Photoplethysmography Signal Quality Assessment using Attentive-CNN Models

Abstract

Due to the rapid popularization of wearable computers such as smartwatches, Health Monitoring Applications (HMA) are becoming increasingly popular because of their capability to track different health indicators, including sleep patterns, heart rate, and activity tracking movements. These applications usually employ Photoplethysmography (PPG) sensors to monitor various aspects of an individual’s health and well-being. PPG is a non-invasive and cost-effective optical technique based on the detection of blood volume changes in the microvascular bed of tissue, capturing the dynamic physiological changes in the body with continuous measurements taken over time. Analyzing PPG as a time series enables the extraction of meaningful information about cardiovascular health and other physiological parameters, such as Heart Rate Variability (HRV), Peripheral Oxygen Saturation (SpO2), and sleep status. To enable reliable health indicators, it is important to have robustly sampled PPG signals. However, in practice, the PPG signal is often corrupted with different types of noise and artifacts due to motion, especially in scenarios where wearables are used. Therefore, Signal Quality Assessment (SQA) plays a fundamental role in determining the reliability of a given PPG for use in HMA. Considering this, in this work, we propose a novel PPG SQA method focused on the balance between storage size and classifier quality, aiming to achieve a lightweight and robust model. This model is developed using recent advances in attention-based strategies to significantly improve the performance of purely Convolutional Neural Network (CNN)-based SQA classifiers.