Alibaba’s DAMO Academy unveils AI for early gastric cancer screening from plain CT in near-100,000-person study

The news

Zhejiang Cancer Hospital (浙江省肿瘤医院) and Alibaba’s DAMO Academy (阿里巴巴达摩院) have announced DAMO GRAPE, an artificial intelligence model that they say is the first to detect early-stage gastric cancer directly from non-contrast CT scans. At a joint press conference, the partners said the system was validated across 20 hospitals in China in a multicenter study involving nearly 100,000 participants and reportedly boosted detection rates of gastric cancer versus standard clinical workflows. The results have been described as appearing in a leading international medical journal, though details of the publication were not immediately disclosed.

Why it matters

Gastric cancer remains a major public health burden in East Asia, especially China, yet early detection is difficult: symptoms are subtle, and the gold standard—endoscopy—is invasive, resource-intensive, and unevenly accessible. If robust, CT-based triage could tap the vast volume of routine abdominal imaging already conducted in health checks, flagging high-risk cases for confirmatory endoscopy and potentially catching malignancies earlier. But can an algorithm trained in Chinese centers generalize across populations, CT vendors, and imaging protocols?

The bigger picture

China has leaned heavily into AI-enabled medical imaging, pairing large hospital networks with tech firms like Alibaba (阿里巴巴) to accelerate clinically oriented models. DAMO Academy (达摩院) has previously reported progress in multi-cancer screening; this gastric cancer milestone is framed as “another step” in that push. Against a backdrop of tightened U.S. export controls on advanced AI chips, Chinese groups have increasingly emphasized healthcare applications where massive datasets and clinical partnerships can offset compute constraints. Any broad deployment of DAMO GRAPE would still require regulatory clearances—such as from China’s National Medical Products Administration—and independent replication.

What to watch

Key next steps include peer-reviewed disclosure of study design and performance metrics, external validation beyond the original 20 hospitals, and head-to-head comparisons with established screening pathways centered on endoscopy. Cost-effectiveness and workflow integration will matter as much as accuracy. Finally, data governance and patient privacy standards will face scrutiny as multi-center AI screening at population scale moves from pilot to practice.