HANGZHOU, July 1 (Xinhua) -- In a landmark development for medical technology, Chinese researchers have unveiled an artificial intelligence (AI) system capable of detecting early-stage stomach cancer through standard CT scans, a breakthrough that could transform cancer detection worldwide.
The breakthrough comes as Chinese scientists and tech firms push to take a global lead in AI-powered healthcare, with far-reaching implications for millions at risk of the deadly disease.
The GRAPE model, developed through a collaboration of Zhejiang Cancer Hospital and Alibaba's DAMO Academy, represents a significant leap forward in early cancer detection.
Published last week in the journal Nature Medicine, the study demonstrates how AI can identify stomach tumors from non-contrast CT imaging, a method previously considered relatively ineffective for gastric cancer screening.
Gastric cancer remains one of the most lethal malignancies, particularly in Asia. According to the National Cancer Center of China, the country records 360,000 new cases and 260,000 deaths annually, with late-stage diagnoses contributing to a dismal five-year survival rate below 30 percent.
In contrast, patients diagnosed at stage one have survival odds exceeding 90 percent. However, traditional screening methods depend on invasive endoscopies, which are expensive, uncomfortable and often deter participation.
The GRAPE addresses these challenges by turning routine CT scans, already widely used in hospitals and clinics, into powerful screening tools.
"Identifying gastric cancer through AI algorithms is highly challenging due to the significant variations in the stomach's distension and shape," said Xia Yingda, an algorithm expert from DAMO Academy Medical AI Team.
Trained on nearly 100,000 scans from 20 medical centers across China, the AI achieved 85.1 percent sensitivity in spotting early tumors, outperforming human radiologists by a substantial margin.
"We registered contrast-enhanced CT scans to non-contrast CT, which enabled the AI to gain intense precision information, accurately identifying gastric cancer locations," explained Xia, the paper's co-first author.
In real-world trials at two hospitals, the system boosted detection rates to as high as 24.5 percent. Notably, approximately 40 percent of the detected cases were stomach cancer patients without symptoms, according to the study.
"The AI model makes image-based gastric cancer screening possible for the first time," said Cheng Xiangdong, a lead researcher from Zhejiang Cancer Hospital.
"After all, even the most advanced and precise surgical treatments cannot match the effectiveness of early diagnosis and treatment," Cheng added.
The research team retrospectively analyzed the CT images of 11 patients before their gastric cancer diagnosis and found that the AI model could detect gastric cancer 2 to 10 months earlier.
A 45-year-old patient diagnosed with locally advanced gastric cancer had undergone a non-contrast CT scan six months earlier for another disease without gastrointestinal symptoms, and the AI model was shown to be capable of detecting signs of gastric cancer at that time.
The GRAPE has already been deployed at hospitals in east China's Zhejiang and Anhui Province, with plans for broader promotion both domestically and internationally, Cheng said.
The breakthrough followed China's earlier success with PANDA, an AI model designed to detect pancreatic cancer, one of medicine's most elusive killers, through ordinary CT imaging.
PANDA demonstrated 92.9 percent sensitivity and 99.9 percent specificity for cancer detection in clinical trials. Its adoption could be particularly transformative in low-income regions where advanced screening methods are unavailable, slashing costs substantially.
MORE DEPLOYMENT
An increasing number of Chinese researchers are harnessing AI to tackle critical health challenges, with systems already in use at top institutions such as Peking Union Medical College Hospital and Shanghai's Ruijin Hospital.
The former uses AI to screen for cognitive impairments related to stroke, Alzheimer's and Parkinson's, while the latter -- RuiPath, developed by Huawei -- analyzes pathological slides to pinpoint lesion areas, cutting diagnosis time for a single slide to just seconds.
In April, a team from the Cancer Hospital under the Chinese Academy of Medical Sciences reported in Nature Communications their use of an AI algorithm to enhance cervical cancer screening accuracy equal to or beyond that of human experts.
The system has proven especially valuable in assisting junior pathologists, helping to reduce diagnostic errors in regions with limited access to specialists.
These innovations reflect China's strategic investment in advancing AI healthcare solutions. DAMO's pilot programs have achieved "one scan for multiple checks" through a single non-contrast CT scan, uncovering early-stage cancers such as pancreatic, gastric, esophageal, colorectal and liver cancer, chronic diseases like osteoporosis and fatty liver, and acute conditions like aortic syndrome.
DAMO PANDA has been officially designated as a breakthrough device by the US Food and Drug Administration this April.
Using DAMO PANDA, the People's Hospital Affiliated with Ningbo University has screened over 40,000 people, detecting two early pancreatic cancer cases missed by routine tests, including a 1.5-centimeter lesion that was subsequently treated surgically.
"AI is giving medicine a new set of eyes," said Xia. "Where humans might overlook subtle signs in complex scans, these models provide consistent, tireless analysis, and they're only getting better." ■