M.

TECH PROJECT MID-TERM REPORT

ROLL NO: 10EE35025
NAME: Satya Narayan Shukla
SUPERVISOR Prof. Amit Patra
1. Project Title: Noninvasive Cuffless Blood Pressure Estimation from PPG signal.

2. Project Field: Biomedical Engineering and Signal Processing.

3. Project Type: Development.

4. Project Aim/Objective/Importance/Summary: Blood pressure (BP) is one of the most
important physiological parameters that can provide valuable information for personal
health care. There are two ways to measure human blood pressure viz. invasive and
noninvasive methods. Though invasive methods have been known to measure BP
continuously and accurately but they are extremely inconvenient to use and might cause
infection. The widely used cuff-based noninvasive methods are also not very convenient
to use especially for injured or obese people and infants. Thus, there has been an unmet
need for a convenient noninvasive cuffless BP measurement technique. While cuffless
methods to measure BP have previously been studied, but most of them were limited to
offline processing of captured signals. This project basically aims at developing a device
to estimate the human blood pressure from photoplethysmographic (PPG) signal.

5. Keywords: PPG Signal, Blood Pressure, Cuffless, Noninvasive.



Supervisors Signature
10EE35025 | Satya Narayan Shukla

2 | P a g e

PROJECT WORK IN 7
th
AND 8
th
SEMESTER
Title: Noninvasive Cuffless Blood Pressure Measurement from Vascular Transit Time (VTT).
Work Done: Developed a noninvasive cuffless BP measurement system that is based on Vascular
Transit Time (VTT) extracted from heart activity signals namely PPG (Photoplethysmograph) and
PCG (Phonocardiogram) signals. The PPG and PCG signals, which are captured simultaneously,
are processed on an LPC1768 Processor Kit. Then the Vascular Transit Time (VTT), which is
defined as the time taken by the blood to travel from the heart to an extremity of the body for
one stroke of the heart, is extracted and transmitted via Bluetooth to an android smartphone.
The user interactive android application, after reading users personal information such as age,
weight and height, estimates and displays the Blood Pressure and Heart Rate (HR) of the user.[1]
The raw signals obtained from the sensors are first amplified and filtered through anti-aliasing
filters and then converted into digital form using ADC of the LPC1768 processor kit. Both the
signals are further processed with appropriate digital filtering for the purpose of correct
identification of peaks of two signals. A peak detection algorithm [2] was applied on both the
waveforms. In the algorithm, a peak function is used for each sample in the waveforms that
computes the average of (i) the maximum among the signed distances of each sample from its
k left neighbors and (ii) the maximum among the signed distances from its k right neighbors. We
have chosen k = 10 for our system. Median Filtering and some more processing was done in
order to remove the false peaks.
Results: Accuracy of Blood Pressure measured using our method and a DBPM is shown in table.
Subject
Systolic Blood Pressure
(mmHg)
Diastolic Blood Pressure
(mmHg)
Heart Rate (bpm)

Accuracy %

Accuracy %
HR Accuracy
%
Measured Actual Measured Actual Measured Actual
1. 120 122 91 85 81 75 98.4 92.9 92.0
2. 115 116 75 79 61 61 99.2 94.9 100
3. 118 115 85 85 93 87 97.4 100 93.1
4. 107 118 62 63 62 65 91.7 98.4 95.4
5. 123 123 70 76 91 86 100 92.1 94.2

Conclusions: The experiment evaluations with results of more than 90% accuracy clearly
indicate that this technique is feasible, convenient, and portable and can be very well used for
normal and non-pathological purposes especially for people with expected fluctuations in BP.
10EE35025 | Satya Narayan Shukla

3 | P a g e

PROJECT WORK IN 9
th
SEMESTER
Title: Noninvasive Cuffless Blood Pressure Estimation from PPG signal.
Literature Review: It is widely accepted that PTT (Pulse Transit Time) varies inversely with blood
pressure changes [3] and can be used for cuffless and continuous estimation of blood pressure.
Usually, time interval between ECG R-wave and the peak of peripheral pulse is selected as the
PTT. In last semester, blood pressure was calculated using the same method but PCG
(Phonocardiogram) signal was used in place of ECG. But this time round, only PPG signal is being
used to estimate the BP.
It has been found that some features of PPG signal (systolic upstroke time and diastolic time)
can be used to estimate blood pressure [4,5] more accurately as compared to PTT (or VTT)
Method [6]. There is a relation, not always linear, between the blood pressure and the pulse
duration, obtained from PPG signal. In order to estimate the blood pressure from the PPG signal,
in this project the Artificial Neural Networks (ANNs) are being used to increase the accuracy of
the estimated blood pressure by implementing a complex function rather than linear regression
analysis. More the number of features used, more is the accuracy of BP and more is the
complexity of ANN [7]. Accuracy can be further improved by preprocessing the input PPG signal
using Kalman Filtering [8] or Continuous Wavelet Transform [4]. It has also been found that for
different age groups of people, BP is related to different features of PPG signal [9]. There are
some more features in PPG signal apart from PTT or the features used in [4], [5], [7], [8] or [9]
which are as better as them for BP estimation [10].
Work Done: To have a wide representation of PPG signals and correspondent beat-to-beat BP
values the distributed freely MIMIC database is being used. It includes multiple recordings of
physiological signals and vital signs captured from monitors for tens of thousands of intensive
care unit patients. Most of them include ECG, BP, PPG, respiratory signals which were recorded
simultaneously with a sampling rate of 125Hz. The features have to be extracted from PPG signal
and used as the input vector of ANN. In total, more than 15000 separate PPG heartbeats with
corresponding BP values for different persons and different time instances are identified. 70%
of them will be used for training the ANN, 15% for validation step and 15% for testing. Initially,
ANN is trained with only two features systolic upstroke time and diastolic time. Objective is to
get an accuracy of more than 95%, so eventually more features will be added to the input of
ANN till desired accuracy is achieved.

10EE35025 | Satya Narayan Shukla

4 | P a g e

WORK PLAN FOR 9
th
AND 10
th
SEMESTER

Work Plan for 9
th
semester:
1. Extract PPG signal data for a wide range of people in order to generalize the formula.
2. Extract different features from PPG signal.
3. Training, validation and testing of the data on Artificial Neural Network (ANN).
4. Increase the number of features (size of input vector of ANN) to increase the accuracy
and at the same time try to optimize between accuracy and complexity of ANN.
5. Try to divide the data into different age groups and observe the dependency of BP among
different age groups on different features of PPG Signal.
6. Finally, get a prominent set of features for a desired accuracy.


Work Plan for 10
th
semester:
1. Implement the ANN on hardware.
2. Take the values of weights calculated by Neural Network from MATLAB and employ it in
C code to implement the whole system on hardware.
3. Preprocess the input PPG signal using Kalman Filter or Continuous Wavelet Transform if
required.
4. Extract the features from practical PPG signal and use it as input to the system.
5. Estimate the blood pressure using the same formula that is used in ANN to get the desired
accuracy.

10EE35025 | Satya Narayan Shukla

5 | P a g e

REFERENCES
[1] Vikram Chandrasekaran et al. Cuffless Differential Blood Pressure Estimation
Using Smart Phones in IEEE Transactions on Biomedical Engineering, vol. 60 no.
4,pg. 1080, April 2013.
[2] G. K. Palshikar, Simple algorithms for peak detection in time-series, in Proc. 1st
Int. Conf. Adv. Data Anal., 2009.
[3] M. H. Pollak and P. A. Obrist, Aortic-radial pulse transit time and ECG Q-wave to
radial pulse wave interval as indices of beat- by-beat blood pressure change,
Psychophysiology, vol. 20, pp. 21-28, 1983.
[4] X. F. Teng and Y. T. Zhang, Continuous and noninvasive estimation of arterial
blood pressure using a photoplethysmographic approach, Proc. of 25th Annual
Inter. Conf. of the IEEE Engineering in Medicine and Biology Society, Cancun,
Mexico, 2003, pp. 31533156.
[5] Y. Yoon, G. Yoon, Nonconstrained blood pressure measurement by
photoplethysmography, Journal of the Optical Society of Korea, vol.10, no.2, pp.
91-95, June 2006.
[6] Giancarlo Fortino, Valerio Giamp, PPG-based Methods for Non Invasive and
Continuous Blood Pressure Measurement: an Overview and Development Issues
in Body Sensor Networks in Proc. IEEE International Workshop on Medical
Measurements and Applications, 2010.
[7] Yu. Kurylyak, F. Lamonaca, D. Grimaldi, "A neural network-based method for
continuous blood pressure estimation from a PPG signal," in Proc. IEEE
International Instrumentation and Measurement Technology Conf., Minneapolis
(MN), 2013.
[8] Yu. Kurylyak et al., Photoplethysmogram-based Blood Pressure Evaluation using
Kalman Filtering and Neural Networks, in Proc. IEEE International Symposium on
Medical Measurements and Applications, 2013.
[9] Rohan Samria et al., Noninvasive Cuffless Estimation of Blood Pressure using
Photoplethysmography without Electrocardiograph Measurement, in IEEE
TENSYMP 2014.
[10] Yanjun Li et al., Characters available in photoplethysmogram for blood pressure
estimation: beyond the pulse transit time, in Australasian Phys Eng Sci Med
2014 , 367-76.