METHOD FOR RECREATING RESPIRATORY WAVEFORMS AND SIMULTANEOUSLY ESTIMATING RESPIRATORY AND HEART RATE FROM POTENTIAL PHOTOVOLUMETRIC SIGNALS (PPG)

Abstract

The invention proposes a method for reconstructing respiratory waveforms, simultaneously estimating respiratory rate and heart rate from a single-channel photovoltaic (PPG) signal. The method performs preprocessing to remove noise and extract three derived feature channels: baseline variation (BW) reflecting the slow change of baseline due to the respiratory cycle, amplitude modulation (AM) reflecting peak-trough amplitude changes, and frequency modulation (FM) reflecting the variation in distance between successive peaks. The three channels are fed into a multi-tasking neural network model with three independent encoders and a shared processor integrating information via a late combining mechanism. The output block uses parallel decoding branches to simultaneously obtain the heart rate, respiratory rate, and reconstructed respiratory waveform values. The invention supports the detection of respiratory abnormalities such as sleep apnea, for applications in wearable medical devices and remote health monitoring.

Images