An efficient deep learning solution for cardiac signal assessment based on optical fiber sensor

Cardiovascular diseases (CVDs) have emerged as the leading cause of mortality worldwide, making the need for continuous heartbeat monitoring especially imperative; however, the conventional approach of attaching multiple electrodes to the body for signal acquisition may be uncomfortable and often unacceptable to elderly patients who need this inspection the most. Ballistocardiography (BCG) provides a non-intrusive alternative for healthcare settings, but there is still no universally accepted standard for evaluating the quality of BCG signals obtained by non-invasive optical devices. In this study, we introduced a non-invasive, cushion-based optical fiber sensor (OFS) first, and then proposed a deep learning model. The primary innovation of our work lies in employing a non-invasive OFS to translate BCG signals into their corresponding ECG signals, which are widely regarded as the gold standard for cardiac assessment. Compared to traditional electrocardiography (ECG) devices, our solution offers a more patient-friendly and convenient method for cardiac data collection. Experimental results demonstrate that our network can generate high-quality ECG signals from raw BCG data. This research provides valuable insights for non-invasive CVD diagnosis using BCG signals.