Related Experiment Videos
A Multi-Frequency Self-Supervised Fusion Model for EEG-Based Dementia Classification
Jinhua Sheng1,2, Jiaqi Lin1,2, Qiao Zhang3,4,5
1School of Computer Science, Hangzhou Dianzi University, 310018 Hangzhou, Zhejiang, China.
Journal of Integrative Neuroscience
|July 30, 2026
Summary
This study introduces an attention-based model to fuse brain activity across multiple frequency bands for improved dementia diagnosis. The model achieved 93.1% accuracy in distinguishing Alzheimer's disease from frontotemporal dementia.
Area of Science:
- Neuroscience
- Biomedical Engineering
- Machine Learning
Background:
- Brain source localization is key for understanding dementia.
- Integrating multi-frequency brain activity data is challenging for dementia diagnosis.
- Current methods struggle with model interpretability and classification performance.
Purpose of the Study:
- To develop a novel model for enhanced dementia classification.
- To improve the integration of multi-frequency brain activity data.
- To enhance the interpretability of dementia diagnostic models.
Main Methods:
- An attention-based multi-frequency self-supervised fusion (AM-SSF) model was developed.
- Self-supervised encoders processed theta, alpha, beta, and gamma frequency bands.
- An attention-guided module fused band-specific representations for classification.
Main Results:
- The AM-SSF model achieved 93.1% accuracy in distinguishing Alzheimer's disease (AD) from frontotemporal dementia (FTD).
- The model significantly outperformed baseline methods like single-band self-supervised learning and average pooling fusion.
- Attention weight analysis revealed theta and beta bands were most influential in classification.
Conclusions:
- The AM-SSF model effectively enhances AD and FTD classification.
- The study provides insights into the discriminative power of specific frequency bands.
- This approach offers improved diagnostic accuracy and model interpretability for dementia.