TCN-Based Diagnostic Model for the Severity of Coronary Atherosclerotic Heart Disease Using Wrist Pulse Wave Sequence

Abstract

The pulse wave at the human radial artery is closely related to the health status of the cardiovascular system. In this paper, the morphological features of the pulse wave were used to establish a diagnostic model for the severity of coronary atherosclerotic heart disease (CAD). Features of waveform variations were extracted from pulse wave sequences by building a deep learning network, Temporal Convolutional Network (TCN), which mined more detailed waveform information and obtained more comprehensive features of waveform morphology than the classical time domain features extraction method, thus established a TCN-based CAD severity diagnostic model (TCSDM) with better performance. The 64 features extracted by TCN have shown significant differences between the three classes of CAD samples at the 0.05 level, which have provided additional basis for the model's classification decisions. The accuracy of TCSDM has reached 91.17%, an 11.93% improvement compared to the Random Forest-based diagnostic model using classical time domain features. The proposed method for the acquisition of pulse wave morphological features can effectively extract the differential features of different pulse waves. And this method has a great application value in the remote diagnosis of CAD severity because it's non-invasive, rapid and low-cost.