ECG Signal Classification Using Temporal Convolutional Network

A large part of the world's population is concentrated in rural areas. In such territories, medical facilities are often deficient or out of reach. The situation is not much better in developed countries, especially in the case of heart diseases where the rate of fatality is significantly high. Thus, it is unlikely that an electrocardiogram (ECG) analysis, responsible for heart visualization, can be continuously performed without specialized methods and equipments. However, the use of remotely-controlled technologies for continuous and better health care results appears as a promising solution to tackle this problem. Indeed, sensing and actuating technologies along with big data analysis provide basic building blocks for Remote Health Monitoring (RHM). These technologies are continuously evolving starting from the miniaturization of the sensors to collectively making sense out of varied data types in different forms of structured, unstructured, video, or sensory data. In this paper, we propose a novel machine learning (ML) architecture for the ECG classification of five heart diseases, based on temporal convolution networks (TCN). The proposed design, which implements a diluted causal one-dimensional convolution on the input heartbeat signals, seems to be outperforming all existing ML methods with an accuracy up to 98.54% and a F2 score of 94.51%, using a reduced number of parameters (10.2k). Consequently, the obtained results make the proposed TCN architecture a good candidate for low power consumption hardware platform implementation.