Classifcation of atrial fbrillation and normal sinus rhythm based on convolutional neural network

• Published in: Biomedical engineering letters pp. 183 - 193
• Main Authors: Mei-Ling Huang, Yan-Sheng Wu
• Format: Journal Article
• Language: English
• Published: 대한의용생체공학회 01.05.2020
• Subjects: 의공학
• ISSN: 2093-9868; 2093-985X
• DOI: 10.1007/s13534-020-00146-9

Electrocardiogram (ECG) technology plays a vital role in detecting arrhythmia. Numerous achievements have been marked in ECG-related research. Most methods frst pre-process ECG signals, then extract features, and fnally classify them. Most of the ECG signals used in the related studies were analyzed in specifc time intervals or using a fxed number of samples. However, it is not always possible to see signifcant changes in a short term, and the symptoms of some patients are relatively short-lived. Misjudgments are possible because the ECG signal was not accurately extracted. This study proposes a computeraided diagnosis (CAD) system for classifcation of Atrial Fibrillation and Normal Sinus Rhythm based on ECG signals through convolutional neural network. The proposed system considers a single heartbeat, rather than a specifc number of seconds. This study eschews the one-dimensional digital ECG signal used in previous studies and uses convolutional neural networks to analyze two-dimensional ECG image. This study explores whether two-dimensional image ECG requires signal fltering. The fnal classifcation results in fltered ECG signals is accuracy of 99.23%, sensitivity of 99.71%, and specifcity of 98.66%. The best result in non-fltered ECG signals achieves accuracy of 99.18%, sensitivity of 99.31%, and specifcity of 99.03%. With no cumbersome artifcial settings, the results of this study are comparable to the related studies. The proposed CAD system has high generalizability; it can help doctors to diagnose diseases efectively and reduce misdiagnosis. KCI Citation Count: 0