An AI Model Can Spot Pancreatic Cancer Three Years Before Doctors Can

Pancreatic cancer is one of the deadliest cancers precisely because it stays quiet. More than 85% of patients aren't diagnosed until it has already spread, and five-year survival sits below 15%. By the time symptoms show up, the window for curative treatment has usually closed.

Mayo Clinic built an AI model called REDMOD to catch it earlier, by reading routine abdominal CT scans for hundreds of subtle texture and structure features that a radiologist wouldn't flag as unusual. In a validation study published in the journal Gut, researchers ran it against nearly 2,000 CT scans, including ones from patients who were later diagnosed with pancreatic cancer but whose scans had originally been read as normal.

REDMOD flagged 73% of those prediagnostic cancers, at a median of about 16 months before the actual diagnosis. In scans taken more than two years before diagnosis, it caught close to three times as many early cancers as would otherwise have gone unnoticed. The study used data from multiple institutions, imaging systems, and scan protocols, deliberately built to mirror real clinical practice rather than a clean lab setup.

Sixteen months doesn't sound dramatic until you consider what it buys: the difference between catching a tumor while surgery can still remove it, and catching it after it has already spread. None of this replaces a doctor's judgment, but it gives doctors something they didn't have before, a second set of eyes that can see patterns humans aren't trained to look for.