A Comprehensive Multi-Strategy Transfer Learning Framework with Knowledge Distillation for ECG Image Classification: A Multi-Dataset Validation Study

Abstract

Cardiovascular diseases remain the leading cause of global mortality, with approximately 17.9 million deaths reported annually worldwide. Electrocardiogram (ECG) analysis is essential for accurate diagnosis but requires expert clinical interpretation, significantly limiting accessibility in resource-constrained healthcare settings. This study proposes a novel multi-strategy transfer learning framework for automated ECG image classification, effectively integrating knowledge distillation, uncertainty quantification, and a confidence-based clinical decision support system. Extensive experiments were conducted on multiple public ECG image datasets to evaluate performance and generalizability. The proposed framework achieved 99.64% accuracy with cross-validation accuracy of 99.42 ± 0.18% and AUC of 0.9997 on the Cardiovascular ECG dataset. Knowledge distillation reduced model parameters by 8.5× while maintaining performance. Uncertainty estimation showed strong calibration (ECE = 0.0017), enabling reliable predictions. The triage system achieved 99.6% accuracy on 97.1% automatically accepted cases. Ultimately, these comprehensive evaluation results successfully demonstrate a highly scalable, computationally efficient, and clinically reliable ECG classification framework that is exceptionally well-suited for seamless integration and real-world deployment within modern, resource-constrained healthcare systems.

Received: 30 November 2025 | Revised: 2 February 2026 | Accepted: 29 May 2026

Conflicts of Interest

The authors declare that they have no conflicts of interest to this work. 

Data Availability Statement

The data that support the findings of this study are openly available in Cardiovascular ECG Images at https://www.kaggle.com/datasets/jayaprakashpondy/ecgimages, National Heart Foundation 2023 ECG Dataset at https://doi.org/10.17632/gwycg64pfx.1, ECG Images Dataset of Cardiac Patients at https://data.mendeley.com/datasets/gwbz3fsgp8/2, and ECG Image Data at https://www.kaggle.com/datasets/erhmrai/ecg-image-data. 

Author Contribution Statement

Md. Rahat: Conceptualization, Methodology, Software, Validation, Formal Analysis, Investigation, Data curation, Writing – original draft, Visualization. Rahat Hossain Faisal: Supervision, Project administration, Writing – review & editing. Khondoker Razinul Karim: Validation, Data curation, Writing – review & editing. Md. Minhaj Ul Islam: Methodology, Validation, Writing – review & editing. Md Zahid Akon: Validation, Writing – review & editing.