A Deep Knowledge Distillation HeartCare Framework for Detection of Multi-label Myocardial Infarction from Multi-lead ECG Signals

Myocardial Infarction (MI) is the most frequent form of Coronary Heart Disease (CHD), which is the world’s leading cause of illness and mortality. More than 15% of deaths per year are caused by MI. An extensively utilized diagnostic technique for cardiovascular problems is the 12-lead Electrocardiogram (ECG). Deep learning algorithms have outperformed traditional feature engineering-based techniques in terms of performance. By analyzing ECG data, deep learning models can generate clinical assessments that are comparable to those of cardiologists. The article proposes a framework for detecting and localizing MI using a convolutional neural network. The method employs a knowledge distillation approach and accepts a 12-lead ECG as input. The framework has been developed and validated using the PTB-XL dataset. The framework achieved the ability to identify and locate five super-diagnostics classes included in the PTB-XL dataset. The framework has been validated successfully, with an average accuracy of \(94.25\%\) $94.25\%$ and a micro-averaged receiver operating characteristic (ROC) of 0.95. In addition, the model is validated using a combined dataset, prepared from the Shaoxing People’s Hospital (SPH) dataset, the China Physiological Signal Challenge (CPSC) 2018 dataset, and a private dataset containing ECG data from Indian patients. This allows the scientific community to assess the applicability of the proposed method to various patient groups and evaluate its performance in detecting MI. During the cross-validation process, the proposed structure achieves a mean accuracy of \(95.80\%\) $95.80\%$ and a sensitivity of \(84.98\%\) $84.98\%$ .