Mayo Clinic researchers say an AI system called REDMOD can spot early pancreatic cancer signals on routine abdominal CT scans long before tumors are obvious to radiologists, according to a new validation study published in Gut.

What the study found

Mayo said REDMOD reviewed nearly 2,000 scans that had originally been read as normal, including scans from patients who were later diagnosed with pancreatic cancer. The model identified 73% of prediagnostic cancers at a median of about 16 months before clinical diagnosis. In the same comparison, radiologists reviewing the scans without AI detected 39%.

A BMJ summary of the paper said the gap widened on scans taken more than two years before diagnosis, where REDMOD detected 68% of cases versus 23% for radiologists. The study also reported stable results across repeat scans and external datasets, suggesting the model may generalize beyond a single hospital workflow.

Why it matters

Pancreatic cancer is often found late, when treatment options are limited. If systems like REDMOD hold up in real clinical use, they could turn already-collected CT scans into an earlier warning layer for high-risk patients.

That said, Mayo and BMJ both frame this as a validation milestone, not a finished screening product. Mayo has already started a prospective follow-up study, AI-PACED, to test whether the tool improves care decisions without creating too many false alarms.