AI detects invisible signs of pancreatic cancer more than a year before diagnosis

Summary

An artificial intelligence model called REDMOD detected very early and normally invisible signs of pancreatic cancer in CT scans that had been considered normal by radiologists. The research was published in the peer-reviewed journal `Gut` and shared by BMJ Group through EurekAlert.

According to the researchers, the model identified subtle signatures of pancreatic ductal adenocarcinoma, the most common form of pancreatic cancer, an average of 475 days before clinical diagnosis. That matters because pancreatic cancer is often detected late, when curative treatment options are limited.

In practice

REDMOD was built to analyse texture patterns in pancreatic tissue on computed tomography scans. These patterns, known as radiomics, can be too subtle for the human eye or conventional radiology review to detect.

In the study, researchers applied the model to abdominal CT scans from 219 patients who showed no visible evidence of disease at the time of imaging, but who were later diagnosed with pancreatic cancer months or years afterward. Those scans were compared with scans from 1,243 people who did not develop the disease up to three years later.

What the results show

The model was more sensitive than experienced radiologists at detecting these pre-clinical signs. According to the release, REDMOD reached 73% sensitivity, compared with 39% for radiologists. For cases more than two years before clinical diagnosis, the gap was also substantial: 68% versus 23%.

Those figures make the finding promising, but they do not mean the technology is ready to replace doctors or be used broadly in clinical practice. The researchers themselves stress that the model needs prospective testing in high-risk patients, such as people with unexpected weight loss and newly diagnosed diabetes.

Why it matters

• Pancreatic cancer remains one of the hardest cancers to detect early.
• An average lead time of 475 days before diagnosis could radically change treatment possibilities.
• The study shows how AI can detect signals in medical images that appear normal to humans.
• Clinical validation is still essential before broad hospital use.

The importance of this news is its potential to shift the disease timeline. If models like REDMOD are validated in real-world clinical settings, AI could help move pancreatic cancer from a disease often diagnosed too late to one detectable at a more treatable stage.

Still, the core message should remain cautious: this is a promising scientific advance, not a clinical tool ready for universal use.