Mayo Clinic Uses AI to Detect Pancreatic Cancer Earlier

When Dr. Mark Truty was in college, his father was diagnosed with pancreatic cancer. Doctors told the family the tumor appeared localized and attempted surgery.

“I remember when the surgeon came out, he said they got most of it. And I never understood that. Like, what do you mean most of them?” Truty recalled, in a Mayo Clinic video. The cancer spread to his father’s liver and abdomen, and he eventually died in his son’s arms.

“That’s probably my inner drive, is that I don’t want this to happen to anyone else,” said Truty, now a surgical oncologist at Mayo Clinic.

Pancreatic cancer is one of the deadliest malignancies, often detected too late for effective treatment. According to the American Cancer Society, the five-year survival rate stands at 13%, but if caught early and confined to the pancreas, that rate jumps to 44%.

“It’s the most deadly — and the least survivable — of all cancers that we know of,” Truty said in a blog post. “That’s what makes it such an awful disease to deal with.”

Researchers at Mayo Clinic believe artificial intelligence (AI) can change that grim equation.

Mayo scientists trained AI models to analyze CT scans and detect pancreatic tumors with far greater precision than radiologists working unaided. While it can take a specialist 20 to 30 minutes to segment a pancreas on a scan — and still not be fully accurate — AI can do it in a fraction of a second across thousands of images.

Dr. Ajit Goenka, a radiologist and nuclear medicine specialist at Mayo Clinic, told PYMNTS that AI was able to detect signs of pancreatic cancer at a median 12 to 13 months early, with some cases up to 3 years ahead of human doctors’ detection.

Goenka said the AI models were more than 90% accurate in determining whether or not patients had pancreatic cancer — and were 95% to 98% accurate in ruling out the disease in healthy patients, meaning they rarely produced false positives.

The healthcare sector recognizes the value of AI. According to a 2024 PYMNTS Intelligence report, healthcare executives are “doubling down” on their generative AI investments, with a long-term growth perspective. The report found that 90% expect a positive return on investment (ROI) and 59% plan to increase their investments in 2025.

On average, respondents expect to fully embed this technology across their businesses in 7.4 years. The top two areas of generative AI deployment are innovation of products and services and automated responses to customer queries, at 58.6% each.

Read also: Healthcare Firms Going Long on GenAI Investment

Goenka said the “holy grail” of medicine is to detect and identify disease before it can be perceived by humans. That head start matters. By the time symptoms develop and a tumor shows up on standard imaging, the disease has usually spread beyond the pancreas.

Truty said identifying the cancer earlier gives patients a much higher chance at survival. “If we can identify a patient with pancreas cancer at a much earlier stage, those patients are much more likely to have a curative procedure for long-term care,” he said.

The research has already shown cases where artificial intelligence picked up tumors invisible to radiologists on earlier scans. Truty described one patient who had a CT scan a year before diagnosis that appeared clear to doctors.

“The computer can see individual pixels within that CT scan,” Truty said, noting that AI flagged the tumor that would only later become clinically apparent.

Asked when this technology will be deployed in the field, Goenka said the Mayo Clinic has launched a clinical trial called AI-PACED (AI-Augmented Pancreas Cancer Early Detection). The study follows two groups of patients and uses a custom AI-powered triage platform that plugs directly into the electronic medical record system.

The software scans imaging data and medical histories to flag people who may be at high risk for pancreatic cancer, alerting doctors while also collecting blood samples to build a biobank for further research.

As part of the trial, high-risk participants undergo regular CT scans, which are read by independent radiologists for clinical decision-making, Goenka said. At the same time, the AI runs in the background, analyzing the images without influencing treatment choices during the study.

If the technology proves it can consistently detect tumors earlier than human eyes alone, Goenka said the system could be rolled out in stages across its hospitals and partner sites within three to five years — a shift that could dramatically improve survival odds of patients.

For hospitals, Goeka said, it would take longer to deploy because they will need the following in addition to accuracy metrics: Prospective proof of clinical utility; regulatory clearance; generalizability and safety; workflow, privacy, and liability readiness; and economic viability.

Beyond pancreatic cancer, the technology could have wider applications.

“It’s not even just specific for pancreas cancer. This can be applied to a variety of other cancers,” Truty said. “Instead of waiting until a patient develops symptoms, we can identify patients who are at potentially higher risk, those that have risk factors, and therefore screen them earlier on, so that we can then intervene at a much earlier date.”

The early diagnosis could stretch survival rates now measured in months to years and decades, and “ultimately, a cure,” Truty said.

For a disease that has long been a silent killer, AI is offering doctors and patients something rare: optimism. “There is definitely hope on the horizon,” Goenka said. “We just have to make the right kind of effort and use cutting-edge technologies like AI to make that dent in this disease.”

Read more:

Can AI-Powered Smart Glasses Help Spot Medication Errors?

Apple Reportedly Developing AI Agent ‘Doctors’ in Latest Health Push

AI Medical Note-Taking Apps Enjoy Healthy Wave of Investment

For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.

The post Mayo Clinic Uses AI to Detect Pancreatic Cancer Earlier appeared first on PYMNTS.com.