Deep Neural Networks for Feature Extraction in Blood Pressure Prediction Using Radar Signals

Non-contact blood pressure (BP) monitoring using radar signals is of significant importance for remote health monitoring and patient care, enabling continuous and unobtrusive measurements. In this study, we investigate the effectiveness of various deep neural network architectures for BP prediction. Specifically, we conduct four key analyses: (1) We compare the performance of shallow and deep convolutional neural networks (CNNs), revealing that deeper architectures outperform their shallower counterparts; (2) We assess the impact of residual connections in deep CNNs, demonstrating that these connections enhance prediction accuracy and computational efficiency; (3) We evaluate the utility of directly extracting temporal information using Gated Recurrent Units (GRUs) and Long Short-Term Memory (LSTM) networks, noting that reliance on these methods alone is insufficient for optimal performance; and (4) We explore a spatiotemporal integration approach that combines waveform feature extraction with temporal information processing, yielding the best results. These findings underscore the potential of deep learning techniques to improve the accuracy of non-contact BP monitoring systems, thereby advancing the fields of remote health diagnostics and patient monitoring.