Rhythm identification of wide complex tachycardia on electrocardiogram using a convolutional neural network

Background Wide complex tachycardia (WCT) on electrocardiogram (ECG) is often difficult to interpret and has important therapeutic implications. Accurate distinction between ventricular tachycardia (VT), supraventricular tachycardia (SVT) with aberrancy, and ventricular-paced (VP) rhythms remains challenging. Artificial intelligence (AI) using convolutional neural networks (CNNs) offers promise in ECG interpretation through pattern recognition. Purpose To develop and evaluate a CNN model capable of classifying WCTs into VT, SVT, or VP rhythms using ECG data and expert electrophysiologist (EP) adjudication as the diagnostic reference. Methods A convolutional neural network (CNN) based on a residual network architecture was trained using95% diagnostic confidence. CNN outputs were compared with cardiologist interpretations, and disagreements were further reviewed by an electrophysiology panel using established diagnostic criteria across available ECGs. Model performance was assessed using receiver-operating-characteristic analysis and confusion matrices against both cardiologist and EP-adjudicated reference standards. Results The mean patient age was 77.5 years, with 35% female. Nearly half (48%) had heart failure, and 28% had left ventricular ejection fraction below 40%. Against EP-adjudicated diagnosis, the CNN achieved high discrimination with area under the ROC curve (AUC) of 0.974 for SVT, 0.964 for VT, and 0.980 for VP rhythms (average AUC = 0.972). Performance improved when benchmarked against EP adjudication compared with original clinical ECG reports. Explainability analysis is underway to identify waveform regions contributing to classification accuracy. Conclusion A CNN trained on ECG waveform data accurately differentiates WCT into VT, SVT, and VP rhythms, achieving very high accuracy against a novel reference standard. AI-based rhythm classification may serve as a valuable adjunct in the interpretation of challenging ECGs and could support clinical decision-making in electrophysiology.