Deep Learning Approaches for Alzheimer’s Disease Detection from EEG Using Attention and Deep Unfolding Networks

Abstract

Alzheimer’s disease (AD) is a progressive neurodegenerative disorder characterized by cognitive decline and memory impairment. Early detection is critical for effective intervention and slowing disease progression. Electroencephalography (EEG) has emerged as a promising non-invasive modality for AD diagnosis due to its ability to capture neural dynamics with high temporal resolution. EEG signals from AD patients typically exhibit reduced synchronization, slowing of rhythms, and decreased signal complexity, making them suitable for automated analysis. Recent advancements in artificial intelligence, particularly deep learning and optimization techniques, have significantly improved EEG-based AD detection. Convolutional neural networks (CNNs), recurrent neural networks (RNNs), and graph-based models have demonstrated strong performance in capturing spatial and temporal dependencies. Furthermore, emerging approaches such as residual attention neural networks and graph convolutional networks enhance feature representation by focusing on relevant brain regions and inter-channel relationships. Dynamic path-controllable deep unfolding networks represent a novel direction by integrating optimization algorithms with deep learning, enabling adaptive feature learning and improved convergence. Studies have shown that combining deep learning with advanced signal processing and optimization techniques leads to improved classification accuracy and robustness. This review analyses recent developments, compares methodologies, and highlights challenges such as noise, data scarcity, and computational complexity, providing future directions for efficient AD diagnostic systems.