Early Detection of Alzheimer's Disease with EEG Signals Using Vision Transformer
Authors :- D Vyawhare, H Telangore, M Sharma, SK Satapathy
Publication :- International Conference on Artificial Intelligence and Machine Vision (AIMV), IEEE, 2025.
Alzheimer’s disease (AZD) is a progressive neurological disorder characterized by cognitive deterioration and structural brain changes. Early detection is pivotal for intervention strategies that could potentially slow its progression. Electroencephalogram (EEG) signals hold promise as a diagnostic tool for AZD. However, the inherent rapid and unstructured nature of EEG data presents challenges for analysis and interpretation. Conventional machine learning models lack interpretability, which is a critical factor for clinical acceptance and trust. To address this gap, this research employs a Vision Transformer (ViT) model, a novel architecture that fundamentally differs from traditional convolutional neural networks (CNNs). A primary focus of this study is not only on achieving high classification accuracy but also on enhancing interpretability. Notably, the ViT model achieved a remarkable accuracy of 99.99% in the best-performing channel O2 demonstrating its efficacy in identifying distinctive EEG patterns associated with AZD. By offering insights into the ViT model’s decision-making process, we aim to provide clinicians with a transparent understanding of how and why specific conclusions regarding AZD classification are reached. This transparency aims to foster credibility and trust in the ViT model’s clinical application, crucial for its adoption in real-world scenarios for early AZD detection and intervention.