SophontAI Releases CortexMAE FMRI Foundation Models And Brainmarks Benchmark
A big focus at Sophont is building foundation models to understand & analyze the brain. This is still a nascent field but we believe such models could eventually help improve diagnosis & treatment of neurodegenerative diseases and mental disorders.
This project is some of our initial efforts in the space. We have trained a family of foundation models on functional MRI (fMRI) neuroimaging data that achieves SOTA performance on a variety of benchmarks. We do so by introducing a novel approach for representing fMRI data called flat maps.
Of course, since it's still an emerging area, there is a lack of systematic benchmarks, which we attempt to fix with our Brainmarks benchmarking suite.
All code and models are completely open-sourced!
As usual, this project was done with the support of our broader @MedARC_AI community. If you're interested in contributing to this line of research, please join the MedARC discord! We are now building multimodal MRI foundation models and participating in the FOMO challenge.
If you are interested in partnering on neuro foundation models, be sure to contact me directly as well!
Sophont@SophontAINEW RELEASE: Today we're releasing CortexMAE: a family of fMRI foundation models trained on 2.1K hours of open fMRI data. We're also releasing Brainmarks: an open benchmark suite for evaluating fMRI foundation models. Full paper is on arXiv (accepted to ICML 2026) A thread:
A big focus at Sophont is building foundation models to understand & analyze the brain. This is still a nascent field but we believe such models could eventually help improve diagnosis & treatment of neurodegenerative diseases and mental disorders.
This project is some of our initial efforts in the space. We have trained a family of foundation models on functional MRI (fMRI) neuroimaging data that achieves SOTA performance on a variety of benchmarks. We do so by introducing a novel approach for representing fMRI data called flat maps.
Of course, since it's still an emerging area, there is a lack of systematic benchmarks, which we attempt to fix with our Brainmarks benchmarking suite.
All code and models are completely open-sourced!
As usual, this project was done with the support of our broader @MedARC_AI community. If you're interested in contributing to this line of research, please join the MedARC discord! We are now building multimodal MRI foundation models and participating in the FOMO challenge.
If you are interested in partnering on neuro foundation models, be sure to contact me directly as well!
Sophont@SophontAINEW RELEASE: Today we're releasing CortexMAE: a family of fMRI foundation models trained on 2.1K hours of open fMRI data. We're also releasing Brainmarks: an open benchmark suite for evaluating fMRI foundation models. Full paper is on arXiv (accepted to ICML 2026) A thread: