Pancreatic cancer has long been called a silent killer because it rarely produces symptoms until it has already spread. A new validation study from Mayo Clinic, published in the journal Gut, suggests artificial intelligence may finally give doctors a head start. Researchers there developed a tool called REDMOD, short for Radiomics-based Early Detection Model, and used it to comb through nearly 2,000 abdominal CT scans that had originally been read as normal by radiologists, including scans from patients who were later diagnosed with pancreatic cancer.
What the study found
According to Mayo Clinic News Network, REDMOD identified 73% of those prediagnostic cancers at a median of roughly 16 months before the patients were formally diagnosed. In scans taken more than two years before diagnosis, the AI flagged nearly three times as many early-stage cancers as would otherwise have gone undetected by standard radiology review. The model works by picking up on faint textural and structural changes in pancreatic tissue that are invisible or easily dismissed by the human eye, but which reliably precede tumor formation.
Why early detection matters so much here
More than 85% of pancreatic cancer patients are diagnosed only after the disease has already spread beyond the pancreas, according to Mayo Clinic, and five-year survival rates remain below 15%, among the worst of any major cancer. Because curative surgery is only an option when a tumor is caught early and localized, even a modest shift toward earlier detection could meaningfully change outcomes for a disease that has seen little survival improvement over decades.
How the research came together
The REDMOD project is described as part of a multiyear Mayo Clinic effort focused specifically on earlier pancreatic cancer detection, building on the institution’s existing radiomics and imaging-analytics research programs. Rather than training the model on a narrow set of confirmed tumors, researchers deliberately tested it against scans that radiologists had already called normal, a more rigorous real-world test than validating against known-positive images alone.
Limitations and open questions
Even with encouraging numbers, the study leaves real gaps before REDMOD reaches clinics. A 73% detection rate still means roughly a quarter of prediagnostic cancers were missed, and the analysis was retrospective, looking backward at scans that had already been collected rather than flagging cancers in real time in a live clinical workflow. Questions also remain about false-positive rates in the broader population, since flagging too many suspicious but benign findings could trigger unnecessary biopsies and patient anxiety. Mayo Clinic researchers have described the work as a milestone rather than a finished product, with prospective clinical validation still needed. Some radiologists have also pointed out that a retrospective study, by design, only tests the model against scans that already exist in a database, which is a different and easier task than flagging a genuinely new scan the moment it is taken, before anyone knows the eventual outcome.
How this fits the broader AI-oncology moment
Mayo’s finding lands amid a wave of similar imaging-AI results across oncology. Radiology as a specialty has become the single largest category of FDA-cleared AI tools, accounting for the majority of the more than 1,500 AI algorithms the agency has cleared to date, according to industry tracking of FDA data, with the agency now clearing roughly 30 new AI applications a month across specialties. That backdrop matters for a tool like REDMOD: it is entering a regulatory and clinical environment where AI-assisted image review has moved from a novelty into a routine, if still closely scrutinized, part of radiology practice, giving Mayo a clearer pathway toward eventual clinical deployment than an entirely first-of-its-kind tool would have faced a few years ago.
What’s next
The natural next step is testing REDMOD prospectively, screening new CT scans as they are taken rather than reviewing old ones, to see whether the same early-warning signal holds up in real time and whether it changes what doctors actually do. If it does, the tool could eventually be layered onto the millions of abdominal CT scans already performed each year for unrelated reasons, effectively turning routine imaging into an opportunistic cancer screen. For a cancer where a diagnosis today often means a devastating prognosis, even a 16-month head start could be the difference between a resectable tumor and a terminal one.
