RESEARCHERS in the US have claimed that AI has surpassed radiologists at detecting early-stage pancreatic cancer in tests, with experts saying “this is what it looks like when AI adoption is done properly”.
An article published in Gut, part of the British Medical Journal (BMJ), claims that an AI system can now detect pancreatic cancer years before any human can see it on a scan.
Mayo Clinic researchers have published a validated AI model called Radiomics-based Early Detection Model (REDMOD) that identified pancreatic cancer on routine CT scans with 73% sensitivity, at a median of 475 days before clinical diagnosis.
Radiologists reviewing the same scans identified the cancer with 39% sensitivity. For scans taken more than two years before diagnosis, the AI was nearly three times more accurate. The study used nearly 1,500 scans across multiple hospitals.
Pancreatic cancer kills around 92% of those diagnosed in the UK within five years but there is no population screening programme for the disease.
This is what it looks like when AI adoption is done properly
The Medicines and Healthcare products Regulatory Agency’s (MHRA) dedicated AI medical device framework is still being written, with publication confirmed for 2026, but not yet delivered.
The study’s own authors acknowledge REDMOD has not yet been tested prospectively or across ethnically diverse populations, which will cause a delay.
It also needs further research for high-risk patients, including those with unexplained weight loss and newly-diagnosed diabetes, before it can be used widely in clinical settings.
Colette Mason, Author & AI Consultant at London-based Clever Clogs AI, welcomed the news: “This is what it looks like when AI adoption is done properly. This is not an ‘AI Booster’ press release about productivity gains or a demo that falls apart in production.
“This is a peer-reviewed clinical study, validated across multiple hospitals, tested head-to-head against the professionals it’s designed to support, with the authors themselves being clear and publishing what it can’t do yet, rather than hype.
“REDMOD needs prospective trials in diverse populations before it reaches patients. That takes time, and for a disease with 8% five-year survival, time is the one thing people don’t have.
“But the alternative, rushing an unvalidated tool into screening pathways because the headlines are exciting, is how you get AI systems that erode the clinical trust they depend on.
“Every organisation deploying AI should be looking at this study and asking why their own adoption process has fewer safeguards than a cancer screening tool that hasn’t even reached patients yet.”
Earlier detection does not equal earlier treatment
Katrina Young, Digital Transformation Strategist at KYC Digital, said the tests are still early and need more work.
She added: “Pancreatic cancer kills over 90% of patients within five years, largely due to late detection. REDMOD changes that on paper. A model identifying cancer around 475 days earlier, with 73% sensitivity versus 39% for radiologists, is not marginal progress.
“But earlier detection does not equal earlier treatment. REDMOD has not yet been prospectively validated or tested across diverse populations.
“The issue is no longer whether AI can detect earlier. It is whether the system can absorb the consequences. Layering AI into those pathways before full validation creates an accountability gap. The MHRA framework is still in development. The breakthrough is technical but the constraint is structural.”
