|
KMID : 0390320220320020007
|
|
Chungbuk Medical Journal
2022 Volume.32 No. 2 p.7 ~ p.17
|
|
|
Comparison of Atrial Fibrillation Classification Performance According to Electrocardiogram Signal Domain Using Long Short-Term Memory Neural Network
|
|
Yang Min-Uk
Lee Tae-Soo
|
|
|
Abstract
|
|
|
Purpose: Atrial fibrillation (AF) is one of the heart arrhythmias in which the atrial contraction disappears and irregular contraction occurs and is a leading risk factor for various cardiovascular disease complications. Recently, although electrocardiogram (ECG) signals have been analyzed on various signal domains to accurately classify the AF ECG, which signal domain is the most useful has not been verified. Therefore, we performed the experiments over time, frequency, and entropy domains to find the most useful signal domains.
Materials and Methods: For the comparison evaluation, all experiments are performed on the PhysioNet/Computing in Cardiology Challenge 2017 database, a public dataset containing more than 8,000 ECG signals. The ECG signal of each domain was analyzed using a long short-term memory neural network, which is a deep learning-based analysis model.
Results: The ECG signal analysis in the frequency domain showed the best performance, with an area under the receiver operating characteristic curve (AUROC) score of 0.9258. In contrast, the signal analysis in the time domain presented the worst performance, with an AUROC score of 0.5767. Furthermore, the analysis performed in both the time and frequency domains achieved an AUROC score of 0.9529, showing a better performance than each single-domain signal analysis.
Conclusion: This study suggests that the method of analyzing multiple-domain signals can extract useful features for atrial fibrillation classification and analyze them.
|
|
|
KEYWORD
|
|
|
Electrocardiogram, atrial fibrillation, deep learning, long short-term memory, signal domain, classification
|
|
|
FullTexts / Linksout information
|
|
|
|
|
|
Listed journal information
|
|
|
|
|