METHODS AND SYSTEMS FOR GENERATING CONDITIONED VOLUME-TO-HEART RATE OPTICAL (PPG) SIGNALS USING CONDITIONED VARIABLE AUTCODING NETWORKS (CVAE)

Abstract

The invention proposes a method and system for generating conditioned photovoltaic (PPG) signals based on heart rate using a Conditional Variational Autocoding Network (CVAE). The method comprises the following steps: (i) Preprocessing the actual signal using a bandpass filter within the physiological heart rate range and normalizing the data; (ii) Conditional coding by simultaneously embedding the signal and the corresponding heart rate label to determine the latent probability distribution; (iii) Sampling the latent code using re-parameterization; (iv) Conditional decoding of the latent code associated with the target heart rate to reconstruct the synthesized PPG signal; and (v) Training the model using KL Annealing. The invention enables the generation of large-scale and diverse PPG datasets across all heart rate ranges, providing a sufficiently large data source for training AI models while reducing risks associated with bioethics and personal privacy, provided there is a suitable validation procedure.

Images