Scientists develop AI that detects pancreatic cancer years before diagnosis in a major breakthrough
Dr Ajit Goenka led the Mayo Clinic team behind the AI that detects pancreatic cancer years before diagnosis. What is the AI breakthrough in pancreatic
Dr Ajit Goenka led the Mayo Clinic team behind the AI that detects pancreatic cancer years before diagnosis. What is the AI breakthrough in pancreatic cancer detection? Why pancreatic cancer is so difficult to detect How the REDMOD AI model works Understanding radiomics What the study found Why this could be a major breakthrough How this differs from liquid biopsy Is the AI available to patients? What the researchers concluded A promising step towards earlier diagnosis Researchers at Mayo Clinic have developed an artificial intelligence (AI) system that could transform how pancreatic cancer is detected by identifying subtle signs of the disease years before it is usually diagnosed. The technology, known as the Radiomics-based Early Detection Model (REDMOD), analyses routine CT scans to detect microscopic changes in pancreatic tissue that are invisible to the human eye. Published in the journal Gut in April 2026, the landmark study found that the AI could identify many future pancreatic cancer cases long before tumours became visible on imaging. Although REDMOD is still undergoing clinical evaluation, researchers believe it could become an important tool for diagnosing one of the world's deadliest cancers at a much earlier stage.The breakthrough centres on REDMOD, an AI model developed by Mayo Clinic researchers that analyses routine contrast-enhanced CT scans for hidden radiomic signatures associated with pancreatic cancer. Unlike traditional imaging methods that rely on spotting an existing tumour, REDMOD identifies tiny structural and textural changes within pancreatic tissue that may appear months or even years before cancer becomes visible.The findings, published in Gut, showed that the AI correctly identified 73% of future pancreatic cancer cases, detecting the disease a median of 16 months (475 days) before clinical diagnosis. In some patients, the AI detected warning signs up to three years before diagnosis, nearly doubling the detection rate achieved by radiologists reviewing the same scans. Researchers believe this could eventually transform routine abdominal CT scans into an early warning system for pancreatic cancer, although the technology is not yet approved for routine clinical use.Pancreatic cancer is one of the deadliest forms of cancer because it rarely causes symptoms during its early stages.
The pancreas is located deep inside the abdomen, making small tumours difficult to detect using conventional imaging. By the time symptoms such as abdominal pain, jaundice or unexplained weight loss develop, the disease has often spread beyond the pancreas.Researchers estimate that more than 85% of patients are diagnosed after the cancer has already advanced, leaving fewer treatment options. Current five-year survival rates remain below 15%, making early diagnosis one of the biggest unmet needs in cancer care. Detecting the disease before symptoms appear could significantly improve a patient's chances of receiving potentially curative treatment.Unlike conventional computer-aided detection systems that search for visible tumours, REDMOD uses radiomics, a technique that converts medical images into hundreds of quantitative measurements describing tissue texture, density, shape and microscopic structural patterns. The AI automatically isolates the pancreas on routine contrast-enhanced CT scans before analysing these hidden imaging features using machine-learning algorithms.Rather than looking for an existing tumour, REDMOD searches for biological changes within pancreatic tissue that occur long before a tumour becomes visible. According to the researchers, these subtle imaging signatures represent early tissue remodelling associated with developing pancreatic cancer.One of REDMOD's biggest advantages is that it does not require patients to undergo a special scan. Instead, it analyses routine CT scans that people may already have undergone for unrelated medical conditions, such as abdominal pain, digestive disorders or kidney stones. Researchers believe this could allow hospitals to identify high-risk individuals without exposing them to additional radiation or imaging procedures.Radiomics is an emerging field that transforms ordinary medical images into large amounts of measurable data. Instead of relying solely on what radiologists can visually observe, specialised computer software extracts hundreds or even thousands of quantitative features from every CT image.These features describe subtle differences in tissue texture, shape, density and spatial organisation that cannot normally be seen by the human eye. Machine-learning algorithms then analyse these patterns to determine whether they resemble healthy tissue or the earliest biological changes associated with disease.In the REDMOD study, researchers initially extracted nearly 1,000 radiomic features from each CT scan before selecting the most informative ones to build the final AI model.