Alibaba’s DAMO Academy, in collaboration with Zhejiang Cancer Hospital, unveiled GRAPE, an AI model that can detect early–stage stomach (gastric) cancer from routine non-contrast CT scans. This breakthrough pushes detection accuracy well beyond human performance.
Outstanding Accuracy
In clinical trials involving over 70,000 patients, GRAPE achieved 85.1% sensitivity and 96.8% specificity, outperforming radiologists by 21.8% in sensitivity and 14% in specificity
Why It Matters
Early detection elevates the five-year survival rate from roughly 30% to over 90% in China. Traditional CT scans struggle to detect early cancer cells in hollow organs like the stomach—GRAPE addresses this by analyzing 3D images and learning subtle patterns beyond the reach of human eyes
Behind the Tech
The project used the world’s largest multicenter dataset of non-contrast CT scans for gastric cancer, tackling challenges like stomach shape variability and early lesion detection. The model is published in Nature Medicine and already deployed in high-risk regions in China newsminimalist
Real-World Impact
In two hospitals, GRAPE identified cancer in 17.7% and 24.5% of high-risk patients, including many without symptoms. In one case, the AI spotted cancer in a CT taken six months before a clinical diagnosis—meaning earlier treatment would have been possible
Global Potential
By integrating non-invasive CT scans with AI, GRAPE provides a more accessible alternative to endoscopy, which many patients avoid due to discomfort. Plans are in motion to expand its use beyond China to international settings
Conclusion
With the development of GRAPE, Alibaba’s AI detects stomach cancer early, offering a non-invasive, high-accuracy alternative for mass screening. This innovation could transform early cancer detection, leading to timely treatment and saving countless lives.
