부산대학교 도서관

Medical practitioners need to understand the critical features of ECG beats to diagnose and identify cardiovascular conditionsaccurately. This would be greatly facilitated by identifying the signifi cant features of frequency components in temporalECG wave-forms using computational methods. In this study, we have proposed a novel ECG beat classifi er based ona customized VGG16-based Convolution Neural Network (CNN) that uses the time-frequency representation of temporalECG, and a method to identify the contribution of interpretable ECG frequencies when classifying based on the SHapleyAdditive exPlanations (SHAP) values. We applied our model to the MIT-BIH arrhythmia dataset to classify the ECG beatsand to characterise of the beats frequencies. This model was evaluated with two advanced time-frequency analysis methods. Our results indicated that for 2-4 classes our proposed model achieves a classifi cation accuracy of 100% and for 5 classes itachieves a classifi cation accuracy of 99.90%. We have also tested the proposed model using premature ventricular contractionbeats from the American Heart Association (AHA) database and normal beats from Lobachevsky University Electrocardiographydatabase (LUDB) and obtained a classifi cation accuracy of 99.91% for the 5-classes case. In addition, SHAP valueincreased the interpretability of the ECG frequency features. Thus, this model could be applicable to the automation of thecardiovascular diagnosis system and could be used by clinicians.