Explainable Rhythm-Based Heart Disease Detection from ECG Signals

Healthcare decision support systems must operate with confidence and trust. Numerous researchers have attempted to automate the identification and classification of cardiovascular conditions from electrocardiogram (ECG) signals. One particular area of research involves utilizing deep learning (DL) for the classification of ECG signals into various heart disease classes. However, DL models do not provide information on why they reached their final decision. This makes it difficult to trust their output in a medical environment. To address this trust issue, ongoing research aims to explain the decisions made by DL models. Some approaches have been used to improve the interpretability of DL models, using the Shapley Value (SHAP). However, SHAP’s explanation happens to be computationally expensive. In this research, we develop a deep learning model that can detect five rhythm-based heart diseases with explainability. We employ visual explanations algorithms, that are Grad-CAM and Grad-CAM++, as an explainability framework. These explainers are relatively lightweight and can be executed quickly on a standard CPU or GPU. Our model was trained using 12-lead ECG signals from the PTB-XL large dataset. Our model was effective, with a classification accuracy of 0.96 and an F1 of 0.88. The outputs of the model were visually explained using Grad-CAM and Grad-CAM++ of which Grad-CAM++ produced more localized explanations. To evaluate the explainability, we gave ten randomly selected outputs to two domain experts. The explanations that were not equally accepted by the experts still had a consensus in many of the individual leads out of the 12.