Hybrid Fast Fourier-Wavelet Neural Networks for Single-Lead ECG-Based Atrial Fibrillation Detection: Methods and Challenges

Abstract

Atrial fibrillation (AF) is a prevalent cardiac arrhythmia associated with increased risk of stroke, heart failure, and mortality. Accurate and early detection of AF using single-lead electrocardiogram (ECG) signals has gained significant attention due to the rise of wearable healthcare devices. This survey presents a comprehensive review of recent methods and architectures aimed at enhancing AF detection accuracy through hybrid signal processing and deep learning techniques. Specifically, it focuses on the integration of Fast Fourier Transform (FFT) and Continuous Wavelet Transform (CWT) for extracting complementary frequency and time-frequency features from ECG signals. These representations are further processed using advanced neural network architectures, including convolutional neural networks (CNNs), residual networks (ResNet), temporal convolutional networks (TCN), and hybrid CNN-LSTM models. Additionally, stochastic pooling is explored as an effective optimization technique for improving model generalization and reducing overfitting. Literature demonstrates significant improvements in classification accuracy, often exceeding 95%. This survey provides a comparative analysis of existing approaches, identifies key challenges such as data imbalance and computational complexity, and highlights future research directions. The findings suggest that hybrid FFT-CWT-based deep learning frameworks represent a promising solution for real-time and reliable AF detection in modern healthcare systems.