Ontario Brain Institute and NeuroFL:federated learning protects privacy without losing research utility

What they're doing

The Ontario Brain Institute (OBI) runs two pieces of infrastructure that fit together. Brain-CODE is the older one, a secure neuroinformatics platform built on more than a decade of provincial investment. It pools clinical, imaging, genomic, and behavioural data from research programs in neurodegeneration, epilepsy, mood disorders, neurodevelopmental disorders, and cerebral palsy. Its governance framework lets more than 40 Canadian research institutions and hundreds of researchers work inside one shared platform, and the platform itself is peer-reviewed and designed for federation, not just storage.

NeuroFL is the newer piece. It is OBI's federated learning platform, built to let researchers train models on sensitive brain data without moving or centralizing it. In March 2026, OBI committed $640,000 across eight Centre for Analytics projects spanning neurodevelopment, concussion, epilepsy, neurodegeneration, and mental health. Three of those projects run on NeuroFL directly. The flagship is Dr. Azadeh Kushki's work at Holland Bloorview Kids Rehabilitation Hospital and the University of Toronto, a multi-modal federated learning pipeline for personalized care of children with brain-based differences. In April 2026, OBI took the NeuroFL approach to the Flower AI Summit in London, presented by President and COO Bryce Pickard and a machine learning engineer from OBI's informatics team.

Why it matters

Privacy is the throttle on Canadian health AI. Most useful brain data sits inside hospitals that cannot legally pool it, and most of the productive AI work needs more data than any one site holds. Federated learning is the way through: the model travels to the data instead of the data travelling to the model. OBI is betting this is the architecture that lets Canada actually use the brain data it already has, and the bet only works because OBI spent more than a decade building the consent, governance, and partner relationships that federated learning quietly depends on. The hard part was never the algorithm. It was the social infrastructure underneath it.

What's working

• ▸More than a decade of data-sharing governance through Brain-CODE, now spanning 40+ research institutions and hundreds of researchers. That is the part no one else can shortcut
• ▸A concrete first cohort: $640,000 across eight Centre for Analytics projects, three of them running on NeuroFL, each with a named clinical lead and a specific condition
• ▸Public, conference-grade presentation of the methodology at the Flower AI Summit in London, April 2026, which pulls international peer scrutiny in early instead of late
• ▸A pairing with Holland Bloorview that anchors the first NeuroFL deployment in pediatric neurodevelopment, a population where data centralization is hardest and the privacy stakes are highest
• ▸Sustained provincial backing, with OBI's CEO framing the work as converting years of provincial data collection into a discovery engine that protects patient privacy and intellectual property

Three takeaways for similar orgs

1. 01If you serve a sector where data centralization is legally or ethically blocked, federated learning is not optional. Map the governance before the technology, because the consent and the contracts are the slow part
2. 02Provincial or sector-wide infrastructure pays back over decades, not quarters. NeuroFL is only possible because Brain-CODE was built first. Plan the long arc and protect the budget line that funds it
3. 03Pair the platform with a named clinical first user. Kushki and Holland Bloorview give NeuroFL a real-world test, a population whose families care about the answer, and a reason for the next set of projects to follow

Action plan: the next two weeks

Three takeaways are easy to nod at and hard to start. Here is what a research institute, hospital, or data-rich mission-driven org could actually move on in the next two weeks.

1. 01Make the list. Write down every dataset you cannot legally pool today, and name the specific legal or consent reason beside each one. That list is your federated-learning roadmap, in priority order
2. 02Book one 30-minute call with whoever owns your data-sharing agreements, usually a privacy officer or legal lead. Ask a single question: what would it take to let a model run against our data without the data leaving our walls? Their answer scopes the whole project
3. 03Name your first user. Find one clinical or program lead willing to be the first real user, and one specific question their data could answer. Federated infrastructure with no first user is a science project, not a program
4. 04Look before you build. Find the partner org that already has the governance you lack, OBI being one example, and ask to plug in rather than stand up your own platform. Most orgs should be joining a pool, not building one
5. 05Write the privacy promise. In plain language, write down the promise you can make to the people whose data this is. If you cannot write that sentence, you are not ready to deploy, and that is the most useful thing to learn in two weeks