Ramanujan filter bank-domain deep CNN for detection of atrial fibrillation using 12-lead ECG

Abstract

Atrial fibrillation (AF) is a type of heart ailment characterized by abnormal and chaotic atrial activity in the heart. Twelve-lead electrocardiogram (ECG) recording is the primary diagnostic test performed in clinical settings to diagnose AF. The automated and early detection of AF using 12-lead ECG using artificial intelligence techniques is challenging for patient monitoring. This chapter proposes a novel approach for optimal lead selection and automated detection of AF-based cardiac ailments using 12-lead ECG signals. The Ramanujan filter bank is introduced to evaluate each lead's time-period representation (TPR) of the ECG signal. A single-channel TPR-domain deep convolutional neural network (CNN) architecture is proposed to select three optimal ECG leads from 12-lead ECG signals. The TPR images of the 3-lead ECG signal and multichannel deep CNN are employed to detect AF automatically. The performance of the suggested AF detection approach is evaluated using a publicly available 12-lead ECG signal dataset. The results show that the proposed AF detection approach has sensitivity, specificity, and accuracy values of 97.21%, 94.49%, and 95.84%, respectively, using the TPR of three selected ECG leads. The results are also compared with those obtained using other deep learning- and time-frequency-based transform-domain CNN methods for detecting AF. © 2024 by Elsevier Inc. All rights reserved, including those for text and data mining, AI training, and similar technologies.