An Explainable Approach to Characterize Heart Diseases Using ECG Images

Cardiovascular diseases are a group of disorders that involve both the heart and the blood vessels and represent one of the main causes of death all over the world. Their early identification plays a critical role in the prevention and effective treatment of these conditions, as cardiovascular diseases often progress silently without obvious symptoms in their early stages. In this regard, identifying the risk factors and the first manifestations of these diseases is crucial to intervene promptly and reduce the risk of serious complications. The electrocardiogram (ECG) is a widely used diagnostic tool to evaluate the electrical activity of the heart. This test provides crucial information about the functioning of the heart and can be useful in diagnosing various cardiovascular conditions. Its clinical utility, together with its simplicity of execution and its wide availability, makes such a test one of the pillars of cardiovascular medical practice. The objective of this study is to develop an approach to predict heart disease through the analysis of ECG images including a glimpse of explainability. Such an approach is based on a customized two-dimensional convolutional neural network (CNN-2D) combined with the explainability provided by the Gradient-weighted Class Activation Mapping (Grad-CAM) tool. The validation was conducted on a public dataset, made of ECG pictures of cardiac patients obtained from the Ch. Pervaiz Elahi Institute of Cardiology of Multan in Pakistan. The results have been very satisfactory and have shown the effectiveness of the proposed methodology, both in terms of performance and interpretability, which could represent a valid tool to support doctors in identifying both myocardial infarctions and arrhythmias in turn.