Detection of Shockable Arrhythmia From Electrocardiogram Signal Using Recurrence Quantification Analysis Based Deep Convolutional Neural Networks

Ventricular fibrillation and ventricular tachycardia are shockable cardiac symptoms. These ventricular arrhythmias (VAs) raise the risk of heart failure by causing heart attacks and other heart problems. The electrocardiogram (ECG) is a commonly used and trustworthy technique for identifying cardiac arrhythmia. In case of emergency patient conditions, acknowledging the limitations of manual cardiac arrhythmia detection methods underscores the pressing need for an automated system. This work presents an automated system for the classification of shockable arrhythmia (SA) and nonshockable arrhythmia (NSA) through ECG signals. The proposed work uses recurrence quantification analysis (RQA) to transform 1-D ECG signal into 2-D RQA image. The RQA images are classified into shockable and nonshockable classes on the first stage of classification using modified versions of the last layer of the pretrained AlexNet, DenseNet, ResNet, and EfficientNet. The second stage of classification refines the distinction between SAs [ventricular fibrillation (VF) and ventricular tachycardia (VT)] and NSAs (normal rhythms and other disorders). Experimental validation using the Physionet ECG database demonstrates the effectiveness of our proposed approach. Notably, the Modified EfficientNet-B0 architecture achieves an impressive overall test accuracy of 98.92%. This study contributes to the field by not only providing a robust method for SA detection but also reducing system complexity. The proposed approach eliminates the need for manual feature extraction or beat detection, enhancing its practical applicability in real-world scenarios.