ideasIdea

Continuous prospective validation for every oncology AI tool after deployment

Cancer AI tools are approved on old test data and then never checked again. Require every deployed tool to report its real-world performance continuously, in public.

Radiology, pathology and prognostic AI tools in oncology are cleared on retrospective datasets; performance drifts with scanners, populations and practice, and post-market surveillance is minimal. The proposal is a regulatory requirement and shared infrastructure: every deployed oncology AI tool feeds outcome-linked performance metrics to a registry, stratified by site and demographic group, with public dashboards, drift alerts and pre-agreed thresholds for suspension, harmonised across regulators.

Hypothesis

Continuous validation detects clinically significant performance degradation in a meaningful share of deployed tools within two years and raises clinician trust and adoption of tools that perform well.

Rationale

Pharmacovigilance is standard for drugs; algorithms change performance more readily and silently, and a shared registry spreads the cost.

What would test it

Pilot registry across three tool categories in ten hospitals; measure detection of drift, time to corrective action, and comparison of registry performance with cleared claims.

Maturity

early clinical

Who has to act

regulator

Cost to try

Medium ($1M to $50M)

Years to first evidence

Bottlenecks it attacks

AI that is built but not validated or deployed · Thousands of cancer AI models are published; a handful are in clinical use, and fewer have shown they help patients.
Regulatory divergence between regions · Regulatory divergence means a drug approved in one country can take years to reach another, or never arrive.

Connected

5top

Pages like this

not linked directly; found by shared links

technologies

AI in radiology Digital pathology & AI Pathology & radiology foundation models

bottlenecks

AI that is built but not validated or deployed Regulatory divergence between regions