Enhanced Alzheimer's detection with EEG source imaging and multi-branch joint attention

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Diagnostics

Background:

• Alzheimer's disease (AD) diagnosis can be challenging with traditional methods.
• Current electroencephalogram (EEG) techniques for AD detection often overlook crucial spatial and frequency band correlations.
• Suboptimal cross-patient performance is a limitation in existing EEG-based AD detection.

Purpose of the Study:

• To develop an advanced deep learning model for more accurate AD detection using EEG.
• To improve the analysis of spatial information and inter-frequency band correlations in EEG signals.
• To enhance the cross-patient diagnostic capabilities for Alzheimer's disease and related dementias.

Main Methods:

• Proposed a multi-branch joint attention network (MJANet) utilizing electrophysiological source imaging (ESI) for enhanced spatial resolution.
• Implemented a multi-branch joint attention (MBJA) mechanism to analyze interactions across different EEG frequency bands.
• Employed a moving shifted window technique for capturing global image features within EEG data.

Main Results:

• Achieved 85.23% accuracy in differentiating Alzheimer's disease (AD) from normal controls (NC), an 8.03% improvement over state-of-the-art methods.
• Demonstrated 75.57% accuracy in distinguishing frontotemporal dementia (FTD) from NC.
• Attained 63.97% accuracy for classifying AD, NC, and FTD simultaneously.
• Utilized GradCAM for visualizing the attention mechanism, offering interpretability.

Conclusions:

• The MJANet model shows significant potential for improving the accuracy of diagnosing neurodegenerative diseases like AD and FTD.
• This approach offers a novel biomarker for enhanced clinical diagnostic methods.
• The findings suggest a pathway towards more precise and reliable diagnostic tools for brain disorders.