Pancreatic cancer has long been one of medicine’s toughest killers, often hiding until it is too late to treat effectively. Now, a new AI tool may offer a way to flip that script – by detecting subtle warning signs years before a tumor appears.
The model, developed by researchers at Mayo Clinic, was tested on routine abdominal CT scans from patients who had originally been scanned for other reasons. Some of those patients were later diagnosed with pancreatic cancer – one of the hardest cancers to catch early.
The study, published in the medical journal Gut, found that the AI system could detect subtle changes in pancreatic tissue up to three years before a clinical diagnosis. Those changes were so faint that the scans had previously been read as normal.
Researchers say the tool, called REDMOD, is not looking for an obvious lump or mass. Instead, it analyzes tiny patterns in the texture and structure of the pancreas – signals that can be too subtle for the human eye to pick up.
That could matter because pancreatic cancer is often diagnosed late, after symptoms appear and the disease has already spread. The pancreas sits deep inside the abdomen, and early warning signs are often vague or absent.
More than 85% of patients are diagnosed after the cancer has spread, according to Mayo Clinic, and five-year survival rates remain below 15%. The disease is also projected to become the second-leading cause of cancer-related death in the U.S. by 2030.
For the study, researchers analyzed nearly 2,000 CT scans, including scans from people who were later diagnosed with pancreatic cancer. The scans had originally been interpreted as normal.
In an independent test group, REDMOD identified 73% of prediagnostic cancers at a median of about 16 months before diagnosis. Mayo Clinic said that was nearly twice the detection rate of specialists reviewing the same scans without AI help.
The advantage appeared even stronger for scans taken more than two years before diagnosis, when the AI identified nearly three times as many early cancers that otherwise went undetected.
Dr. Ajit Goenka, a Mayo Clinic radiologist and senior author of the study, said researchers believed the disease must leave clues before it becomes visible.
“We knew, based on the biology of the disease, that this is not something which is coming all of a sudden in three months,” he told NBC News. “We knew that the signal was there. We just needed to find a way to be able to detect it.”
For everyday patients, the idea is simple: a CT scan that looks normal today may still contain patterns that point to trouble developing in the background.
The research does not mean the AI tool is ready to be used for routine screening just yet, though. Experts say it needs to be tested further in clinical settings, including how often it produces false alarms and how doctors should respond when it flags a scan.
That caution is important as in cancer screening, false positives can lead to anxiety, extra tests and unnecessary procedures. The researchers have also noted that further validation is needed, including across more diverse patient groups.
But for now, Mayo Clinic is advancing the work through a clinical study called AI-PACED, which will evaluate how AI-guided detection could be integrated into care for people considered at higher risk. That includes tracking early detection, false positives and patient outcomes over time.
If the tool proves reliable, doctors say it may be especially useful for people already known to face a higher risk of pancreatic cancer, such as those with a family history or certain forms of new-onset diabetes.
The goal would not be for AI to diagnose cancer on its own. Instead, it could act as an extra set of eyes, flagging scans that deserve closer follow-up before symptoms begin.
“The greatest barrier to saving lives from pancreatic cancer has been our inability to see the disease when it is still curable,” Goenka said in a Mayo Clinic release.
He added that the model can identify “the signature of cancer” in a pancreas that appears normal.
Researchers are still years away from knowing whether the technology can improve survival. The clinical trial will need to follow participants over time to see who develops cancer and whether earlier warnings change outcomes.
But for a disease that is so often found too late, the study offers a hopeful step toward a future where doctors may be able to act before pancreatic cancer has a chance to fully reveal itself.
