Artificial intelligence is reshaping early cancer detection — and China has become one of its most ambitious testing grounds.
A Diagnosis That Arrived Before Symptoms
Three days after a routine medical exam for diabetes, Qiu Sijun, a 57-year-old retired construction worker in eastern China, received an unexpected phone call.
The caller was not his usual doctor, but the head of the hospital’s pancreatic department. He was asked to return immediately.
“I knew it couldn’t be good news,” Qiu later recalled.
The diagnosis confirmed his fear: pancreatic cancer — one of the deadliest and hardest-to-detect forms of the disease. Yet there was a crucial difference from most cases. The tumor had been discovered early, before symptoms appeared, allowing surgeons to remove it successfully.
The detection was not made by a doctor alone, but by an artificial intelligence system quietly analyzing routine CT scans in the background.
Why Pancreatic Cancer Is So Hard to Catch
Pancreatic cancer has one of the lowest survival rates among major cancers, with a five-year survival rate of roughly 10%. The reason is brutally simple: symptoms usually appear only after the disease has advanced.
Widespread screening is discouraged because confirmatory tests, such as contrast-enhanced CT scans, expose patients to significant radiation. Safer alternatives, like non-contrast CT scans, produce less detailed images, making early abnormalities difficult for radiologists to identify.
This diagnostic gap has persisted for decades — until artificial intelligence entered the equation.
How AI Learned to See What Humans Miss
At the Affiliated People’s Hospital of Ningbo University, doctors began testing an AI system called PANDA (Pancreatic Cancer Detection with Artificial Intelligence). Developed by researchers affiliated with Alibaba’s DAMO Academy, the system was trained to detect pancreatic tumors using non-contrast CT scans.
To overcome the lack of image clarity, researchers manually annotated contrast-enhanced scans from over 2,000 confirmed pancreatic cancer patients. These annotations were algorithmically mapped onto corresponding non-contrast scans, allowing the AI to learn subtle visual patterns invisible to the human eye.
When tested on more than 20,000 non-contrast CT scans, the model correctly identified pancreatic lesions in 93% of confirmed cases, according to a study published in Nature Medicine.
Since late 2024, PANDA has analyzed over 180,000 routine CT scans at the Ningbo hospital, helping detect nearly two dozen pancreatic cancer cases — 14 of them at an early stage.
China’s Rapid Push to Apply AI in Medicine
China’s healthcare system offers a unique environment for AI experimentation. Large patient volumes, widespread use of routine imaging, and close collaboration between hospitals and major technology companies allow systems like PANDA to be tested at scale.
In April, Alibaba announced that the U.S. Food and Drug Administration granted PANDA “breakthrough device” designation, accelerating its review process for potential commercialization.
While similar AI-assisted detection efforts exist in other countries, China’s ability to deploy such tools quickly has turned hospitals into real-world laboratories for medical AI.
The Risks: False Alarms, Trust, and Infrastructure
Despite its promise, PANDA is not without controversy.
Since deployment, the system has flagged roughly 1,400 scans as high-risk, but only about 300 required further medical follow-up. False positives can cause anxiety, unnecessary testing, and invasive procedures.
Medical experts caution that AI tools must reduce false alarms before they can be widely adopted. Others note that such systems may benefit hospitals with fewer specialists more than elite medical centers.
There are also practical challenges. Outdated hospital hardware has struggled to process large AI models, and staff shortages limit the ability to follow up with every flagged patient. In some cases, the system has even crashed under heavy computational loads.
A Tool — Not a Replacement
Doctors involved in the project emphasize that PANDA does not replace specialists. It serves as a second set of eyes, reviewing scans already ordered for other reasons and highlighting cases that might otherwise go unnoticed.
For patients like Qiu, the technology made all the difference.
“I don’t use AI, and I don’t really understand how it works,” he said after his recovery. “But the doctor told me I was lucky. All I could feel was relief.”
The Future of Early Detection
The success of PANDA illustrates both the potential and the complexity of AI-driven medicine. Detecting cancer earlier can save lives — but it also raises ethical, technical, and social questions that healthcare systems around the world will need to confront.
As AI continues to mature, its greatest impact may not come from replacing doctors, but from quietly changing when — and how — life-threatening diseases are discovered.
Sources & References
- The New York Times — AI Is Detecting Deadly Tumors Doctors Often Miss in China
- Nature Medicine (2023) — Study on AI-assisted pancreatic cancer detection
- Alibaba DAMO Academy — PANDA project disclosures
- U.S. Food and Drug Administration — Breakthrough Devices Program
