Abstract
Abstract
Photoplethysmography (PPG) is a cost-effective optical technology that detects changes in blood volume within tissues, providing insights into the body’s physiological dynamics over time. By analyzing PPG data as a time series, valuable information about cardiovascular health and other physiological parameters such as heart rate variability, peripheral oxygen saturation, and sleep stages can be estimated from commercial wearable devices. However, these indicators require reliable signals, which are often contaminated with movement noise, variable ambient lighting conditions, and other environmental effects. Hence, signal quality assessment is crucial in determining the trustworthiness of such data for wearable health applications. Modern approaches for PPG signal quality assessment are either based in the application of signal processing logical conditions or adoption of Convolutional Neural Networks. In this work, we propose the addition of differential and integral attention mechanisms composite with a two-stage rule and Convolutional-based procedure for the classification of segments. The approach is implemented in a way to balance storage size and classifier accuracy in the resulting models, increasing robustness across signals from different commercially available devices. Our proposed two-stage differential attention methods achieve accuracy values between 0.8952 and 0.9737, and F1-scores between 0.9183 and 0.9865 across four expert-annotated datasets.