A new AI model reads EEG brainwave data to flag possible Alzheimer's disease, and its makers say it beats every other system tested so far.
The model, called LEAD, is the first foundation model built specifically for EEG-based Alzheimer's detection. Researchers assembled a dataset of 2,238 subjects, the largest EEG-Alzheimer's corpus to date, to get around the small, single-site datasets that have limited earlier deep-learning attempts. LEAD uses a "gated temporal-spatial Transformer" that can handle EEG recordings of different lengths, channel counts, and sampling rates, plus a subject-level training method meant to improve how well it generalizes across different people. The team pretrained it with medical contrastive learning on 13 datasets, four of them Alzheimer's-specific and nine covering other neurological disorders, then fine-tuned and tested it on five separate Alzheimer's datasets, where it topped the rankings across all 20 evaluations.
The bigger story here is cost and access. Alzheimer's diagnosis today leans on PET scans and spinal taps, both expensive and not widely available outside specialist centers. EEG equipment is cheap and common, so a model that reliably reads Alzheimer's signals off routine EEG data could push early screening into clinics that will never have a PET scanner.
Winning on five benchmark datasets in a paper is not the same as working in a doctor's office, and at 2,238 subjects, LEAD's training set is a rounding error next to the datasets behind most other AI foundation models - though the code is open-source on GitHub, so outside labs can now test that claim themselves.