Improving heart rate variability information consistency in Doppler cardiogram using signal reconstruction system with deep learning for Contact-free heartbeat monitoring

Abstract

A contact-free continuous heart rate variability (HRV) analysis is required to conduct daily heart monitoring and minimize physical contact during medical remedies owing to COVID-19. This paper suggests a Doppler cardiogram (DCG) signal processing and reconstruction system that enables the standard deviation of normal-to-normal peaks (SDNN) obtained from DCG to be used as an actual HRV index. The heartbeat signals of twelve healthy adults were recorded. Three electrodes and a Doppler radar module were used to record the electrocardiogram (ECG) and DCG signals, respectively. To optimize the performance of the signal reconstruction system, two signal processing methods were applied to the dataset. These DCG signals were reconstructed into a signal that mimicked the ECG using a variational autoencoder (VAE), to enhance the consistency of the SDNN. The synthetic signal quality was assessed by comparing the SDNN of the synthetic ECG with that of the reference ECG. A total of 1,430 signals were reconstructed to achieve a valid SDNN. A unified analysis of the signal reconstruction results using different signal processing methods was built up to raise the consistency growth. The final result of the signal reconstruction system represented a consistency improvement of 75.5%, compared to the SDNN of the input DCG.