AI for Pancreatic Cancer Detection

With a 5-year relative survival rate of just 13%, pancreatic cancer is one of the most lethal cancers in the world. The primary reason for its lethality is delayed diagnosis due to the disease being asymptomatic until later stages. New research published in Gut offers hope against this challenge and could potentially change the trajectory of pancreatic cancer diagnosis worldwide.

In a paper published on May 8, 2026, a team of scientists found that an artificial intelligence (AI) model could detect pancreatic cancer up to three years before physicians diagnose a tumor.

The AI model, developed at the Mayo Clinic in Rochester, was used to analyze about 2,000 computer tomography (CT) scans previously classified as ‘normal,’ i.e., bearing no signs of disease. The CT scans were from patients who had first been screened for other medical conditions unrelated to pancreatic cancer, and about one-seventh of those patients had later been diagnosed with the disease.

The AI model could discern irregularities in pancreatic tissue that later developed into cancer with almost three times the sensitivity of radiologists’ assessments. The sensitivity of a diagnostic tool refers to its ability to identify true positive cases of disease.

The difference in sensitivity between the AI tool and the radiologists’ detection was the highest in cases with a lead time greater than 24 months, implying that the AI tool is best used to identify early signs of disease.

“We knew that the signal was there. We just needed to find a way to detect it,” remarked Ajit Goenka, MD, who is a radiologist at the Mayo Clinic and the lead scientist on the team.

Though pancreatic cancer is the 12th most common cancer in the world, it is the 3rd leading cause of cancer-related deaths in the United States, since over 80% of cases are diagnosed at an advanced stage. Early signs of the disease are almost invisible to the naked eye, even on a diagnostic scan. Furthermore, symptoms such as abdominal pain or weight loss appear only once the disease has progressed to later stages; in some cases, such symptoms are fairly general and may not adequately signal the presence of cancer until it is too late.

Therefore, the development of an AI tool for the early diagnosis of pancreatic cancer is a particularly remarkable step towards improving the lives of those at risk of the disease, as early detection can significantly enhance treatment and survival outcomes. The tool would be particularly useful for those with a family history of the disease or those at risk due to environmental or medical factors. The AI model can be combined with bloodwork and additional imaging to detect early signs of the disease in high-risk individuals.