Researchfoundation modelsfmrineuroimagingself supervised
Omni-FMRI Introduces Atlas-Free Voxel Foundation Model
8.1
Mo Wang et al. (arXiv, submitted Jan 30, 2026) present Omni‑fMRI, an atlas-free foundation model operating on voxel-level fMRI signals and pretrained on 49,497 sessions across nine datasets. The method introduces dynamic patching to cut computational cost while preserving spatial detail, and establishes an 11-dataset benchmark showing consistent gains over prior foundation models. Code and logs are provided.
Key Points
- 1Introduces atlas-free voxel-level foundation model pretrained on 49,497 fMRI sessions across nine datasets.
- 2Applies dynamic patching to reduce compute while preserving spatial detail, enabling scalable pretraining.
- 3Improves transfer performance on 11 benchmark datasets, enabling reproducible atlas-free brain representation learning.
Scoring Rationale
High novelty and strong scalability across large fMRI datasets, limited by preprint status and lack of peer-reviewed validation.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
