RespTrack-Net: Respiration Parameters Tracking From PPG Signal Using Deep Learning Model

Abstract

Photoplethysmography (PPG) signals are widely used for nonintrusive health monitoring, but existing methods often struggle with noise susceptibility and computational complexity, limiting their practical utility. This research introduces two key innovations: the wearable low-cost PPG acquisition device (WeLOVE) and the RespTrack-Net model. The WeLOVE device is designed to provide high-quality PPG signal acquisition at low cost, addressing the accessibility challenges of current systems. The RespTrack-Net model introduces a novel architecture tailored for extracting respiration rate (RR) and cardiovascular parameters with enhanced robustness to noise and motion artifacts. The proposed approach was validated using two datasets: an experimental database (eight subjects) collected in this study and the publicly available CapnoBase database (42 subjects). RespTrack-Net achieved mean absolute errors of 1.58 $\pm$ 1.30 and 3.16 $\pm$ 3.36 for RR estimation on these datasets, respectively, outperforming State-of-the-Art methods. These contributions demonstrate the system's novelty and potential for reliable, real-time health monitoring in diverse settings. Future research will explore the use of the proposed device for sleep apnea detection, offering a cost-effective and comfortable alternative to current polysomnography (PSG) methods.