Three days later, Qiu Sijun, a retired mason from eastern China, ChinaWhile on his way to a routine check-up for his diabetes, he received a call from a doctor he had never met before. The doctor, head of the hospital’s pancreas department, wanted him to come back for follow-up.
“I knew it couldn’t be a good thing,” Qiu, 57, recalled.
He was partly right. The bad news was that Qiu had Cancer of the pancreas. But there was also good news: the tumor was detected early. The doctor, Zhu Kelei, managed to remove it.
This was only possible thanks to a new AI-based tool the hospital was testing, which detected Qiu’s routine scan before he even showed symptoms. The tool is an example of how Chinese tech companies and hospitals are racing to apply AI to some of medicine’s most persistent problems.
Pancreatic cancer is one of the deadliest forms of cancer, with a five-year survival rate of around 10%, largely because early detection is very difficult. Symptoms usually only appear when the cancer is in an advanced stage.
Tests used to confirm its presence, such as contrast-enhanced CT scans, involve large amounts of radiation – which is why many experts advise against large-scale screening. But alternatives with less radiation, such as non-contrast CT scans — in which no contrast material is injected into the patient’s bloodstream — produce less well-defined images, making it difficult for radiologists to identify abnormalities.
THE artificial intelligence can change that. Dr. Zhu’s hospital tool, developed by researchers affiliated with Chinese tech giant Alibaba, was trained to look for pancreatic cancer on non-contrast CT scans.
The tool is called PANDA, an acronym for “pancreatic cancer detection with artificial intelligence”. At the hospital where Dr. Zhu works at the Affiliated People’s Hospital of Ningbo University in eastern China, doctors began using it in a clinical trial in November 2024.
Since then, the tool has analyzed more than 180,000 CT scans of the abdomen or chest, helping doctors detect about two dozen cases of pancreatic cancer, including 14 at an early stage, Dr. Zhu said. The tool identified 20 cases of intraductal adenocarcinoma, the most common and deadliest type of pancreatic cancer. Qiu had a neuroendocrine tumor, a rarer and less aggressive cancer.
All of these patients had arrived at the hospital with complaints such as bloating or nausea and had not initially seen a pancreas specialist, Dr. Zhu said. Several of their CT scans hadn’t triggered any alerts until they were flagged by the AI tool.
“I think we can say with 100% certainty that AI saved their lives,” he stressed.
In April, Alibaba said the U.S. Food and Drug Administration (FDA) had granted PANDA “breakthrough device” status, meaning its evaluation would be fast-tracked to help bring it to market. The tool is also the subject of several clinical trials in China.
The researchers cautioned that more real-world data was needed to demonstrate whether the tool could detect enough cases at an early stage to offset the risks of false positives and unnecessary testing. Scientists elsewhere are investigating other AI-assisted approaches to early pancreatic cancer detection that focus more specifically on high-risk groups, largely because the prevalence of this cancer is low.
Several experts not involved in the Chinese research said they were skeptical that non-contrast CT scans could offer as much valuable information as other forms of imaging.
Even the engineers behind PANDA initially shared this concern, said Ling Zhang, a senior algorithm engineer at Damo Academy, Alibaba’s research arm, and one of the tool’s creators.
To solve this problem, they had a radiologist manually annotate the contrast-enhanced CT scans of more than 2,000 known pancreatic cancer patients, indicating the location of their lesions. The engineers then algorithmically mapped the lesions highlighted on non-contrast CT scans of the same patients. These non-contrast CT scans were then fed to the AI model so that it could learn to detect possible cancer even in less detailed images.
When the tool was subsequently tested on more than 20,000 non-contrast CT scans, it correctly identified 93% of people with pancreatic damage, according to a study published in Natural medicine in 2023.
“The efficiency really surprised us,” Mr. Zhang said.
At Ningbo Hospital, the system is used to analyze tests that doctors had already ordered for other reasons, so there is no additional cost to the hospital or patients. (In China, many people have regular non-contrast CT scans as part of their annual check-ups; at Ningbo Hospital, a non-contrast CT scan costs about $25, before insurance.)
Dr. Zhu and his team review all tests that the system marks as high risk and, if necessary, call patients for more detailed examinations.
The model still cannot be compared to a pancreas specialist, Dr. Zhu said.
Sometimes it highlights cases of pancreatitis and cannot tell whether a tumor originated in the pancreas or has spread from another organ. Since its launch, the model has issued alerts for about 1,400 tests, but only about 300 of them required follow-up, according to the doctors’ decision.
Dr. Ajit Goenka, a radiologist at Mayo Clinic who studies early diagnosis of pancreatic cancer, said it’s crucial to reduce the number of false alarms. It’s possible that hundreds of people in Ningbo “faced the terror of a possible pancreatic cancer diagnosis, endured unnecessary booster shots, and likely underwent expensive and invasive follow-up tests — only to discover they were healthy,” he wrote in an email.
The tool also might be more useful for doctors in training than experienced specialists, said Dr. Diane Simeone, a pancreatic surgeon at the University of California, San Diego. Some of the tumors identified by the tool during the study of Natural medicine they should have been “very obvious” to well-trained radiologists, even without AI, she said.
But she acknowledged the tool could be a valuable resource for hospitals where there is a shortage of specialists. PANDA is also being tested in a clinic in rural Yunnan province.
“You will have different skill levels in different centers depending on where you are in the world or the clinical volume,” Dr. Simeone said.
In Ningbo, the technology’s apparent success has posed new challenges. The hospital currently does not have enough staff to contact all patients requiring follow-up, Dr. Zhu said. Additionally, your old hardware has difficulty handling the large amount of model data. Several times when Dr. Zhu tried to open PANDA on his computer, it crashed.
Detecting cancer before patients show symptoms can also create its own problems. In China, widespread corruption in the medical field has undermined public trust in doctors. Some people may refuse to follow up, Dr. Zhu said, because they fear the hospital is only trying to make money.
Qiu was not one of them. He didn’t hesitate when Dr. Zhu recommended removing his tumor, even though he later said he had neither used AI nor understood how it worked. During a follow-up doctor’s visit in November, Qiu said he felt perfectly healthy and was busy growing vegetables on the family farm.
“He said I was very lucky,” Qiu said. “Then there was nothing more to say. I could only be relieved.”/Contributed by Siyi Zhao
This content was translated using artificial intelligence tools and reviewed by our editorial team.Learn more in our AI policy.
About The Author
