Hybrid Deep Learning Approaches for Single-Lead ECG-Based Atrial Fibrillation Detection

Abstract

Atrial fibrillation (AF) is one of the most prevalent cardiac arrhythmias, significantly contributing to stroke, heart failure, and mortality worldwide. Early and accurate detection is essential for effective treatment and prevention. Recent advancements in artificial intelligence and signal processing techniques have enabled improved AF detection using single-lead electrocardiogram (ECG) signals. This systematic review focuses on hybrid approaches combining Fast Fourier Transform (FFT) and Continuous Wavelet Transform (CWT) with stochastic pooling-based neural networks for enhanced classification accuracy. The integration of FFT enables efficient frequency-domain feature extraction, while CWT provides superior time–frequency localization, making them highly complementary. Deep learning models, particularly convolutional neural networks (CNNs) enhanced with stochastic pooling, further improve generalization and reduce overfitting. Literature demonstrates significant improvements in AF detection performance, achieving accuracy rates exceeding 95% in many studies. Additionally, hybrid architectures incorporating residual networks, temporal convolutional networks, and multi-branch CNN frameworks have shown robust performance in handling imbalanced datasets. This review presents a comprehensive comparative analysis of recent methodologies, highlighting their advantages, limitations, and future research directions. The findings indicate that hybrid signal processing combined with advanced neural architectures represents a promising direction for reliable, real-time AF detection using wearable and portable ECG devices.