YiheWang/LEAD

This is the pre-trained checkpoint for the paper LEAD: An EEG Foundation Model for Alzheimer's Disease Detection (Accepted by Transactions on Machine Learning Research 2026).

Authors: Yihe Wang, Nan Huang, Nadia Mammone, Marco Cecchi, Xiang Zhang

Paper link: TMLR, Arxiv

We train the world's first EEG foundation model for Alzheimer's Disease (AD) detection, named LEAD, by leveraging the largest EEG-based AD detection dataset to date, which includes 2,238 subjects and 427.81 hours of EEG recordings. The LEAD model is adaptable to EEG data of any length, channel topologies, and diverse sampling rates, enabling flexible end-to-end subject-level AD detection. The motivation for this work is to provide a comprehensive work for EEG-based AD detection, including existing challenges, EEG-AD datasets, robust model design, a training pipeline, post-hoc processing, correct evaluation protocols, current progress, and future directions.