Photoplethysmograph-based time-frequency and machine learning applications on biomedical signal analysis for medical diagnosis

Abstract

Machine learning (ML) integration in biomedical signal processing and medical diagnosis has the potential to revolutionize healthcare by improving diagnostic accuracy. This paper focuses on the applications of different ML algorithms for analyzing real-time physiological data collected from Photoplethysmography (PPG) sensors. Heart rate variability (HRV) analysis using electrocardiography (ECG) signals makes the process longer and bulky. Therefore, this paper demonstrates the real-time generation of HRV signals using a simple, low-cost, and non-invasive PPG sensor which is further processed using the Arduino ATMEGA328P microcontroller and then interfaced to a PC for display to investigate the usefulness of HRV feature analysis. HRV features have been computed using time domain analysis (TA), and frequency domain analysis (FA). At last, these TA and FA indices have been given to different ML models that could predict the gender, age group, and physiological conditions of a human being. Prediction of the physiological conditions using TA, FA, and ML models simultaneously makes the proposed approach more novel than the other existing methods. Comparative analysis of different ML approaches using ROC curves and confusion matrices has been shown to find the effectiveness and precision of different proposed models. It shows random forest ML approach has achieved 91% accuracy in identifying the physiological conditions. This simple yet accurate real-time PPG-based time-frequency ML system might be useful in medical assessment with faster response.