Noninvasive Human Ballistocardiography Assessment Based on Deep Learning

Abstract

Ballistocardiography (BCG) is a vibration signal of human cardiac activity, which can be obtained by optical fiber sensor in a non-invasive way. The proposed optical fiber sensor as a low power consumption, non-contact, non-invasive real-time health monitoring instrument, has been developed into an effective health care monitoring method. However, when people need to monitor BCG for a long time, a large number of BCG data needs to be collected, which is time-consuming, costly and labor-intensive. To solve this problem, in this paper, we proposed a novel deep learning model, termed BCGNET. Firstly, convolutional neural network (CNN) is used to extract the short-term dependence between multivariate loads. Then, the recurrent neural network (RNN) model is used to capture the long-term dependence of load sequence, and the ultra-long-term repetitive pattern of load sequence is fully studied by using the long-short term memory network with recurrent-skip. Finally, autoregressive layer and full connection layer are used for combined prediction. Extensive experimental results demonstrate the superiority of BCGNET, the accuracy is 91.43% achieved by the proposed BCGNET compared with CNN (89.61%), RNN (89.88%) and multi-head attention network (MHA-Net) (90.22%), and also show the proposed model has good performance in BCG prediction and assessment.