Explainable Deep Learning with Transfer Learning for Scalable ECG-based Cardiac Abnormality Detection

Cardiovascular diseases (CVDs) have continued to be one of the leading causes of mortality in the world, which requires prompt and precise diagnostic methods. The interpretation of the electrocardiogram (ECG) data is usually difficult because of the subtle waveform variations, inter-patient variations and domain variations across datasets. Despite promising results of deep learning-based methods in automated ECG classification, the current methods are limited by low generalization, poor interpretability, and low cross-dataset adaptability. To overcome these issues, this paper suggests GENCardioXplain, a hybrid deep learning model of generalized and explainable ECG-based cardiac abnormality detection. The model combines Temporal Convolutional Networks (TCN) and Bidirectional Gated Recurrent Units (Bi-GRU), to learn multi-scale temporal dependencies, and Adaptive Batch Normalization (AdaBN) to achieve easy domain adaptation to achieve better cross-dataset performance. Moreover, a guidance-based explainability module that integrates attention mechanisms with SHAP-based attribution gives clinically interpretable explanations. The effectiveness of the proposed framework is demonstrated by extensive experiments on multiple benchmark ECG datasets, achieving an accuracy of 94.22, precision of 93.84, F1-score of 93.43 and MCC of 0.9011. The studies of ablation, cross-dataset analyses, and statistical significance analysis are additional evidence that it is robust and reliable. The findings demonstrate the promise of GENCardioXplain as a scalable, interpretable and clinically applicable solution to real-time cardiac diagnosis, especially in telemedicine and remote healthcare settings.