OnCo
ideasIdea

A federated learning consortium of cancer centres that jointly own the models

Hospitals could train shared AI models on all their patients' scans and records without any data leaving the building, and jointly own the results, if someone built and governed the network.

A consortium of cancer centres and trial groups operating a federated learning infrastructure (models travel, data stay) for pathology, radiology and multimodal outcome prediction, with a governance agreement that gives members joint ownership of trained models, shared validation protocols and a route to regulatory submission. Owkin's federated networks and the EU's several federated health data projects show technical feasibility; the missing piece is a durable, member-owned institution with IP terms that reward data contribution rather than data extraction. Prospective validation of consortium models in member trials closes the loop with the AI validation bottleneck.

Hypothesis
A member-owned federated consortium of twenty centres trains models that outperform single-centre models on held-out external validation for at least three clinical tasks within three years, and at least one consortium model enters a prospective clinical trial.
Rationale
Federated training has matched centralised training in published pathology and radiology tasks; the barrier to scale is governance and IP, not algorithms. Member ownership addresses the concern that centres give away data value to vendors.
What would test it
Constitute the consortium with a governance charter, run three federated training tasks against single-centre baselines and pre-register external validation.
Maturity
early clinical
Who has to act
engineering
Cost to try
Medium ($1M to $50M)
Years to first evidence
3
Bottlenecks it attacks

Connected

9top

Pages like this

not linked directly; found by shared links