Leveraging 3D Heart Visualisation and Data Balancing Techniques for ECG Classification.

Cardiovascular diseases are among the most prevalent global health conditions, making the accurate diagnosis and classification of cardiac abnormalities crucial for effective treatment and patient management. While the electrocardiogram (ECG) is the primary tool for assessing cardiac electrical activity, its manual analysis is often time-consuming and susceptible to interpretive error. To address these limitations, this work proposes a comprehensive deep learning pipeline for the automated classification of arrhythmias, incorporating specific strategies to mitigate the challenge of imbalanced datasets. Furthermore, we introduce a novel three-dimensional (3D) visualisation framework that provides interactive, anatomically precise renderings of the heart regions implicated by the ECG classification, thereby delivering enhanced diagnostic insight. Our evaluation demonstrates that the proposed data balancing techniques yield significant performance gains, and under our current experimental setup, the results are competitive with or exceed several previously reported methods. We acknowledge that a more rigorous inter-patient cross-validation is needed to fully establish generalisation. The resulting 3D visualisations not only enable precise anatomical localisation of arrhythmia substrates but also serve as a powerful interactive tool for clinical practice and medical education.