NPJ Digital MedicineNov 2, 2026, 19:34 GMT

AI model outperforms others in human brain scan diagnosis

Across nine diagnostic tasks, the largest model averaged 0.751 on a score measuring how well it separated diagnostic groups, outperforming comparison models.

Paper by Kangfu Han, Dan Hu, Ya Wang, +4Gang Li · University of North Carolina at Chapel Hill

Our read

AI & data

Studied:: Computational study of 22,911 people aged 0 to 100, against comparison models
What changed:: FedFound averaged 0.751 on a score measuring how well it separated diagnostic groups, outperforming comparison models
The catch:: Tests were computational classification; clinical decisions and patient outcomes were not measured

Editor of record: Boris Djordjevic · AgingResearch.News

Paper

In a computational study, researchers developed and tested FedFound using structural brain scans from 22,911 people aged 0 to 100. They drew on more than 20 datasets. The artificial intelligence model first learned patterns by reconstructing missing portions of maps showing structural similarities between brain regions. It then used federated learning, which coordinates model training across separate datasets, to learn disorder-specific patterns.

FedFound outperformed comparison models across nine diagnostic tasks covering neurodevelopmental, psychiatric and neurodegenerative disorders. Analyses of its predictions highlighted both shared and disorder-specific structural patterns. The model also preserved age-related structural characteristics when reconstructing the maps. Tests on separate datasets favored FedFound over competing models, supporting the authors’ conclusion that it could complement clinical expertise.

Why it matters

For age-related neurodegenerative diseases, the work addresses how brain imaging might support diagnosis while accounting for structural variation across the lifespan.

Caveats

These were computational classification tests, not a test of whether using the model improved clinical decisions or patient outcomes. The highlighted brain patterns do not establish causes of disease.