AI surpasses radiologists at detecting early-stage pancreatic cancer in tests

RESEARCHERS in the US have revealed that AI has surpassed radiologists at detecting early-stage pancreatic cancer in tests.

Pancreatic cancer kills around 92% of those diagnosed in the UK within five years. That statistic exists almost entirely because detection comes too late. 

An article published in Gut, part of the BMJ on 28 April 2026 reports an AI system can now detect pancreatic cancer up to three 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 caught 39%. 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.

The UK has no population screening programme for pancreatic cancer.

The MHRA's 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.

The detection capability is real. The work now is proving it holds up in diverse populations and live clinical settings.

Colette Mason, Author & AI Consultant at London-based Clever Clogs AI, said

She added: “This is what it looks like when AI adoption is done properly. 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.”

Katrina Young, Digital Transformation Strategist at KYC Digital, said 

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. More critically, a 19% false positive rate in a cancer pathway means patients undergoing invasive follow-up for disease they do not have. 

"The issue is no longer whether AI can detect earlier. It is whether the system can absorb the consequences. NICE NG12 already defines the high-risk cohort. Layering AI into those pathways before full validation creates an accountability gap. The MHRA framework is still in development. The breakthrough is technical. The constraint is structural."