ECG abnormalities classification using Deep Learning

Electrocardiography (ECG) is the key to diagnosis of ECG abnormalities including arrhythmia, which represents a great dilemma for physicians. The topic of how to better use deep learning to aid ECG diagnosis has been widely studied in recent years. However, it has not yet been used in the identification of ECG abnormalities with wide QRS complex in normal heart beat. We constructed a dataset of ECGs with 5 distinct kinds of ECG abnormalities with wide QRS complex in normal heart beat, and the normal ECGs (N) served as control group. We developed an AI-enabled system to generate diagnoses of these abnormalities in this dataset. For evaluating the performance of the model, the F1 score, area under the curve (AUC) of receiver operating characteristic curve (ROC), accuracy, sensitivity, specificity, and the 95% CI for both sides were calculated separately for the model and the diagnoses result of 3 physicians. Our deep learning model resulted in an accuracy of 96.6%, which is higher than the doctors' average of 86.8%. Our model is promising to distinguish the abnormal part of the normal heart beat containing the wide QRS complex, laying an essential foundation for computer-aided diagnosis (CAD) systems in clinical applications.