China leads AI push to detect hard-to-find cancers early
A new artificial intelligence system is being integrated into routine hospital scans in China, helping identify pancreatic tumors earlier.
-
A doctor examines a magnetic resonance image of a human brain during the CHAIN Cup at the China National Convention Center in Beijing, Saturday, June 30, 2018 (AP)
Hospitals across China are increasingly turning to artificial intelligence to identify deadly cancers that can escape even trained medical eyes, offering new hope in cases where early detection is notoriously difficult, The New York Times reported.
Three days after Qiu Sijun, a 57-year-old retired bricklayer in eastern China, went in for a routine diabetes exam, he received an unexpected call from a senior doctor he had never met. The head of the hospital’s pancreatic department asked him to return for further tests.
“I knew it couldn’t be anything good,” Qiu told The New York Times.
The diagnosis confirmed his fears, pancreatic cancer, but also delivered unexpected relief: the tumor had been detected early and could be surgically removed.
How AI flagged what humans missed
Mr. Qiu had no symptoms. His cancer was discovered only because the hospital was testing an artificial-intelligence system that scans routine CT images for early signs of pancreatic tumors.
The tool flagged his scan automatically, prompting doctors to intervene before the disease advanced.
The system is part of a broader push by Chinese hospitals and technology firms to apply AI to illnesses that have long resisted early diagnosis.
Pancreatic cancer is among the deadliest malignancies, with a five-year survival rate of roughly 10 percent, largely because symptoms usually appear only at advanced stages.
Solving the imaging dilemma
Confirming pancreatic cancer typically requires contrast CT scans, which expose patients to high levels of radiation and are unsuitable for mass screening.
Lower-radiation alternatives, such as noncontrast CT scans, are widely used in China but produce images that are harder for radiologists to interpret.
Artificial intelligence may bridge that gap.
The system used at the Affiliated People’s Hospital of Ningbo University is called PANDA, short for “pancreatic cancer detection with artificial intelligence.”
Developed by researchers affiliated with Alibaba, PANDA was designed to detect cancer using noncontrast CT scans. Doctors at the Ningbo hospital began using it in a clinical trial in November 2024.
Early results show promise
Since its rollout, PANDA has analyzed more than 180,000 abdominal and chest CT scans. According to Dr. Zhu Kelei, it helped identify around two dozen pancreatic cancer cases, including 14 detected at an early stage.
The tool flagged 20 cases of intraductal adenocarcinoma, the most common and lethal form of pancreatic cancer. Qiu’s tumor was a rarer and less aggressive neuroendocrine type.
All of the patients initially presented with vague complaints such as bloating or nausea and were not being evaluated by pancreatic specialists.
“I think you can 100 percent say A.I. saved their lives,” Dr. Zhu said.
Training AI to see the invisible
Even PANDA’s creators initially doubted whether noncontrast CTs could provide sufficient detail.
To overcome this, engineers first had radiologists manually label tumor locations on contrast CT scans from more than 2,000 pancreatic cancer patients. They then digitally mapped those lesions onto the patients’ noncontrast scans, allowing the algorithm to learn how cancer appears even in less detailed images.
When tested on more than 20,000 noncontrast CT scans, PANDA correctly identified 93 percent of patients with pancreatic lesions, according to a 2023 study published in Nature Medicine.
“The effectiveness actually surprised us,” said Ling Zhang, a senior algorithm engineer at Alibaba’s Damo Academy.
Regulatory momentum and ongoing trials
In April, Alibaba announced that the US Food and Drug Administration had granted PANDA “breakthrough device” status, enabling an accelerated review process. Multiple clinical trials are also underway in China.
At the Ningbo hospital, the system analyzes scans already ordered for other medical reasons, adding no extra cost for patients.
Despite encouraging results, researchers and outside experts cautioned that broader data is needed to determine whether early detection benefits outweigh the risks of false alarms.
Dr. Ajit Goenka of the Mayo Clinic warned that many patients may have experienced unnecessary anxiety and invasive follow-up tests after being flagged by the system.
Others noted that PANDA may be more valuable for trainee doctors or hospitals with limited specialist expertise.
“You’re going to have different skill sets at different centers, depending on where in the world you are or the clinical volume,” said Dr. Diane Simeone of the University of California, San Diego.
Operational strain, public trust issues
At Ningbo, the system’s success has introduced logistical challenges. The hospital lacks sufficient staff to contact all flagged patients, and aging computer systems sometimes struggle to process PANDA’s data, occasionally crashing.
There are also social hurdles. In China, past medical corruption scandals have eroded trust, leading some patients to suspect that follow-up requests are financially motivated.
Qiu did not hesitate when doctors advised surgery, even though he said he did not understand how A.I. worked. By November, he said he felt healthy and had returned to tending his farm.
“He said I was very lucky,” Qiu recalled.
“So there was nothing else for me to say. I could only be relieved,” he told The New York Times.