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. According to the US National Cancer Institute, more than 85% of patients aren't diagnosed until the disease has already spread, and survival rates sit below 15%. By the time symptoms appear, the window for surgery that could cure it has usually closed.

Researchers at Mayo Clinic have built an AI model designed to find it much earlier, on scans doctors have already looked at and judged to be normal.

How the model works

The model is called REDMOD, short for Radiomics-based Early Detection Model. It reads routine abdominal CT scans, the kind patients often get for unrelated reasons, and measures hundreds of quantitative features of the pancreas's tissue texture and structure. Those are patterns too subtle for a radiologist to flag, because the organ still looks normal to the human eye.

"This AI can now identify the signature of cancer from a normal-appearing pancreas," said senior author Dr. Ajit Goenka, a Mayo Clinic radiologist and nuclear medicine specialist.

What the study found

In a validation study published in the journal Gut, the team tested REDMOD on nearly 2,000 CT scans, including scans from patients who were later diagnosed with pancreatic cancer but whose earlier images had been read as normal.

• REDMOD flagged 73% of those prediagnostic cancers, at a median of about 16 months before the patients were actually diagnosed.
• That was roughly double the detection rate of specialists reading the same scans without computer help.
• On scans taken more than two years before diagnosis, it found nearly three times as many early cancers as would otherwise have been missed.

The researchers deliberately tested the model on scans from multiple hospitals, imaging machines, and scanning protocols, to mirror the messiness of real clinical practice rather than a tidy lab dataset, and its performance held up. Repeat scans of the same patients taken months apart also gave consistent results, which matters for any tool meant to track people over time.

Who it's meant for

REDMOD isn't meant to screen everyone. It is aimed at people at higher risk of pancreatic cancer, such as adults who develop diabetes for the first time later in life, a known warning sign. Many of them already have abdominal CT scans on record, so the AI can look for early signs without anyone needing an extra test.

What comes next

The next step is a prospective clinical study called AI-PACED (Artificial Intelligence for Pancreatic Cancer Early Detection), funded by the US National Institutes of Health. It will test how doctors can use the AI's findings for high-risk patients in practice, and will track false positives and long-term outcomes, the questions that decide whether a promising model actually saves lives. The work is part of Mayo Clinic's Precure initiative, which focuses on spotting the earliest biological signs of disease.

The study didn't report false-positive rates, and the model has not been approved for routine clinical use. But sixteen months is the difference between catching a tumour while surgery can still remove it and finding it after it has spread. REDMOD doesn't replace a doctor's judgment. It gives doctors a second set of eyes that can see patterns they were never trained to look for.