An Explainable CNN-BiGRU Model for Arrhythmia Classification from ECG Signals

Electrocardiogram (ECG) signals have served as valuable tools by medical professionals for diagnosing heart arrhythmias, but their complexity makes classification difficult for even the most skilled individuals. Deep Learning (DL) models allow the process to be sped up, allowing for accurate and auto-matic classification. However, they prompt the new issue of being uninterpretable, which is problematic in scenarios where lives are at risk. In this paper, I propose a hybrid DL model with an Explainable Artificial Intelligence (XAI) framework, alleviating the black box effect associated with standard deep learning models. The classification of 5 different heart rhythms was performed by a 9-layer neural network consisting of Convolutional Neural Network (CNN) and Bidirectional Gated Recurrent Unit (BiGRU) layers. Transparency is done by application of the Locally Interpretable Model-agnostic Explanation method (LIME) to the time series data associated with ECG signals. The model achieved an accuracy of 94.44%, precision of 94.33%, recall of 94.36%, and F1-score of 94.35%. The successful results of the model and XAI indicate strong potential in clinical applications, increasing trust in AI systems.