<![CDATA[Arrhythmia is one of the heart abnormalities which probably not a life threat in a short time but could cause a long-term interference in electricity of the heart. Even so, it should be detected earlier to have proper treatment and suggest a better lifestyle. Arrhythmia diagnosis is usually made by performing a long recording ECG by using Holter monitoring then analyzing the rhythm. Nevertheless, the observation takes time, and using Holter in several days may affect the patient’s physiological condition. Previous research has been conducted to build an auto-detection of arrhythmia by using various datasets, different features, and detection methods. However, the biggest challenges faced by the researcher were the computation and the complex features used as the algorithm input. This study aims to review the latest research on the data used, features, and deep learning methods that can solve the time computation problem and be applied in wearable devices. The review method started by searching the related paper, then studied on the data used. The second step was to review the used ECG features and the deep learning method implemented to detect arrhythmia. The review shows that most researchers used the MIT-BIH database, even it requires a lot of effort on the pre-processing. The CNN is the most used deep learning method, but time computation is one of the considerations. The ECG interval features in the time domain are the best feature analysis for rhythm abnormality detection and have a low computation cost. These features will be the input of the deep learning process to reduce computation time, especially on wearable device applications.]]>
Copyrights © 2022
