Noninvasive sleep and cardiovascular health monitoring from ballistocardiogram signals using bed sensor

Abstract

Monitoring sleep and cardiovascular health plays a crucial role in assessing overall well-being and detecting early signs of physiological disorders. Conventional sleep assessment relies on polysomnography (PSG), which, although comprehensive, is intrusive, expensive, and limited to short-term clinical use. Recent advances in wearable and non-wearable sensing technologies have enabled more comfortable and continuous home-based monitoring. In this study, we propose a noninvasive bed sensing system based on ballistocardiogram (BCG) signals for sleep and cardiovascular monitoring. The system captures subtle body vibrations associated with cardiac and respiratory activity without direct skin contact. We investigated multiple datasets spanning diverse populations, from young, healthy participants in a public dataset to older adults with sleep and cardiovascular disorders in a private dataset. Various analytical approaches were explored, ranging from hand-crafted feature extraction to deep learning models, including convolutional neural networks (CNN), long short-term memory (LSTM) networks, Transformer-based architectures for sleep stage classification and heart beat detection. The results demonstrate the capability of using BCG signals to estimate sleep stages and cardiovascular parameters while partially addressing challenges related to signal quality, inter-subject variability, and data imbalance. This work highlights the potential of unobtrusive BCG-based sensing, combined with deep learning approaches as a practical alternative for long-term, in-home sleep and cardiovascular health monitoring.