The invisible killer is losing its hiding spots.
Pancreatic cancer has long been one of medicine’s toughest battles, often hiding in the body until it is already advanced. But new research suggests AI could be changing that timeline, spotting warning signs years before doctors can.
A study led by researchers at the Mayo Clinic has shown that an AI system can detect early signs of pancreatic cancer on routine CT scans up to about three years before a formal diagnosis is made.
The system, known as REDMOD (Radiomics-based Early Detection Model), was trained to pick up extremely subtle changes in the pancreas that are invisible during standard radiology reviews. These scans are often taken for unrelated issues, such as abdominal pain, and are usually read as normal at the time.
Researchers tested nearly 2,000 CT scans, including hundreds from patients who were later confirmed to have pancreatic cancer. Many of those earlier scans showed no signs of disease.
According to a study published in Gut, the AI identified about 73% of these pre-diagnostic cases, on average, roughly 16 months before diagnosis. In comparison, radiologists reviewing the same scans without AI assistance identified far fewer early cases.
Stronger performance years before diagnosis
The advantage of the AI system became even more noticeable in earlier scans. For images taken more than two years before diagnosis, the system detected significantly more cases than human experts reviewing the same data.
Reports also indicate that in some comparisons, the model performed nearly twice as well as radiologists in identifying early warning signs, highlighting how difficult these early-stage changes are to spot without computational support. The system works by analyzing radiomic features, tiny patterns in tissue texture and structure that may reflect early biological changes long before a tumor becomes visible.
Why this matters for survival
Pancreatic cancer is often diagnosed late because it rarely shows symptoms in its early stages. As a result, most patients are diagnosed only after the disease has spread, when treatment options are limited. Data from health agencies show survival rates remain very low compared to many other cancers, largely due to late detection.
Researchers involved in the study say that shifting the diagnosis earlier could significantly change outcomes. One of the key arguments from the study is that moving more cases into earlier, localized stages could improve survival rates. Unlike experimental models that rely on highly controlled data, REDMOD was tested using scans from multiple hospitals and different imaging systems. This was done to better reflect real clinical conditions.
The AI is also designed to run automatically on routine CT scans, without needing extra steps or special preparation. Researchers say this makes it more realistic for use in everyday hospital workflows, especially for patients already considered at higher risk.
Moving toward clinical testing
The next step is real-world testing. The Mayo Clinic team is currently advancing a prospective study called AI-PACED, which will evaluate how the system performs when used in clinical settings and how it affects patient care decisions over time. The research effort is part of a larger push at Mayo Clinic to use AI for early prediction and prevention of disease before symptoms appear.
In other AI news: For more on how AI is reshaping jobs beyond healthcare, read how Amazon’s latest layoffs reflect a broader shift toward automation and workforce restructuring.