🍿 An important milestone for the lab!🍿
Today, we are sharing our work on NeuroFM, a foundation model for individualised brain health estimation from structural MRI, on medRxiv and releasing the model to the community. Particular thanks to Austin, who led much of the day-to-day work behind this project! Paper: link.
Over the past few years, the team worked to test whether structural MRI can be organised into stable, reusable patterns of brain health across datasets and domains. We evaluated this idea across five neuroscience domains using approximately 136k MRI volumes from multiple external cohorts.
More broadly, this project reflects a shift toward population-level representations that support individual-level inference from neuroimaging. Our hope is that tools like this will help researchers and clinicians move toward more standardised, quantitative, and scalable approaches to brain-health assessment from a single scan.
Code, model, and more results are available on the project website: https://rocknroll87q.github.io/NeuroFM/
Austin’s LinkedIn post: