HANGZHOU, June 26 (Xinhua) -- Chinese researchers have developed an AI model that can spot early signs of stomach cancer in routine CT scans, helping doctors identify cases earlier.
The DAMO GRAPE model, co-developed by the Zhejiang Cancer Hospital and Alibaba’s DAMO Academy in east China’s Zhejiang Province, was tested on more than 70,000 patients, yielding promising results.
The study was published in the journal Nature Medicine on Tuesday.
"This is an interesting finding as such a tool may be able to identify more patients with earlier stage gastric cancer in the future," commented a reviewer of the journal.
Stomach cancer results in 260,000 deaths every year in China. While early detection can increase five-year survival rates from under 30 percent to over 90 percent, many patients avoid invasive endoscopy, and the quality of screening varies among hospitals, said Cheng Xiangdong, Party chief of the hospital.
Zhang Ling, the corresponding author and a senior algorithm expert at DAMO Academy, said that non-contrast CT is a common non-invasive examination used for checkups and outpatient care, as it is both cost-effective and efficient. However, it is traditionally considered unsuitable for scanning hollow organs like the stomach.
The team challenged this conventional belief by building the world’s largest multicenter dataset of gastric non-contrast CT images, overcoming difficulties such as the stomach’s variable shape, internal content interference and early-stage cancer cells limited to the mucosal layer.
"The model achieved 85.1 percent sensitivity and 96.8 percent specificity, outperforming radiologists by 21.8 percent and 14 percent, respectively," said Zhang. "This marks the first time non-contrast CT can detect early gastric cancer."
At two regional hospitals, the model detected cancer in 17.7 percent and 24.5 percent of scanned patients at high risk, including many without symptoms.
The DAMO GRAPE model could fill the gap in opportunistic detection of early gastric cancer during imaging examinations, said Hu Can, the leading author and a gastric surgeon at Zhejiang Cancer Hospital.
During the study, the team reviewed earlier CT scans from 11 patients diagnosed with stomach cancer and found that the AI model detected abnormalities in scans taken two to 10 months before their diagnosis.
One notable case involved a 45-year-old patient diagnosed with late-stage stomach cancer. The AI successfully identified cancerous markers when reanalyzing a CT scan that she had received for another condition six months prior to diagnosis.
"If this tool had been available earlier, the patient could have received treatment much sooner," said Hu.
Cheng said that the AI model has been deployed in high-risk areas for gastric cancer in China, with plans for further expansion at home and abroad.
原文地址：http://en.people.cn/n3/2025/0626/c90000-20333175.html