Explainable AI-Based ECG Heartbeat Classification Using Deep Learning Models

The ECG is fundamental in detecting cardiovascular ailments, and its thorough analysis is imperative. The analysis of conventional ECG depends on interpretation based on the expertise of the professional, which is slow and prone to mistakes. Deep learning models, which include CNNs, LSTMs, and Attention techniques introduced recently, have demonstrated promising results in the automation of ECG classification with high accuracy. However, a major drawback regarding these models is the ‘black box’ nature, which goes against clinical usage and comparison. This study adopts SHapley Additive exPlanations (SHAP) values and Local Interpretable Model-agnostic Explanations (LIME) techniques under deep learning algorithms for identifying ECG heartbeats. The model is trained and tested on the MIT-BIH Arrhythmia Database, and the performance seems to be accurate with an overall accuracy of 98.25%. Local Interpretable Model-Agnostic Explanations involving the construction of a local model, such as LIME and Shapely Additive explanations, or SHAP, help to provide more explanation regarding what the model is doing, thus increasing its reliability for clinical use. In this paper, the focus is presented on the experiment's method and outcomes, with a discussion of the interpretability of the final model as a crucial factor in its application in medical diagnostics.