AI tool catches pancreatic cancer in routine scans before symptoms appear

According to physician Zhu Kelei, AI has definitively saved the lives of patients whose scans were only flagged by PANDA, an AI tool developed by Alibaba researchers. The system analyzes non-contrast CT images—scans where even experienced radiologists can easily miss tumors.

Pancreatic cancer ranks among the deadliest cancers, with a five-year survival rate of just around 10 percent. The main reason is that early detection is extremely difficult, as symptoms often don't appear until the cancer has already progressed. Contrast-enhanced CTs show abnormalities more clearly, but the high radiation exposure makes them unsuitable for widespread screening. Non-contrast CTs produce less detailed images, making it harder for radiologists to spot tumors.

Training AI on low-detail images

That's where PANDA comes in. The engineers had a radiologist manually annotate contrast-enhanced CTs from over 2,000 known pancreatic cancer patients. These marked lesions were then algorithmically transferred to non-contrast scans from the same patients. This approach taught the AI model to recognize potential cancer sites even in less detailed images. Ling Zhang, lead algorithm engineer at Alibaba's research division Damo Academy, says the effectiveness actually surprised the team.

A clinical trial with the tool has been running at Ningbo University Hospital in eastern China since November 2024. Since then, PANDA has analyzed over 180,000 abdominal or chest CTs and detected around two dozen cancer cases, 14 of them in early stages. All of these patients had come to the hospital for other complaints like bloating or nausea and hadn't initially seen a pancreatic specialist. Several of their CT scans hadn't raised any red flags with human doctors until the AI tool flagged them.

FDA fast-tracks approval process

The FDA granted PANDA "Breakthrough Device" status in April of last year, enabling an accelerated approval process in the United States. Several clinical trials are underway in China, including one at a rural clinic in Yunnan province.

But widespread use also brings side effects: of 1,400 alerts from the system, only about 300 actually required follow-up examination. Ajit Goenka, a radiologist at the Mayo Clinic, points out that potentially hundreds of people went through the terror of a potential pancreatic cancer diagnosis, underwent unnecessary follow-up examinations, and probably endured expensive, invasive follow-up tests, only to find out they were healthy.

Diane Simeone, a pancreatic surgeon at the University of California San Diego, also expresses skepticism: some tumors detected by the tool should have been obvious to well-trained radiologists, even without AI. Still, she acknowledges that the system could serve as a valuable safety net at hospitals with fewer specialists.