12-Lead ECG Heartbeat Classification Using the Progressive Moving Average Transform (PMAT) and Convolutional Neural Networks

Abstract

This study presents an automated method for classifying electrocardiogram (ECG) heartbeats into Normal, Supraventricular, and Ventricular categories. To improve diagnostic reliability, we developed a hybrid CNN-MLP model that processes two ECG leads simultaneously. The CNN extracts morphological features from PMAT transformed images, while the MLP analyzes RR-interval timing. In this experiment, we systematically trained every possible dual-lead combination to identify the optimal pairings. Results show that leads 4 and 10 achieve the best performance for critical arrhythmia detection, reaching 98.52% accuracy. Meanwhile, leads 7 and 10 provide the most reliable configuration for general monitoring, achieving 97.31% accuracy. Despite challenges with rare Supraventricular beats, this approach effectively combines waveform shape and rhythm patterns, offering a practical tool for automated ECG analysis.