Optimized Convolutional Neural Network by Genetic Algorithm for the Classification of Complex Arrhythmia

Complex arrhythmia is a serious cardiovascular disorder. Delayed diagnosis of complex arrhythmia can cause serious consequences, such as blood clots, stroke, heart failure, and sudden death. However, for accurate diagnosis and treatment, correct identification of the type of arrhythmia is important. Therefore, in this study, for automatic multiclassification of surface electrocardiogram signals (i.e., atrial fibrillation [AFIB], atrial flutter [AFL], supraventricular tachycardia [SVTA], ventricular tachycardia [VT], ventricular flutter [VFL], sinus rhythm [N], and NOISE), an optimized 17-layer-deep convolutional neural network (CNN) was constructed using a genetic algorithm and grid search. Various, indicators were applied to evaluate the classification performance of the constructed CNN. The resulting accuracy, sensitivity, specificity, precision, and macro_F1 score of the constructed CNN were 95.87%, 95.78%, 99.18%, 96.02%, and 95.87%, respectively. The results showed that the proposed CNN can be a powerful tool for multiclassification of complex arrhythmia.