‹ back to 2026-08-13

GenFAR: A generalized representation of brain structure, derived from 49,246 multi-cohort MRIs via deep learning

arXiv cs.CV Medical/Healthcare AI Vishnu M. Bashyam, Guray Erus, Junhao Wen, Pratik Chaudhari, Randa Melhem, Sindhuja Govindarajan Tirumalai, Gareth Harman, Yong Fan, Colin L. Masters, Paul Maruff, Sterling C. Johnson, Jurgen Fripp, Duygu Tosun, John C. Morris, Daniel S. Marcus, Pamela LaMontagne, Tammie Benzinger, Susan R. Heckbert, Mark Espeland, Marilyn S. Albert, Andrew J. Saykin, Paul M. Thompson, Timothy J. Hohman, Susan M. Resnick, R. Nick Bryan, Murat Bilgel, Yang An, David A. Wolk, Li Shen, Haochang Shou, Ilya M. Nasrallah, Christos Davatzikos 2026-08-12

TL;DR - GenFAR is a modular brain-MRI representation framework trained on 49,246 participants across 11 cohorts and 17 clinical and biological tasks. Its reusable features improve accuracy and sample efficiency when training specialized downstream models.

Sequential learning lets tasks progressively build on previously learned MRI representations.
Testing 5,000 task sequences identified six tasks as the optimal sequence length.
A new Donor Score highlighted Age, AD/MCI, MMSE, Hypertension, and Hyperlipidemia as strong contributors to downstream performance.
The learned representation generalized to tasks outside training and supported more data-efficient secondary predictors.

view merged work →