EEGDecoder-x: an explainable deep learning framework for cross-subject EEG-based detection of Alzheimer's and Creutzfeldt–Jakob disease - Summary - MDSpire
Advertisement
EEGDecoder-x: A Transparent Deep Learning Approach for Cross-Subject Detection of Alzheimer's Disease and Creutzfeldt–Jakob Disease Using EEG Data
To develop an interpretable framework for decoding EEG signals to distinguish between Alzheimer's disease, Creutzfeldt–Jakob disease, and healthy controls.
Approach:
EEGDecoder-Net: A hybrid attention network combining convolutional neural networks with dual attention mechanisms for efficient spatio-temporal feature extraction.
EEGDecoder-XAI: An explainability module providing local and global insights into the model's learning process.
Key Findings:
Achieved 97.22% classification accuracy using a Leave-One-Subject-Out evaluation on a dataset of 36 subjects (12 with AD, 12 with CJD, and 12 healthy controls).
Outperformed baseline models, demonstrating both effectiveness and interpretability.
Interpretation:
Limitations:
The study relies on a small dataset of 36 subjects, which may limit generalizability.
Potential challenges in real-world clinical settings due to EEG signal contamination, which may affect model performance.
Conclusion:
EEGDecoder-x presents a novel approach for EEG-based classification of neurodegenerative diseases, balancing robustness, evaluation reliability, and interpretability.
