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.

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.

What is the AI breakthrough in pancreatic cancer detection?

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.

Why pancreatic cancer is so difficult to detect

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.

How the REDMOD AI model works

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.

Understanding radiomics

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. Many of the strongest predictive signals came from wavelet-filtered images, which enhance subtle tissue characteristics invisible during routine clinical review.

What the study found

The researchers trained and validated REDMOD using nearly 2,000 abdominal CT scans collected from multiple healthcare institutions. Many of these scans had originally been reported as normal but belonged to patients who were later diagnosed with pancreatic cancer.During validation, the AI correctly identified 73% of future pancreatic cancers, detecting them a median of about 16 months (475 days) before clinical diagnosis. In scans obtained more than two years before diagnosis, REDMOD identified nearly three times as many cancers as radiologists reviewing the same scans without AI assistance.The researchers also found that REDMOD maintained consistent performance across CT scans obtained using different scanner manufacturers, imaging protocols and healthcare institutions. This suggests the model could be integrated into a wide range of hospitals rather than being limited to a single imaging system or clinical setting.

Why this could be a major breakthrough

The greatest challenge in pancreatic cancer treatment has always been detecting the disease while it is still curable. By identifying microscopic tissue changes before tumours become visible, REDMOD could provide doctors with valuable time to investigate suspicious cases and begin treatment earlier.Explaining the importance of the findings, Dr Ajit Goenka, radiologist at Mayo Clinic and one of the study’s lead researchers, said the technology identifies imaging biomarkers before a tumour becomes visible, creating an opportunity to intervene when treatment is likely to be far more effective.The researchers also emphasised that REDMOD is designed to support clinicians rather than replace them. Instead of making a diagnosis on its own, the AI could help radiologists recognise subtle warning signs that might otherwise go unnoticed.Researchers further suggested that REDMOD could eventually function as an opportunistic screening tool, automatically analysing routine CT scans performed for unrelated medical conditions and alerting clinicians when hidden imaging features indicate an increased risk of pancreatic cancer.

How this differs from liquid biopsy

Another promising strategy for early pancreatic cancer detection is the liquid biopsy, which analyses blood samples for tumour DNA, RNA, proteins and other cancer-related biomarkers released into the bloodstream.Unlike REDMOD, which analyses medical images, liquid biopsy searches for molecular evidence of cancer. Researchers believe the two technologies could complement each other rather than compete. Combining AI-powered radiomics with blood-based biomarkers may improve diagnostic accuracy and enable even earlier detection in the future.

Is the AI available to patients?

Not yet. Although the findings are highly encouraging, REDMOD remains an investigational technology and has not been approved for routine clinical use or population screening.To determine whether the AI improves patient outcomes in real-world healthcare, Mayo Clinic has launched the AI-PACED (Artificial Intelligence for Pancreatic Cancer Early Detection) prospective clinical trial. The study will evaluate REDMOD among people considered to be at increased risk of pancreatic cancer while measuring its impact on diagnosis, treatment decisions and long-term outcomes.

What the researchers concluded

The researchers concluded that REDMOD demonstrates strong potential to identify pancreatic cancer during its pre-diagnostic phase by detecting radiomic signatures that remain invisible during routine CT interpretation. However, they stressed that prospective clinical trials are still needed to determine whether integrating the technology into everyday healthcare can improve survival and reduce pancreatic cancer deaths.According to Mayo Clinic, REDMOD has the potential to transform ordinary abdominal CT scans into an early warning system for pancreatic cancer by identifying hidden imaging biomarkers before tumours become visible. If validated through future clinical studies, the technology could allow doctors to investigate suspicious cases much earlier than is currently possible.

A promising step towards earlier diagnosis

The REDMOD study highlights how artificial intelligence is moving beyond simply recognising visible abnormalities and beginning to detect the earliest biological fingerprints of disease. Rather than replacing radiologists, the technology has the potential to enhance routine imaging by revealing warning signs that human observers cannot see.While REDMOD is not yet part of standard clinical care, researchers believe it represents one of the most promising advances in pancreatic cancer detection in recent years. If ongoing clinical trials confirm the findings, routine CT scans performed for unrelated medical conditions could one day become powerful tools for identifying pancreatic cancer before symptoms develop, giving thousands of patients a far better chance of receiving life-saving treatment.