Studying brain activity with fMRI is limited by data access; large existing datasets require agreements and mostly capture rest states. Open-fMIND provides a uniformly preprocessed, openly available corpus of fMRI timeseries from 11,379 subjects across hundreds of studies, including diverse tasks. Pretraining on this corpus achieves performance comparable to using restricted data, enabling wider access for brain model development.
Open brain activity dataset rivals restricted UK Biobank for AI pretraining
The Open-fMIND corpus of 2.4 million timeseries from 322 datasets yields downstream performance comparable to models trained on UK Biobank
Research Lab
Carlo Alberto Barbano · Giovanni Marraffini · Demian Wassermann
Université Paris-Saclay · Inria · CEA
Research Digest··2 min read

In one line
Open-fMIND is an open-access corpus of 2.4 million fMRI timeseries from 11,379 subjects, enabling pretraining of brain foundation models.
Before you read this
An idea this paper builds onRead it as
The authors introduce Open-fMIND, a uniformly preprocessed fMRI corpus from 322 OpenNeuro studies comprising 11,379 subjects and 45,731 functional runs, intended for self-supervised brain representation learning.
The authors collected 322 openly licensed fMRI datasets from OpenNeuro, preprocessed them uniformly with fMRIPrep, and parcellated timeseries under 52 atlases from 14 families.
How it works
stage by stageThe Daily Edition
The front page in your inbox each weekday morning: the lead, the picks, and why each one earned its place.