WavelNet을 활용한 초단기 심전도로부터의 심방세동 검출

Abstract

Detecting atrial fibrillation in early stages is crucial for preventing its progression and cardiovascular complications. However, its symptoms in the early stages are intermittent and brief, making it difficult to be detected using existing solutions. We propose WavelNet-based automatic atrial fibrillation detection model, which can detect atrial fibrillation from ultra-short-term electrocardiograms while minimizing manual preprocessing. As a result of development and evaluation using the MIT-BIH Atrial Fibrillation Database, WavelNet-based models outperformed vanilla convolutional neural network- and SincNet-based models. Particularly, a WavelNet-based model with adopting Symlet-6 mother wavelet achieved the best performance in terms of accuracy and F1-score, as 86.8% and 0.883, respectively.