Photo-Plethysmography
Noise Cleaning
Algorithm
WVU Biomedical Engineering Design Project 2018-2019
Objective
​
Develop an algorithm to achieve heart rate (HR) monitoring as reliable as ECG to allow for optimized PPG implementation in human performance and patient monitoring.
Background
With the growing market in athletics and at home healthcare monitoring, the race is on to optimize photoplethysmography (PPG) heart rate signals obtained from wearable devices to be as accurate as electrocardiography (ECG) signals taken directly from the chest. PPG devices have the benefit of being comfortable, easy to use, and commercially available, but have inherent complications with regards to motion artifacts and ambient lighting. Most PPG commercial devices incorporate filtering algorithms that are proprietary and ineffective in nature, i.e., do not output raw heart rate data that is as accurate as ECG. Noise from motion must be reduced using signal processing techniques to increase the overall accuracy and effectiveness of a PPG device.
General Overview
Data Collection
-
9 sets of physiological data were collected for two different activities at the Rockefeller Neuroscience Institute.
-
5 Sitting​
-
4 Walking
-
-
25 sets of synthetic data were manipulated from a preexisting PPG data base (Kings College London):​
-
5 Amplitude Modulation​
-
5 Frequency Modulation
-
5 Baseline Wander
-
5 Combined
-
5 Gaussian Static Noise
-
-
9 sets of physiological data were collected for two different activities at the Rockefeller Neuroscience Institute.
-
5 Sitting​
-
4 Walking
-
-
25 sets of synthetic data were manipulated from a preexisting PPG data base (Kings College London):​
-
5 Amplitude Modulation​
-
5 Frequency Modulation
-
5 Baseline Wander
-
5 Combined
-
5 Gaussian Static Noise
-
-
9 sets of physiological data were collected for two different activities at the Rockefeller Neuroscience Institute.
-
5 Sitting​
-
4 Walking
-
-
25 sets of synthetic data were manipulated from a preexisting PPG data base (Kings College London):​
-
5 Amplitude Modulation​
-
5 Frequency Modulation
-
5 Baseline Wander
-
5 Combined
-
5 Gaussian Static Noise
-
Our Algorithm
-
Deconstruction:
-
The original PPG signal is decomposed into correlation and detail coefficients, low and high frequency components, respectively.
-
-
Filtering:
-
Selected wavelet(s) are filtered using a lowpass filter to remove baseline wander disruption ​
-
-
Peak Counting
-
​The peaks of the reconstructed PPG signal are identified in this step and then utalized for heart rate and other calculations
-
-
Deconstruction:
-
The original PPG signal is decomposed into correlation and detail coefficients, low and high frequency components, respectively.
-
-
Filtering:
-
Selected wavelet(s) are filtered using a lowpass filter to remove baseline wander disruption ​
-
-
Peak Counting
-
​The peaks of the reconstructed PPG signal are identified in this step and then utalized for heart rate and other calculations
-
-
Deconstruction:
-
The original PPG signal is decomposed into correlation and detail coefficients, low and high frequency components, respectively.
-
-
Filtering:
-
Selected wavelet(s) are filtered using a lowpass filter to remove baseline wander disruption ​
-
-
Peak Counting
-
​The peaks of the reconstructed PPG signal are identified in this step and then utalized for heart rate and other calculations
-
Outputs
-
As seen above, the post-processed PPG signal bears a characteristic heart rate waveform.
-
MATLAB outputs the post-processed signal and average PPG heart rate.
-
As seen above, the post-processed PPG signal bears a characteristic heart rate waveform.
-
MATLAB outputs the post-processed signal and average PPG heart rate.
-
As seen above, the post-processed PPG signal bears a characteristic heart rate waveform.
-
MATLAB outputs the post-processed signal and average PPG heart rate.
Meet the Team
