilized for imaging liver,
kidney, uterus in light of its noninvasive nature additionally it II. LITERATURE REVIEW
is versatile, transferable, versatile, have astounding worldly
In literature, several contributions on noise reduction using
goals, and they are moderately modest. These should prompt a
novel methods based on wavelet thresholding for the
ultrasound evaluation of ovaries. These assessments give
suppression of speckle noise were introduced. Weighted
enormous understanding into the state of the ovary and
variance is used for estimating the threshold value. In
number of follicle exhibits in it. Because of low quality of
automatic segmentation, many filters disturb the sharpness of
the regions, but adaptive wavelet threshold does not degrade
C. Gopalakrishnan, Research Scholar, SITE School, VIT, Vellore. the sharpness of the regions in an image [5]. This is because of
E-mail: arungopalit@gmail.com the fact that wavelet transform produces better results when
M. Iyapparaja, Associate Professor, SITE School, VIT, Vellore.
E-mail: Iyapparaja.m@vit.ac.in compared to the median and Wiener filter [6].
DOI:10.9756/BIJSESC.9017

ISSN 2277-5099 | © 2019 Bonfring


Bonfring International Journal of Software Engineering and Soft Computing, Vol. 9, No. 2, April 2019 27

A modified morphological image cleaning algorithm is contour models are proposed by Caselles et al these methods
introduced which depends on the mathematical morphology based on speed term [16].
operation. It is based on set theory. The disadvantage in this
Mohammed Ammar et al. introduced thresholding to
algorithm is that the morphological filter loses its features
initialize the level set segmentation. The binary mask is
corresponding to the size of the structuring element [7].
obtained from thresholding [17].
Michailovich et al. introduces a despeckling method for
ultrasound images, these method without disturbing the
III. MATERIALS AND METHODS
medical image anatomical content remove the noise from the
image [8]. a. Dataset Description
Due to the insufficiency of hormones the egg does not
mature properly and results in failure of ovum release. In The dataset of ovarian ultrasound images are collected
medical diagnosis, identifying follicles from ultrasound image after consulted with professional radiologist and gynecologist
is a critical issue. from Nandhini Sri Diagnostic Center.

Modified watershed algorithms are used to extract follicles The image acquisition is done by the GE LOGIQ
from ovary which results in over-segmentation. The output of ultrasound system with the transvaginal transducer frequency
the watershed algorithm is given as an input to the clustering of 24 Hz. In total, 65 ultrasound ovarian images are used to
algorithm produces better recognition rate [9]. Edge-based conduct this experiment.
segmentation method results in broken edges. Morphological
operations are used to overcome these problems. Shrinkage of b. Processing of Ultrasound Images
images may lose vital details from an image. For medical
imaging, accurate results are needed for effective diagnosis • RGB-Gray Scale Conversion
[10]. Transformation of a shading image into a grayscale image
To efficiently segment follicles region growing methods comprehensive of notable highlights is a confused procedure.
were introduced. These methods initially require a seed point The changed over grayscale image may lose contrasts,
which is manually selected by a physician which also comes sharpness, shadow, and structure of the shading image. To
under manual detection of follicles. safeguard contrasts, sharpness, shadow, and structure of the
shading image another calculation has proposed.
Region growing methods are less sensitive to noise and
increase the recognition rate [11]. Morphological operators To change over the shading image into grayscale image the
often take a binary image. Morphological opening and closing new calculation performs RGB guess, decrease, and expansion
operations were used to extract the bright features from the of chrominance and luminance.
ultrasound image background by using Top hat transform. The
enhanced image is given as input to the Scan line thresholding • Image Thresholding
for segmentation.
Image thresholding is a straightforward, yet powerful,
When compared to manual detection, the error rate is less
method for dividing a image into a forefront and foundation.
in this method [12]. Histogram equalization is applied to
This image investigation procedure is a sort of image division
enhance the contrast of the de-speckled image. Further, the
that confines by changing over grayscale images into double
negative transformation is applied to the histogram equalized
colored (0,1) pixel imputed images. Image thresholding works
image as the proposed segmentation method works on high
well with large amounts of complex images.
intensity valued objects. In the segmentation stage, the active
contour without edges is used. The result image after applying • Speckle Noise Elimination
active contour method contains segmented regions with it
[13]. Speckle noise in ultrasound images are undermined with
Thresholding is the easiest and effective methods to dot clamor characteristically, which can cause negative
achieve optimal threshold value to binarize the grayscale impacts on the based translation and indicative methods.
image. Otsu method is a global threshold selection method Detection of speckle noise is an imperative advance
widely used because of its simplicity and quality. The proceeding in the preparation and investigation of the
problems related to energy minimization approach were restorative ultrasound images.
solved by new active contour model following the level set
The mathematical expression of multiplicative noise model
framework. During the curve evolution the changes in the
is given by
contour solved implicitly this is the main advantage of the
level set method [14].
[a, b] k[a, b] + η[a, b]k[a, b]
k=
In medical imaging to segment region of interest level set
methods are used. Chan and Vese introduces active contours Let k [a,b] be the original ultrasound ovary image. k [a, b]
method also consider information inside the region they solve be the digitized version of noise and η[a,b] is the noise
this problem using level set methods [15], the geometric active function.

ISSN 2277-5099 | © 2019 Bonfring


Bonfring International Journal of Software Engineering and Soft Computing, Vol. 9, No. 2, April 2019 28

Fig. 1: Workflow of Proposed Method


• Edge Detection the summed up Hough change. Each bunch of at least 3
includes that concur on an item and its posture is then subject
Segmentation of an image is characterized as the way to additionally point by point show confirmation and
toward apportioning it into important sub-parts that show accordingly anomalies are disposed of. At last the likelihood
wanted highlights. Fragmenting follicles from ovary image is that a specific arrangement of highlights demonstrates the
a troublesome errand because of the nearness of delicate nearness of an item is registered, given the exactness of fit and
tissue, common tissue and veins. Thresholding is a optimal number of plausible false matches. Item coordinates that
procedure to separating objects from foundation which breeze through every one of these tests can be recognized as
depends on the rule of minimization work delivers an ideal right with high confidence.
edge esteem. The decision of limit must be made exact by
thinking about the idea of their image and the quantity of Table 1: Segmentation Results of Image with different
articles anticipated from the image. The Binarization Methods
technique converts the image into two color pallets dependent Proposed
on the rule of minimization of inside class change. Original Image Otsu
Method
The processes involved in canny edge detection method are:
• Pre-processing
• Gradient Calculation
• Non-maximum suppression
• Thresholding No of Objects Identified: 8
Identified:4
Present: 7
c. Feature Descriptor Extraction Using SIFT
d. Supervised Classification
Scale-Invariant Feature Transform (SIFT) is a component
discovery calculation in computer vision to distinguish and Support Vector Machine is a mathematical model, follows
portray neighborhood includes in images. SIFT key points of parametric estimation, works based on the kernel function and
items are first extricated from a lot of reference images and hyperplane otherwise called a decision boundary. The data
put away in a database. An item is perceived in another image points are projected in 2D vector space. Based on the kernel
by separately looking at each component from the new image function, the decision boundary is identified that splits the data
to this database and discovering hopeful coordinating into different groups depends upon the classes it belongs to.
highlights dependent on Euclidean separation of their element With the help of cost function, (RMSE in this case), the error
vectors. From the full arrangement of matches, subsets of key rate of the learning model is adjusted to an optimal point
points that concur on the item and its area, scale, and iteratively. In this work, the feature points extracted from the
introduction in the new image are distinguished to sift through images are considered as features to train the model.
great matches. The assurance of predictable groups is
performed quickly by utilizing a proficient hash table usage of

ISSN 2277-5099 | © 2019 Bonfring


Bonfring International Journal of Software Engineering and Soft Computing, Vol. 9, No. 2, April 2019 29

e. Performance Evaluation Table 2: Performance Evaluation of different ML Algorithms


The performance of the model is calculated using few with Various Metrics
standard metrics such as Mean Squared Error, Accuracy and Algorithms/Performance Accuracy MSE NAE
Normalized Absolute Error. K-Fold cross-validation method is Support Vector Machines 94.40 0.15 0.28
adopted to evaluate the model performance with 10 folds each. Naïve Bayes 88.25 0.36 0.53
a. MSE measures the total amount of difference between Decision Tree 87.54 0.42 0.69
the original image and filtered image [18].
R C V. CONCLUSION
∑∑
1
=MSE (G ( s, k ) − G ' ( s, k ) 2 This work exhibits a computationally effective strategy for
RC =s 1=
k 1 follicle identification and division of ovarian follicles from the
b. The normalized absolute error of filtered image is ultrasound ovary image. In this paper, canny edge detection
defined as method is applied as an underlying cover for the dimension set
R C R C definition to detect the edges of the follicles. Also, this
=NAE ∑∑
=s 1 =
G ( s, k ) − G ' ( s, k ) ∑∑ G(s, k )
=s 1 =
technique identifies the definite number of follicles from the
k 1 k 1 ultrasound ovary image. SIFT algorithm detects the feature
descriptors of follicles to identify the presence of syndrome.
IV. RESULTS AND DISCUSSION For training and classification of the data, SVM is adopted.
In this experiment, various segmentation techniques are This proposed strategy performed well than the other
applied on ultrasound image to identify the location of the benchmarking algorithms. Thus these proposed techniques are
follicles. Using SIFT algorithm, the feature descriptors are more appropriate for extraction and detection of follicles from
detected. The identified vectors are fed into robust Machine the ovary images. In future, more robust algorithms are
Learning algorithms namely Support Vector Machine (SVM), deployed in this framework to improve the predictability of
Naïve Bayes (NB) and Decision Tree (DT). Among them, the system for different kind of images.
SVM outperformed other two algorithms based on the results
obtained from various performance metrics. REFERENCES
[1] A. Pellicer, P. Gaitán, F. Neuspiller, G. Ardiles, C. Albert, J. Remohí
In Fig 2 and 3, the performance of MSE and NAE are and C. Simón, “Ovarian Follicular Dynamics: From Basic Science to
depicted using a graph. Table 3 represents the results of three Clinical Practice”, Journal of Reproductive Immunology, Vol. 39, No. 1-
ML Algorithms obtained from this experiment. 2, Pp. 29-61, 1998.
[2] Affiliates of Medifocus.com, “Medifocus Guidebook: Polycystic Ovary
Syndrome”, Medifocus.com, Inc, 2007.
[3] T.W. Kelsey and Wallace W.H.B. “Ovarian Volume Correlates Strongly
with Number of Non Growing Follicles in the Human Ovary”,
Obstetrics and Gynecology International, Vol. 2012, Pp. 1-5, 2012.
[4] C. Battaglia, P.G. Artini, A.D. Genazzani, R. Gremigni, M.R. Slavatori
and M.R. Sgherzi, “Color Doppler Analysis in Oligo and Amenorrheic
Women with Polycystic Ovary Syndrome”, Gynecological
Endocrinology, Vol. 11, No. 2, Pp. 105-110, 1997.
[5] S. Sudha, G.R. Suresh and R. Sukanesh, “Speckle Noise Reduction in
Ultrasound Images by Wavelet Thresholding based on Weighted
Variance”, International Journal of Computer Theory and Engineering,
Vol. 1, No. 1, Pp. 7-12, 2009.
[6] M.C. Nicolae, “Comparative Approach for Speckle Reduction in
Medical Ultrasound Images”, Romanian Journal of Bio-Physics, Vol.
20, No. 1, Pp. 13-21, 2010.
Fig. 2: Performance Graph for Mean Square Error [7] R.T. Jeyalakshmi and K. Ramar, “A Modified method for Speckle Noise
Removal in Ultrasound Medical Images”, International Journal of
Computer and Electrical Engineering, Vol. 2, No. 1, Pp. 54-58, 2010.
[8] O.V. Michailovich and A. Tannenbaum, “Despeckling of Medical
Ultrasound Images”, IEEE transactions on ultrasonics, ferroelectrics,
and frequency control, Vol. 53, no. 1, Pp. 64-78, 2006.
[9] Y. Deng, Y. Wang and P. Chen, “Automated Detection of Polycystic
Ovary Syndrome from Ultrasound Image”, 30th Annual International
IEEE Engineering in Medicine and Biology Society Conference
Vancouver, British Columbia, Canada, Pp. 20-24, 2008.
[10] P.S. Hiremath and Jyothi R. Tegnoor, “Automatic Detection of Follicles
in Ultrasound Images of Ovaries using Edge Based Method”,
International Journal of Computer Applications Special Issue on Recent
Trends in Image Processing and Pattern Recognition Pp. 15-16, 2010.
[11] Y. Deng, Y. Wang and Y. Shen, “An Automated Diagnostic System of
Polycystic Ovary Syndrome based on Object Growing”, Journal of
Artificial Intelligence in Medicine, Elsevier Science Publishers Ltd.
Essex, UK, Vol. 51, No. 3, Pp. 199-209, 2011.
[12] P. Mehrotra and C. Chakraborty, “Automated Ovarian Follicle
Fig. 3: Performance Graph for Normalized Absolute Error Recognition for Polycystic Ovary Syndrome”, International Conference
on Image Information Processing, Pp. 1-4, 2011.

ISSN 2277-5099 | © 2019 Bonfring


Bonfring International Journal of Software Engineering and Soft Computing, Vol. 9, No. 2, April 2019 30

[13] P.S. Hiremath and R. Jyothi Tegnoor, “Automatic Detection of Follicles


in Ultrasound Images of Ovaries using Active Contours Method,”
International Journal of Service Computing And Computational
Intelligence Vol. 1, No. 1, Pp. 26-30, 2011.
[14] C. Phillips, “The Level-Set Method,” MIT Undergraduate Journal of
Mathematics, 1999.
[15] T. Chan and L. Vese, “Active contours without edges”, IEEE
Transactions Image Processing, Vol. 10, No. 2, Pp. 266-277, 2001.
[16] V. Caselles, R. Kimmel and G. Sapiro, “On geodesic active contours”,
International Journal of Computer Vision, Vol. 22, No. 1, Pp. 61-79,
1997.
[17] M. Iyapparaja and M. Tiwari, “Security policy speculation of user
uploaded images on content sharing sites”, IOP Conf. Series: Materials
Science and Engineering, Vol. 263, Pp. 1-8, 2017.
[18] M. Iyapparaja, P. Sivakumar, “Metrics Based Evaluation for Disease
Affection in Distinct Cities”, Research J. Pharm. and Tech., Vol. 10, No.
8, Pp. 2487-2491, 2017.

ISSN 2277-5099 | © 2019 Bonfring