Smart Myocardial Infarction Detection System Using Deep Learning, Ensemble Classifiers, and Explainable Artificial Intelligence from Single-Lead ECG Signals

Abstract

Myocardial infarction (MI) is one of the most important cardiovascular diseases that causes a large proportion of deaths worldwide, and thus, it is important to have reliable and timely detection methods. The purpose of this paper is to present a smart myocardial infarction detection system using a hybrid of deep learning, ensemble classifiers, and EAI techniques by applying single-lead electrocardiogram (ECG) signals. The system is based on Discrete Wavelet Transform (DWT) technique to extract time–frequency features of ECG signals, and an autoencoder network to perform unsupervised feature compression. The compressed representations have been classified using ensemble learning models to identify and localize MI into 4 classes: Normal, Anterior MI, Inferior MI, and Lateral MI, namely a Random Forest and an XGBoost. SHAP (SHapley Additive exPlanations) is included as a means to improve the interpretability and clinical reliability of the model, by emphasizing the features that are most important to the prediction. The proposed framework is designed to be memory efficient for processing and deployable using a FastAPI based backend with a web front-end, allowing for real-time analysis and access to the system in telemedicine and remote healthcare settings. The results show good classification rate and the proposed approach is effective for the support system of cardiac diagnosis systems, as it enhanced the sensitivity for the classes of MI.