AI Breakthrough Detects Pancreatic Cancer Up to Three Years Before Human Doctors

Pancreatic cancer is one of the most lethal diagnoses in medicine, primarily because it hides until it is too late. Now, a new artificial intelligence model developed by researchers at the Mayo Clinic and the University of Texas MD Anderson Cancer Center is changing that timeline. The system, called REDMOD, can detect the earliest signs of the disease on routine abdominal CT scans up to three years before human doctors can spot a tumor.[1][2][3][4]

The findings, published in the medical journal Gut, represent a massive leap forward in oncological diagnostics. By analyzing scans that had previously been cleared as "normal" by human radiologists, the AI successfully identified 73% of prediagnostic cancers at a median of 16 months before actual diagnosis.[1][3][4]

This nearly doubles the detection rate of human specialists reviewing the exact same scans without AI assistance, whose pooled sensitivity sat at just 38.9%. The advantage was even more pronounced at earlier timeframes: for scans taken more than two years prior to diagnosis, the AI caught 68% of cases, compared to a mere 23% by human doctors.[2][4]

The stakes for this early detection are life and death. Pancreatic cancer rarely presents symptoms in its nascent stages. As a result, more than 85% of patients receive their diagnosis only after the cancer has metastasized to other organs. This late detection is the primary reason the five-year survival rate remains a grim 13% in the United States.[1][2][3]

"The greatest barrier to saving lives from pancreatic cancer has been our inability to see the disease when it is still curable," explained Dr. Ajit Goenka, a Mayo Clinic radiologist and the study's senior author. If doctors can identify the cancer years earlier, patients can undergo surgical resection and begin targeted therapies while the disease is still localized.[1][2][3]

How does the AI see what highly trained human eyes miss? REDMOD, which stands for Radiomics-based Early Detection Model, does not look for visible tumors. Instead, it converts the 2D images from a CT scan into a complex 3D mathematical puzzle. It evaluates the pancreas pixel by pixel, extracting hundreds of quantitative imaging features that describe tissue texture and structure.[1][2][4]

These mathematical features capture the faint, multiscale biological changes—what researchers call "visually occult" signatures—that occur as healthy tissue begins its slow transformation into a malignancy. The AI is essentially reading the microscopic structural degradation of the organ long before a physical mass forms.[2][4]

However, the technology is not designed to replace human radiologists. While REDMOD is vastly superior at catching early cancer, human doctors still hold a slight edge in ruling out healthy patients. The AI correctly identified disease-free patients 81.1% of the time, compared to an average of 92.2% for human radiologists. Experts emphasize that the future lies in "AI augmentation," where the algorithm flags suspicious mathematical patterns for a human physician to evaluate.[2][4]

The Mayo Clinic is already advancing the technology into clinical testing through a prospective study known as AI-PACED. This next phase will evaluate how doctors can integrate the AI-guided detection into real-time care for patients at elevated risk, tracking false positives and long-term clinical outcomes.[1]

The diagnostic breakthrough arrives alongside a wave of promising new treatments for the disease. The FDA recently greenlit expanded access for daraxonrasib, a new drug targeting the KRAS mutations that drive the vast majority of pancreatic cancers. Clinical trial data presented at the American Society of Clinical Oncology showed the drug nearly doubled survival rates for study volunteers.[5]

Furthermore, the FDA has begun granting "breakthrough device" designations to similar AI diagnostic tools, signaling a regulatory willingness to fast-track these life-saving algorithms into hospitals.[5]

Combined, these advancements offer the first genuine hope in decades for a disease that has stubbornly resisted medical progress. By pairing AI-driven early detection with next-generation targeted therapies, the medical community is laying the groundwork to finally turn the tide against pancreatic cancer.[1][5]