Enhanced Arrhythmia Diagnosis via Three-Heartbeat Multi-Lead Ecg And Deep Learning Fusion Models

Abstract

Accurate classification of cardiac arrhythmias is essential for effective clinical diagnosis and treatment. In this study, we propose a novel methodology that, for the first time, utilizes Three-Heartbeat Multi-Lead (THML) ECG data, where each data segment includes three complete heartbeat cycles across multiple ECG leads. This enriched temporal and spatial representation provides a more comprehensive view of cardiac activity, enabling enhanced pattern recognition. Two classification models are developed and compared: a Weight Convolutional Neural Network (WCNN) and a hybrid Light Gradient Boosting Machine-Convolutional Neural Network (LGBM-CNN). The performance of the models was rigorously evaluated using accuracy (Acc), sensitivity (Sen), and positive predictive value (PPV). Experimental results demonstrate that the LGBM-CNN significantly outperforms the WCNN across all metrics, indicating its superior capability in capturing the complex morphological features of arrhythmic ECG signals. These findings underscore the effectiveness of THML ECG representation combined with advanced hybrid architectures for automated arrhythmia classification.