A Hybrid System for Myocardial Infarction Classification with Derived Vectorcardiography

The 12-lead electrocardiography (ECG) remains the most rapid and widely used diagnostic test for patients with myocardial infarction (MI). Most wearable ECG devices only provide single limb-lead measurement, limiting their practical applicability for MI diagnosis. The ability to transform from single-lead ECG to 3-lead vectorcardiography (VCG) enables wider use of wearable devices in clinical diagnostics. This study presents a patient-specific transformation for VCG synthesis using temporal convolutional networks in variational mode decomposition domain. MI-induced changes in morphological and temporal wave features are extracted from the derived VCG via spline curve approximation. After feature extraction, a multilayer perceptron classifier is used to classify different types of MI. Experiments on the PTB diagnostic database show that the proposed system achieves satisfactory performance in differentiating MI patients from healthy subjects and localizing infarcted area.