- DAMO LiON can detect liver tumors as small as 1 centimeter in CT scans
- The model is designed to catch subtle cancers doctors may miss, cutting reading time by 27%
Alibaba Group’s DAMO Academy has launched an AI model that can identify liver tumors as small as 1 centimeter in enhanced CT scans, targeting one of the hardest challenges in early cancer detection: lesions that are too small or subtle to stand out.
Developed with Shengjing Hospital of China Medical University in northwestern China’s Shenyang and other institutions, DAMO LiON detected 15 previously missed malignant tumors during a two-month prospective real-world clinical trial.
Most were around 1 centimeter in diameter, according to Alibaba.
The model is designed to work as an “AI safety officer” alongside radiologists, rather than replace them.
It can identify primary liver cancer while also focusing on liver metastases, which can be particularly easy to overlook when doctors are concentrating on a patient’s original tumor, DAMO Academy said in a statement.
Small, faint and easy to miss
Many of the lesions detected by DAMO LiON were characterized by researchers as “small, faint and unusual.”
They averaged about 1 centimeter, had low contrast with surrounding liver tissue or appeared in less common anatomical locations, said Yan Ke (闫轲), an algorithms expert at the academy.
That makes them difficult to spot during routine image reading, particularly in patients whose attention is already drawn to a known primary tumor elsewhere in the body.
In testing, the AI model achieved higher accuracy in detecting malignant tumors than radiologists. With AI assistance, doctors reduced image-reading time by 27% while improving sensitivity to malignant tumors by 11.5%, Alibaba said.
Building a cancer-screening suite
The launch is the latest step in DAMO Academy’s push into AI-assisted cancer screening.
Since April 2025, the research arm has released AI models targeting four major cancers — pancreatic, gastric, colorectal and liver cancer.
Related research has been published three times in Nature Medicine and has received the US Food and Drug Administration’s Breakthrough Device designation.
The broader goal is to build a multi-cancer screening system combining routine CT scans with AI, potentially allowing hospitals to identify suspicious lesions without requiring additional specialized imaging.
Why it matters globally
For international healthcare systems facing rising cancer burdens and shortages of specialist radiologists, the approach points to a broader role for AI in medical imaging: not replacing doctors, but acting as a second set of eyes for the tiny abnormalities most likely to be missed.
