Pancreatic cancer kills more than 85 percent of those diagnosed, largely because it is almost always caught too late.

By the time symptoms appear, the disease has usually spread beyond the reach of surgery.

As the researchers behind a promising new study put it, “the timing of diagnosis is the single most critical determinant of survival outcomes.”

That timing may finally be shifting.

Researchers at the Mayo Clinic, working with collaborators at MD Anderson and the University of Washington, have developed an artificial intelligence framework called REDMOD that can identify pancreatic cancer on routine CT scans more than a year before a patient is diagnosed.

The findings, published in Gut, mark a meaningful step toward catching the disease while it is still curable.

How AI Detects Pancreatic Cancer Before Tumors Appear

The challenge has always been visibility.

In its earliest stage, pancreatic cancer leaves no visible mass on a CT scan—even expert radiologists cannot see it.

But the tissue is changing in subtle ways at the microscopic level, long before any tumor takes shape.

REDMOD was trained to read these invisible patterns, called radiomic signatures, by analyzing nearly 1,500 CT scans from patients who later developed pancreatic cancer alongside healthy controls.

The results were striking. On an independent test set, REDMOD detected pancreatic cancer with 73 percent sensitivity at a median of 475 days—roughly 16 months—before clinical diagnosis.

That was nearly double the accuracy of board-certified radiologists reviewing the same scans.

At lead times beyond two years, the AI’s advantage grew to almost threefold, identifying 68 percent of cases that radiologists caught only 23 percent of the time.

Just as important, the model proved consistent.

It maintained 90 to 92 percent agreement when re-tested on follow-up scans, and held its accuracy across different hospitals, scanner brands, and image qualities—practical hurdles that have stalled earlier AI tools.

The Future of Pancreatic Cancer Diagnosis

The researchers envision REDMOD eventually serving as a screening aid for high-risk patients, particularly older adults newly diagnosed with diabetes, a known early warning sign.

Prospective trials are now underway.

The findings offer a glimpse of a future in which one of medicine’s deadliest cancers might be caught before it is too late.

_______________________________________________

Adapted from: Mukherjee S, et al. “Next-generation AI for visually occult pancreatic cancer detection in a low-prevalence setting with longitudinal stability and multi-institutional generalisability.” Gut. 2026. doi:10.1136/gutjnl-2025-337266