Deep Unfolding and Residual Attention Networks for Alzheimer’s Disease Detection Using Central Lobe EEG Signals

Abstract

Alzheimer’s Disease (AD) is a progressive neurodegenerative disorder characterized by cognitive decline, memory loss, and impaired decision-making. Early detection is critical for effective intervention, yet traditional diagnostic techniques such as neuroimaging and clinical assessments are often expensive and subjective. Electroencephalography (EEG), particularly signals obtained from the central lobe, offers a non-invasive and cost-effective alternative for early diagnosis. Recent advances in Artificial Intelligence (AI), especially deep learning, have significantly enhanced EEG-based AD detection. This study reviews recent developments in AD identification using advanced architectures such as Dynamic Path-Controllable Deep Unfolding Networks and Residual Attention Neural Networks. These models enable adaptive feature extraction and improved representation learning by dynamically controlling information flow and emphasizing relevant EEG features. Deep learning approaches have demonstrated strong capability in identifying subtle neurological patterns that are not easily detectable using traditional methods. Residual attention mechanisms further improve classification performance by focusing on important brain regions and suppressing irrelevant noise, achieving high diagnostic accuracy. Despite these advancements, challenges such as limited datasets, signal noise, and model interpretability persist. This review highlights current trends, evaluates state-of-the-art techniques, and discusses future directions for developing efficient, reliable, and clinically applicable AD diagnostic systems.