Pancreatic cancer is notoriously difficult to catch early. By the time most patients notice symptoms, the disease has already spread, making curative treatment nearly impossible. A new artificial intelligence tool developed at the Mayo Clinic aims to change that by spotting the cancer years before a conventional diagnosis.

The Radiomics Based Early Detection Model, or RED MOD, analyzes CT scans to identify subtle tissue changes that are invisible to the human eye. According to research published in the journal Gut, the AI system can detect early signs of pancreatic cancer up to three years before a patient would normally receive a diagnosis.

"It helps our team see patterns in the tissue that the human eye was never designed to see," said Dr. Ajay Goenka, senior author of the study and a radiologist at the Mayo Clinic.

How the AI works

RED MOD uses a technique called radiomics, which extracts hundreds of quantitative features from medical images that radiologists cannot perceive. These features include variations in texture, shape, and density that can signal the earliest stages of malignancy, even when no tumor is yet visible on a scan.

The team trained the model using nearly 2,000 CT scans, including images from patients who were eventually diagnosed with pancreatic cancer. The AI learned to associate specific patterns in seemingly healthy pancreatic tissue with the subsequent development of the disease.

In testing, RED MOD correctly identified early signs of cancer in about 73 percent of cases. On average, it flagged these signals 16 months before the patient received a clinical diagnosis. That detection rate is roughly double the performance of specialist radiologists reviewing the same scans without AI assistance.

Why pancreatic cancer is so hard to catch

Pancreatic cancer ranks among the deadliest cancers, with a five-year survival rate of around 12 percent. The pancreas sits deep inside the abdomen, and early-stage tumors rarely cause noticeable symptoms. When they do produce warning signs, such as jaundice, abdominal pain, or weight loss, the cancer has often already spread to other organs.

"We often catch these cancers too late to offer curative therapy," said Dr. Wesley Talcott of Lenox Hill Hospital, who was not involved in the study but commented on its significance. He called the AI tool "another tool for doctors in the fight against this aggressive disease" and noted there is "a lot of room for optimism."

A delay of even a few months can be the difference between a tumor that can be surgically removed and one that has metastasized beyond treatment. The Mayo Clinic researchers believe that moving the moment of diagnosis forward by even a year could transform patient conversations from managing an incurable disease to discussing a cure.

RED MOD in the clinic today and tomorrow

The model is not yet ready for widespread use. Dr. Goenka told CBS News that the team wants to deliver "a technology that is battle tested." The next step is a clinical trial to validate the tool in real-world hospital settings across a larger and more diverse patient population.

One open question is how the AI will perform on scans from different CT machines and imaging protocols used at various institutions. The training data came from a single healthcare system, so generalizability remains to be proven. There are also concerns about false positives, which could lead to unnecessary biopsies, anxiety, and additional procedures.

"Still a long ways to go, but certainly a lot of room for optimism," Talcott added.

Broader momentum for AI in diagnostics

The Mayo Clinic announcement arrives alongside other advances in AI-assisted medicine. Researchers at Harvard Medical School and Beth Israel Deaconess Medical Center recently found that an AI reasoning model outperformed two experienced physicians when diagnosing patients and managing their care, using only electronic health records and limited clinical information that had been available to the doctors at the time.

That study, published separately, suggests that large language models and reasoning systems are becoming capable of handling complex diagnostic tasks. Combined with specialized tools like RED MOD, these technologies could eventually serve as a second opinion for clinicians or even as a screening front-end for high-risk populations.

What it means for patients

For now, the Mayo Clinic AI remains a research tool. But if it passes clinical trials, the potential impact is enormous. Pancreatic cancer is the third leading cause of cancer death in the United States. The American Cancer Society estimates that more than 66,000 people will be diagnosed with the disease in 2024, and roughly 52,000 will die from it.

Early detection is the single most effective way to improve survival. When pancreatic cancer is caught while still localized to the pancreas, the five-year survival rate jumps to 44 percent. That is a dramatic improvement over the single-digit rates for distant-stage disease.

RED MOD may never replace the annual physical or a doctor's judgment. But it could become an inexpensive, noninvasive screening layer for patients with risk factors such as family history, new-onset diabetes, or chronic pancreatitis. A CT scan is already a common part of diagnostic workups for abdominal complaints. Running it through an AI model adds negligible cost and time.

The road ahead

The Mayo Clinic team is actively planning the clinical trial, which will need to demonstrate not just detection accuracy but also that the tool leads to better patient outcomes. That means proving that patients whose cancers are caught early by the AI actually live longer or require less aggressive treatment.

Until then, RED MOD joins a growing list of promising AI diagnostics that have yet to cross the chasm from published paper to everyday practice. The technology works in the lab. The harder problem is making it work in the messy, variable world of real hospitals.

But the direction is clear. AI is learning to see what doctors cannot. For a disease as deadly as pancreatic cancer, that extra sight could make all the difference.