Classification of Phonocardiogram Based on Multi-View Deep Network

A phonocardiogram (PCG) is a plot of high-fidelity recording of the sounds of the heart obtained using an electronic stethoscope that is highly valuable in clinical medicine. It can help cardiologists diagnose cardiovascular diseases quickly and accurately. In this paper, we propose a multi-view deep network for the classification of PCG signals that can extract rich multi-view features from different modalities of PCG for classification. The model is mainly composed of two branches. In the first branch, we divide each PCG signal into three equal-length sub-signals, using Gramian Angular Fields to encode them from audio modality to two-dimensional image modality, and then Res2Net is applied to extract the image view features. In the second branch, we propose MobileNet-LSTM to extract the features of another view from preprocessed PCG signals. Finally, the features from these two views are fused and fed into the classifier for classification. Experiments show that our proposed method achieves 97.99% accuracy on the 2016 PhysioNet/CinC Challenge dataset, which is very competitive compared with the existing baseline models. In addition, the ablation experiment proves the necessity and effectiveness of our proposed method.