NeuroFormer: A Deep Learning Framework for Alzheimer's Detection Using EEG Signals.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Diagnostics

Background:

• Alzheimer's disease (AD) and Frontotemporal dementia (FTD) are neurodegenerative disorders with overlapping symptoms, complicating diagnosis.
• Accurate and timely diagnosis is critical for effective dementia treatment and intervention planning.
• Current diagnostic methods face challenges due to symptom heterogeneity and complex pathological mechanisms.

Purpose of the Study:

• To develop a novel deep learning architecture, NeuroFormer, for classifying EEG signals.
• To differentiate between Alzheimer's disease (AD), Frontotemporal dementia (FTD), and healthy controls (CN) using non-invasive neurophysiological measurements.
• To establish a robust diagnostic framework for early dementia detection.

Main Methods:

• Proposed NeuroFormer, a deep learning model integrating spectral processing, neural dynamics, capsule routing, and attention mechanisms.
• Utilized EEG signals captured via non-invasive neurophysiological instruments.
• Employed dynamic spectral gating, graph convolutions, and multi-scale adaptive feature pyramids for spatiotemporal feature extraction.

Main Results:

• NeuroFormer achieved a classification accuracy of 95.76% for AD, FTD, and CN.
• The model significantly outperformed benchmark models across all key evaluation metrics.
• Demonstrated effective extraction of discriminative spatiotemporal features from raw EEG data.

Conclusions:

• NeuroFormer shows significant potential as a measurement-driven diagnostic framework for dementia.
• The model enables early and accurate classification of dementia-related disorders using EEG instrumentation.
• This approach offers a non-invasive method for distinguishing between AD, FTD, and healthy individuals.