Deep Learning-Based Classification of Premature Ventricular Contractions and Supraventricular Tachycardia Using ECG Signals

This study provides comparative evaluation of deep learning models for the classification of PVC and SVT from ECG signals. We evaluate ResNet, DenseNet, VGGNet, InceptionNet, and WaveNet performance on various measures like accuracy, precision, recall, F1 score, AUC, MCC and log loss. The findings are that ResNet performs the best for both PVC and SVT classification tasks with 94% for PVC and 98% for SVT. Whereas all the models perform best for PVC detection, DenseNet experiences a steep drop in performance for SVT classification, indicating that it requires optimization. The results show the effectiveness of deep learning models, ResNet and WaveNet, in the correct diagnosis of arrhythmias from ECG signals.