Artificial Intelligence for Alzheimer's Disease Identification from EEG Using Deep Unfolding and Residual Attention Networks

Abstract

Alzheimer’s Disease (AD) is a progressive neurodegenerative disorder that significantly affects cognitive functions and memory. Early detection remains challenging due to subtle neurological changes in initial stages. Electroencephalography (EEG), particularly central lobe EEG signals, provides a non-invasive and cost-effective method for identifying abnormal brain activity associated with AD. Recent advancements in artificial intelligence (AI), especially deep learning, have significantly enhanced EEG-based diagnostic systems. Models such as Convolutional Neural Networks (CNN), Graph Neural Networks (GNN), and attention-based architectures have demonstrated strong capabilities in extracting complex spatial-temporal features from EEG signals. For instance, deep pyramid convolutional neural networks (DPCNN) have achieved classification accuracy exceeding 97% for AD detection by effectively capturing hierarchical EEG features. Additionally, graph-based approaches model functional connectivity between EEG channels, improving classification performance by representing brain networks more effectively. Optimization-based approaches such as deep unfolding networks further enhance model efficiency by integrating iterative optimization into deep architectures. Despite these advancements, challenges such as data scarcity, computational complexity, and limited interpretability persist. This review presents a comprehensive analysis of AI techniques for AD detection using EEG signals and highlights the potential of dynamic path-controllable deep unfolding networks and residual attention neural networks for improved diagnostic accuracy and robustness.