Close Banner

Integrative Health

Mayo Clinic's New AI Tool Could Transform Pancreatic Cancer Diagnosis

Author: Sela Breen

May 03, 2026

Assistant Health Editor

By Sela Breen

Assistant Health Editor

Sela Breen is the Assistant Health Editor at mindbodygreen. She is a graduate of the Medill School of Journalism at Northwestern University, where she studied journalism, international studies, and theatre.

Image by TommL / iStock

May 03, 2026

Pancreatic cancer is known to be one of the deadliest forms of the disease. Not because it can't be treated, but because it's almost always caught too late. Over 85% of cases are diagnosed at a stage when the disease has already spread beyond what surgery can address.

Pancreatic cancer is essentially invisible on standard imaging when it's in its earliest, most treatable phase. But a new AI tool developed by researchers at Mayo Clinic may change that.

Why pancreatic cancer hides so well

Unlike many cancers that form a distinct mass early on, pancreatic cancer often develops without any visible changes on imaging. During this invisible stage, the pancreas looks completely normal, even to expert radiologists reviewing the scans.

By the time a tumor becomes visible, it's often too advanced for treatment that could lead to a cure. And because there's no established early detection strategy for the majority of pancreatic cancer cases, most people aren't diagnosed until symptoms appear, at which point, options are limited.

How AI sees what radiologists can't

The REDMOD (Radiomics-based Early Detection MODel) developed at Mayo Clinic doesn't look for a visible tumor. Instead, it analyzes subtle textural patterns in pancreatic tissues, called radiomic signatures, that indicate early cancer development is underway, even when the pancreas appears normal to the human eye.

The AI uses specialized image filters to detect microscopic disruptions in tissue structure. According to the study, 90% of the predictive features REDMOD relies on come from these filtered images, which capture patterns the human eye simply cannot perceive.

A radiologist can only identify a visible mass on scans, but REDMOD can detect the cellular-level changes that happen before a mass even forms.

RELATED READ: Recent Study Shows Creatine Intake May Lower Your Cancer Risk

What the research found

A new study, published in Gut, established REDMOD's efficacy by testing it across a large dataset from multiple institutions. Tests were designed to simulate real-world early detection conditions. Here's what stood out:

- 73% sensitivity: REDMOD correctly identified 73% of pre-diagnostic pancreatic cancers, compared to just 38.9% for radiologists. That's nearly double the current detection rate.

- Over one year of lead time: The AI detected cancer a median of 475 days before clinical diagnosis.That's about 16 months before doctors are normally able to catch it.

- Nearly 3x better at longer lead times: For cancers detected more than 24 months before diagnosis, REDMOD's sensitivity was 68% versus just 23% for radiologists.

- Strong reliability: The model showed over 90% consistency across repeat scans, meaning its predictions remained stable over time.

The AI also demonstrated strong accuracy in correctly identifying normal pancreases across both a multi-institutional set (81.3%) and a public NIH dataset (87.5%).

Who could benefit most

REDMOD isn't available for routine screenings yet, but the research points to specific groups who stand to gain the most from this developing technology.

People with new-onset diabetes have a nearly 20-fold higher risk of pancreatic cancer compared to the general population. In fact, some guidelines already recommend urgent abdominal imaging for individuals aged 60 and older with new-onset diabetes and weight loss.

But even with imaging, conventional CT scans often miss early-stage disease. REDMOD could serve as a critical addition to screening protocol, flagging high-risk individuals for closer monitoring or additional testing before a visible tumor forms.

What this means for the future

This represents a shift from reactive, symptom-driven diagnosis of pancreatic cancer to proactive detection before symptoms appear. The researchers note that even modest improvements in early diagnosis could more than double survival rates of this disease.

REDMOD is now heading into real-world clinical trials as part of a study called AI-PACED (Artificial Intelligence for Pancreatic Cancer Early Detection). This will explore how well the tool works in practice.

For now, the technology isn't available outside of research settings. But it's important to stay proactive about monitoring and discussing screening options with your doctor if you have risk factors for pancreatic cancer, like new-onset diabetes after age 60, a family history of the disease, or chronic pancreatitis.

The takeaway

Pancreatic cancer's deadliness has always been tied to late detection. REDMOD offers a glimp