AI Detects Pancreatic Cancer 3 Years Before Symptoms

Pancreatic cancer has always been a race against time that patients rarely win, but a new AI system might finally change those odds.

Mayo Clinic researchers have developed REDMOD, an artificial intelligence tool that detected pancreatic cancer on routine CT scans up to three years before traditional diagnosis. The study, published in Gut by the British Society of Gastroenterology, shows the system caught nearly three in four early cases that human radiologists missed.

Lead researcher Sovanlal Mukherjee and his team tested REDMOD on 493 CT scans. The AI correctly identified 73 percent of pre-diagnostic scans, while radiologists reviewing the same images caught less than 39 percent. When looking at scans taken more than two years before diagnosis, REDMOD performed nearly three times better than human experts.

The difference comes down to what REDMOD can see. The system hunts for "imaging-occult" cancers, meaning cases where no visible tumor exists even when experts review the scans. These aren't tumors that radiologists simply overlooked. They're microscopic changes in pancreatic tissue that signal disease is already underway, invisible to the human eye but detectable through mathematical analysis.

REDMOD analyzes nearly 1,000 features from each pancreas image and narrows them to 40 key signals. About 90 percent of those signals come from wavelet-filtered images, a technique that reveals subtle texture disruptions in tissue. Researchers believe these disruptions reflect early biological remodeling before any tumor appears.

The system combines three machine learning algorithms that work together to produce a final classification. Doctors can adjust the detection threshold depending on clinical context, balancing sensitivity against false positives. The model already exceeds UK health standards for acceptable cancer referral accuracy.

Testing showed the system maintained 90 to 92 percent consistency across repeat scans from the same patients. It also held its accuracy across different CT scanner brands and image quality settings, proving it works in real-world conditions.

Why This Inspires

Pancreatic cancer kills most patients because symptoms don't appear until the disease has spread. By the time someone feels sick enough to get scanned, treatment options have often run out. REDMOD offers something patients have never had: a head start.

The precision threshold of 36 percent might sound low, but it's twelve times better than the 3 percent standard considered acceptable for initial cancer screening. That gap represents thousands of lives that could be saved with earlier intervention.

Mayo Clinic is now planning a clinical trial called AI-PACED to test REDMOD in real-world high-risk populations before bringing it to standard care.

For the first time, routine scans might catch the cancer that's always been too fast to stop.