Sunnybrook Research Institute:where AI in cancer imaging gets industrialized

Academic health sciences centre with major research institute~10,000 staffTorontoClinical, research, oncology

What they're doing

Sunnybrook Research Institute (SRI) runs one of the deepest AI imaging programs in Canadian oncology. Dr. Anne Martel, Senior Scientist and Tory Family Chair in Oncology at SRI, leads a lab that has spent more than a decade building machine learning models for digital pathology and radiology. Her group develops AI to detect occult invasive cancers in breast tissue, predict recurrence-free survival from histopathology images, and classify cancerous regions in microscopy data with enough precision to compress pathology turnaround times.

The infrastructure underneath the work is a $485,676 Canada Foundation for Innovation grant that established the Artificial Intelligence Platform for Precision Medicine (AIPPM) at SRI, a computing system built for imaging-scale datasets. Sunnybrook also leads the Industry Consortium for Image-Guided Therapy, a pan-Canadian network backed by a $49M federal investment that integrates AI into image-guided therapy.

Why it matters

Most hospital AI programs are still trying to get one model into production. Sunnybrook has been running a portfolio of imaging AI projects for years, and it has the infrastructure (AIPPM, image-guided therapy network) and the people (Martel and a deep bench) to keep producing them. The lesson is about industrialization, not invention. Sunnybrook treats AI in oncology as a permanent research-and-deployment function, not a one-off project.

What's working

A named senior scientist (Martel) with a decade-plus track record and an endowed chair that anchors the program
Federal funding stacked deliberately: CFI for the platform, $49M for the consortium, CIHR for individual model studies
A national consortium (Industry Consortium for Image-Guided Therapy) that pulls in industry partners and other Canadian hospitals
Cross-modal integration (pathology plus radiology plus clinical data plus genomics) that mirrors how oncologists actually make decisions
A pipeline of trainees who earn top finishes in international AI pathology challenges (SemiCOL), which feeds the next generation of the field

What's harder than it looks

Imaging AI has a translation gap. Models that perform well in research datasets often degrade on real clinical workflows, and Sunnybrook publishes the methodology but not always the failed pilots
The pathology AI story is moving toward foundation models trained on millions of slides. Hospital-scale labs like Martel's compete with industry labs that have an order of magnitude more compute and data
Image-guided therapy commercialization runs through industry partners. The science stays with Sunnybrook, but the deployed product does not always carry the hospital's name into clinical practice

Three takeaways for similar orgs

01Treat AI in your specialty as a permanent function, not a project. Anchor it with a chair, an endowed lead, or a budget line that does not depend on the next grant
02Pair the AI program with a CFI-grade infrastructure investment. The compute platform is what makes the second, third, and tenth model possible
03Multi-modal data integration is where the next decade of clinical AI value lives. Single-modality models will increasingly be commodities