A groundbreaking artificial intelligence model now detects pancreatic cancer up to three years before physicians can spot it on CT scans. Researchers published this finding on April 28, 2026, in the journal Gut. The tool, called REDMOD (Radiomics-based Early Detection Model), analyzed nearly 2,000 CT scans previously cleared as “normal.” It identified tiny structural changes in the pancreas that later developed into tumors. This advance could transform how doctors approach one of medicine’s most feared diagnoses.
Why Early Detection of Pancreatic Cancer Is So Critical
A Disease That Hides Until It Is Too Late
Pancreatic cancer consistently ranks among the deadliest cancers worldwide. The five-year survival rate in the United States sits at just 12% to 13%. That grim statistic exists for one key reason: doctors rarely catch the disease early enough for treatment to succeed.
“The five-year survival rate is about 12% to 13% because of our inability to detect it at a time when therapeutic options could work their magic,” said study co-author Dr. Ajit Goenka, a radiologist and nuclear medicine specialist at the Mayo Clinic in Rochester, Minnesota.
The early stages of pancreatic cancer trigger almost no symptoms. Consequently, most patients receive their diagnosis only after the cancer reaches an advanced, often terminal stage. Standard diagnostic methods — including tissue sampling and CT imaging — work well in many other cancers. Yet for pancreatic cancer, tumors tend to stay invisible until treatment options narrow sharply.
The Gap Science Has Yet to Close
Medical research has produced dramatic improvements in detecting and treating many cancers over recent decades. Breast cancer screening, colorectal surveillance, and cervical cancer prevention programs all demonstrate real gains. However, no comparable breakthrough has emerged for pancreatic cancer. This gap has pushed researchers to explore AI-assisted imaging as a new path forward.
How the REDMOD AI Tool Works
Finding Signals That Humans Cannot See
The cancer development process does not begin months before diagnosis — it begins years, or even decades, earlier. Dr. Goenka explained the concept clearly: “The basic science research tells us that the process of cancer development is not something that starts six months earlier. It starts 10 to 15 years earlier, which means that there was a signal in the pancreas and that signal was outside the purview of human detectability.”
REDMOD leverages AI to detect exactly those hidden signals. The model operates through two core steps. First, it segments the pancreas from a standard 2D CT scan and builds a full 3D structural model of the organ. Then it evaluates that model pixel by pixel, searching for subtle deviations from normal tissue.
Mathematics at the Core of the Model
The tool essentially converts a medical image into a mathematical problem. “It’s taking each and every pixel in that image and it is quantifying the degree to which it differs from the rest of the organ,” Goenka explained. “At the end of the day, it’s mathematics. It converts that image into a mathematical representation and extracts those mathematical features.”
The model compares those extracted features against a control group of scans without expected pathological changes. Differences too subtle for the human eye to register become clear patterns in the data.
What the Test Results Revealed
Strong Performance on Overlooked Scans
The research team tested REDMOD on a sample of approximately 2,000 existing CT scans. Clinicians had originally collected those scans for conditions unrelated to cancer. Moreover, radiologists had previously reviewed and cleared all of them as normal. About one-seventh of those scans belonged to patients who later developed pancreatic cancer.
REDMOD successfully flagged 73% of those early-stage cases. Furthermore, the scans the model analyzed came, on average, from 16 months before the patient’s actual cancer diagnosis — meaning the AI spotted warning signs that human reviewers had missed more than a year earlier.
A Twofold — and Threefold — Advantage Over Radiologists
The performance gap between the AI and human radiologists grew even wider at earlier time windows. “The sensitivity gain over radiologists was nearly twofold across the spectrum,” Goenka noted. “When you look at even earlier — more than two years prior to diagnosis — that sensitivity gain was almost threefold.”
In other words, the further back in time researchers looked, the more pronounced REDMOD’s advantage became. That finding suggests the tool could enable intervention at a point when treatment remains genuinely curative.
Limitations and the Path Forward
The AI Is Not Yet a Replacement for Radiologists
REDMOD’s current version does carry notable limitations. Specifically, it generates more false positives than human radiologists do. The model correctly identified healthy patients 81.1% of the time, while human radiologists achieved an accuracy rate of 92.2%. That gap matters in clinical settings, where false alarms lead to unnecessary procedures and patient anxiety.
“There is a complementary role for both of them — physician expertise combined with AI augmentation,” Goenka acknowledged. Rather than replacing radiologists, the tool works best as an additional layer of scrutiny that catches what human reviewers miss.
Clinical Trials Are Already Under Way
Despite these limitations, the trajectory is encouraging. Dr. Goenka aims to see REDMOD implemented routinely in clinical settings within the next five years. The team is currently running clinical trials to validate the detection strategy in real-world practice, not just retrospective data sets.
What Experts Say About the Future
High-Risk Groups Stand to Benefit Most
Tatjana Crnogorac-Jurcevic, a professor of molecular pathology and biomarkers at Queen Mary University of London and an independent expert not involved in the study, praised both the study design and its implications. She also pointed out that general population screening for pancreatic cancer remains impractical given the disease’s relative rarity. Instead, the tool’s greatest near-term value lies with defined high-risk groups.
“There are defined high-risk groups for which surveillance will be possible,” she explained — particularly individuals with a family history of pancreatic cancer, those carrying relevant cancer-linked gene mutations, and patients with new-onset diabetes.
Combining AI With Biomarker Tests Could Be Transformative
Looking further ahead, Crnogorac-Jurcevic described an exciting convergence on the horizon. Her own team develops urine-based biomarker tests that target the same goal: catching pancreatic cancer as early as possible. Combining such liquid biopsy approaches with an AI imaging tool like REDMOD could multiply the gains seen in either method alone.
“Having an AI imaging tool to combine with our body fluid biomarkers would be fantastic,” she said. “It’s highly likely that they will be complementary, which would increase the sensitivity and accuracy of early detection massively.”
Together, these converging technologies hold real promise for shifting pancreatic cancer from a disease almost always caught too late into one that doctors can intercept before it turns fatal.
