OnCo
ideasIdea

Digitise the nation's pathology slides and link them to outcomes

Scan the millions of cancer slides already sitting in hospital basements and connect each to what happened to the patient, creating the world's largest training set for pathology AI.

Pathology departments hold decades of glass slides with matched registry outcomes. Digitising them at scale (tens of millions of slides) and linking to registry survival, treatment and recurrence data would create a public resource that dwarfs any commercial pathology AI training set. Precedents include the PathLAKE and iCAIRD centres in the UK, the TCGA image collection, and the NHS digital pathology programme. Cost is dominated by scanning and storage, both of which have fallen sharply.

Hypothesis
A linked archive of more than 10 million slides with outcomes, released under governed access, will produce prognostic and predictive pathology models that outperform current commercial models on external validation within three years of release.
Rationale
Foundation models scale with data; the best current pathology models were trained on around one to three million slides from single institutions. Public linked data would also allow independent validation, which is what commercial models lack.
What would test it
Fund one region to scan 500,000 archival slides linked to registry outcomes; release under a TRE; run an open benchmark against existing models for five-year survival prediction in three cancers.
Maturity
early clinical
Who has to act
philanthropy
Cost to try
Large (over $50M)
Years to first evidence
4
Bottlenecks it attacks
  • Data silos · Records, scans, genomes and outcomes sit in separate systems that cannot talk. Every patient's experience is lost to the next.
  • 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.

Connected

7top

Pages like this

not linked directly; found by shared links

fronts

1
AI & Computation

technologies

3
Digital pathology & AIHistopathology & immunohistochemistryPathology & radiology foundation models

collections

1
TCGA / NCI Genomic Data Commons

bottlenecks

2
AI that is built but not validated or deployedData silos