Deep learning models for atrial fibrillation detection: A review

Abstract

Detection of atrial fibrillation (AF) presents one of the main tasks of modern cardiology. In the last few years, the deep learning (DL) emerges as the most frequent approach for accomplishing the task. When deciding to apply DL model for AF detection researchers are facing different choices bringing specific advantages but also imposing specific restrictions. The expansion of publishing, and advancements in this field, demand frequent review of the state of the art. The initial set of 370 papers filtered by keywords of interest, were systematically narrowed to 32 papers in focus. The objective of the paper is to present a comprehensive overview of commonly used ECG databases, signal preprocessing techniques, inputs formatting, DL models used, choice of output classes, and performance metrics achieved.