Alibaba DAMO Academy Launches DAMO LiON, a Liver Cancer AI That Spots 1cm Tumors

Alibaba DAMO Academy, together with Shengjing Hospital of China Medical University and other institutions, has launched DAMO LiON, a liver cancer diagnosis AI model that identifies tiny malignant lesions on contrast-enhanced CT scans. The findings were published in Nature Medicine.

In a two-month prospective real-world clinical trial, DAMO LiON flagged 15 malignant tumors that radiologists had originally missed — most around 1 cm in size — enabling patients to receive timely surgery or drug treatment.

Liver malignancies include primary liver cancer and the often more common liver metastases that spread from colorectal, pancreatic, and other sites. For both, earlier detection means better outcomes. But small lesions are easily overlooked under enhanced CT because of cirrhosis, fatty liver, and complex anatomy — and in metastasis cases, the radiologist's attention is often drawn to the primary tumor.

Positioned as an "AI safety officer" beside the radiologist, the model identified malignant tumors with higher accuracy than radiologists. With AI assistance, reading time dropped 27% and sensitivity for malignant tumors rose 11.5%, cutting missed diagnoses and bringing junior doctors up to senior level.

The team then deployed the model in real hospital reading workflows: when AI and the initial physician diagnosis disagree, the case goes to a senior radiologist for review and, if necessary, an MDT discussion. Over two months, the AI read contrast-enhanced CT scans of more than 10,000 patients, helped catch 15 overlooked liver metastases, and changed treatment plans for those patients.

In one case, a 65-year-old bladder cancer patient with normal liver function and tumor markers was initially diagnosed with only a calcified lesion — the AI correctly identified liver metastasis. DAMO algorithm expert Yan Ke said the lesions the AI caught are typically "small, faint, off-center": around 1 cm in diameter, low contrast with surrounding tissue, or located in uncommon anatomical positions.

DAMO LiON uses an improved network architecture that captures the relationship between the lesion and the whole liver while preserving local texture and boundaries, boosting performance on difficult cases such as fatty liver, cirrhosis, and post-surgical livers. It also iteratively fuses multi-phase imaging to detect pixel-level differences between phases.

DAMO Academy has worked on medical imaging AI since 2017, previously releasing DAMO PANDA for pancreatic cancer screening on plain CT, DAMO GRAPE for gastric cancer, and DAMO COCA for colorectal cancer. Its models have appeared in Nature Medicine three times, entered the NMPA innovation channel, and received FDA Breakthrough Device designation.

Watch next: whether DAMO LiON expands its hospital deployment and advances through regulatory approval, and whether DAMO's medical AI pipeline can replicate this real-world validation playbook across more cancer types.