Abnormal Heart Sound Detection by Using Temporal Convolutional Network

Abnormal heart sound detection is great of significance because of the frequent occurrence of heart diseases. However, the automatic diagnosis for abnormal heart sound has a high requirement for domain knowledge and the signal noise poses an increased difficulty of diagnosis. In this paper, we propose a temporal convolutional network (TCN) to automatically detect abnormal heart sounds. Specifically, a noise removing technology is applied to original signals. Then, a TCN architecture is carefully designed to adapt the properties of heartbeat sound. The proposed method is tested on the Physionet dataset, and the results show our method contains potential ability in abnormal heart sound detection.