Mayo Clinic AI spots pancreatic cancer on routine scans up to three years early

Pancreatic cancer is among the deadliest of all cancers, largely because it is usually found too late. By the time symptoms appear and a tumor is visible, the window for curative treatment has often closed. Researchers at Mayo Clinic reported in the journal Gut a validation study for an artificial-intelligence tool designed to change that grim timeline.

The model, called REDMOD (Radiomics-based Early Detection Model), analyzes routine abdominal CT scans — the kind millions of people already receive for unrelated reasons — and measures hundreds of quantitative imaging features that describe the texture and structure of tissue. Those features can capture faint biological changes that begin as cancer develops, long before a tumor is large enough for the human eye to notice. In effect, the AI is reading signals that are present but invisible in a scan a radiologist would read as normal.

The results are striking. Across nearly 2,000 CT scans — including scans from patients later diagnosed with pancreatic cancer that had originally been interpreted as normal — REDMOD identified 73% of prediagnostic cancers a median of roughly 16 months before clinical diagnosis, nearly doubling what specialists detected without the tool. The advantage was greatest far in advance: beyond two years before diagnosis, the model found nearly three times as many early cancers that would otherwise have been missed. Crucially, the study was designed to mirror real clinical practice, drawing on CT scans from multiple institutions, imaging systems and protocols.

“The greatest barrier to saving lives from pancreatic cancer has been our inability to see the disease when it is still curable,” said senior author Ajit Goenka, framing exactly why early detection matters so much for this cancer.

The path from a validation study to routine clinical use still requires prospective testing and regulatory review, and the researchers are careful not to overstate what a single study proves. But the promise is profound: because the tool works on scans people already have, it could one day flag high-risk patients opportunistically, without any new test or radiation. For a cancer where months can mean the difference between a curable and an incurable diagnosis, an extra year of warning is not a small thing — it is, potentially, a life.