A Deep Learning Approach to Alzheimer's Diagnosis Using EEG Data: Dual-Attention and Optuna-Optimized SVM

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder characterized by amyloid-β plaques and tau tangles.
• Early diagnosis and management of AD are crucial for improving patient quality and duration of life.
• Real-time processing of mobile electroencephalography (EEG) data presents challenges for rapid AD detection.

Purpose of the Study:

• To identify specific EEG frequency bands associated with Alzheimer's disease.
• To accelerate Alzheimer's disease detection methods using EEG data.
• To achieve accurate and computationally efficient classification of AD.

Main Methods:

• EEG recordings from 48 individuals (24 AD, 24 healthy controls) were analyzed.
• Data were segmented into Alpha, Beta, Delta, Gamma, and Theta frequency bands.
• MobileNetV2 architecture, Dual-Attention Mechanism, and Support Vector Machine (SVM) with Optuna hyperparameter optimization were employed.

Main Results:

• Delta and Beta frequency bands were identified as most significant for AD detection.
• Attention mechanisms improved MobileNetV2 model performance by 2%.
• SVM with optimized hyperparameters showed an approximate 3% performance increase; feature fusion enhanced detection in larger datasets.

Conclusions:

• Frequency band analysis and feature fusion show potential for enhancing accuracy and efficiency in EEG-based AD diagnosis.
• The study highlights the utility of specific frequency bands and advanced machine learning techniques.
• Results are promising but require caution regarding generalizability to broader populations.