A robust deep learning strategy for long-term sequence mapping of BCG and ECG based on an optical fiber sensor

Abstract

Since cardiovascular diseases (CVDs) have become the leading cause of death globally, continuous heartbeat monitoring holds particular importance, especially for the elderly population; however, the use of multiple electrodes attached to the body during signal collection may not be acceptable for patients. At present, ballistocardiogram (BCG) signal offers an unobtrusive solution for medical institutions, yet BCG has no uniformly confirmed gold standard for the non-contact diagnosis as electrocardiogram (ECG). In this paper, we proposed a non-invasive cushion based optical fiber sensor (OFS) with a deep learning model. The key novelty of our research is to develop a deep learning model feasible for non-contact OFS to achieve BCG-ECG mapping. Compared with traditional ECG measurement, ours presents a more user-friendly and convenient solution for the patients to collect cardiac data using OFS. We evaluated the generated ECG by using quantitative metrics: signal-to-noise ratio (SNR) and root mean square error (RMSE); and evaluation plots: Bland-Altman Plot, histogram distribution of prediction data and ground-truth data, and histogram of prediction errors. The involved datasets for training and testing are a public dataset and self-collected dataset. In comparison to other recent methods, our proposed model presents a lower prediction error of 0.21, higher SNR of 12.53, and better predicted data distribution on both the public and self-collected dataset. Simultaneously, the model presents an appropriate number of training parameters and training time in comparison to the state-of-the-art, giving the possibility of embedding the network into OFS for inference. Our research would provide a meaningful value for both the non-invasive CVD diagnosis based on OFS and BCG-ECG mapping.