Abstract:

Abstract Finger pulse signal, popularly known as Photoplethysmogram (PPG), can provide important information on circulatory functions in the human body. PPG signal is prone to motion artifact (MA) due to peripheral body movement. In this paper, a new method of motion artifact reduction is proposed. Here, the accelerometer signal is utilized for its frequency domain analysis for detecting the presence of MA in the PPG signal. From the frequency peak analysis of both the PPG signal and the accelerometer signal, the heart rate (HR) is estimated. Once the HRs are calculated in the 8-sec moving time window, the calculated time series-based HRs are sent to the HR updation unit, where the HRs are analyzed and modified using the LSTM algorithm to further reduce the effect of motion artifacts. The result showed a significant improvement in the Average Absolute Error (AAE) calculated with respect to the ground truth HR given. The mean AAE was 2.05, whereas the popular literature demonstrated an AAE of 2.42(TROIKA) and 1.285(JOSS). Although the algorithm didn’t give the best result among the literature, the fact that it didn’t require any reference clean signal for its functioning and the presence of the LSTM algorithm also makes the algorithm adaptive person to person, case to case, making this work significant. Since the algorithm is not trained once but is constantly getting trained on the consecutive input HRs, it will be very adaptive and can provide good results for critical care unit patients, whose cardiac vitalities vary to a larger extent.

Keywords: