Alibaba DAMO and Shengjing Hospital Unveil Liver Cancer AI Model Detecting 1cm Tumors and 15 Missed Cases

On August 24, Alibaba DAMO Academy, together with Shengjing Hospital of China Medical University and other institutions, developed a liver cancer diagnostic AI model named DAMO LiON, which can identify tiny malignant liver lesions on CT scans. In a two-month real-world prospective clinical trial, the AI model detected 15 malignant tumors that had been missed, most of which were around 1cm in size, helping patients receive timely surgical or drug treatment. The related paper was published in the international academic journal Nature Medicine.

Acting as an "AI safety officer" to assist doctors in reading scans, the DAMO LiON model can accurately identify both primary liver cancer and easily overlooked liver metastases. Experimental results showed that the AI model achieved higher accuracy in identifying malignant tumors than radiologists. When doctors used the AI model to assist in reading scans, reading time decreased by 27%, and sensitivity to malignant tumors increased by 11.5%, effectively reducing missed diagnoses. With AI assistance, junior doctors reached the level of senior doctors.

The research team deployed the AI model in hospitals for daily scan reading. When the AI and the doctor's initial diagnosis differed, the case was referred to a senior radiologist for review, and if necessary, escalated to a multidisciplinary team discussion. Within two months, the AI read enhanced CT scans of more than 10,000 patients, helping doctors identify 15 liver metastases that had been overlooked and prompting changes in treatment plans for these patients.

According to algorithm experts at DAMO Academy, the malignant lesions detected by the AI generally had an average diameter of about 1cm, showed low contrast with liver tissue, or were located in less common anatomical positions. The DAMO LiON model uses an improved network architecture that captures the relationship between lesions and the entire liver while preserving local texture and boundaries, enhancing performance on difficult cases such as fatty liver, cirrhosis, and postoperative livers. Additionally, the AI iteratively fuses images from different phases, capturing pixel-level differences between phases to identify tiny lesions in enhanced CT scans.

Since its establishment in 2017, DAMO Academy has focused on medical AI, developing AI models for pancreatic cancer screening (DAMO PANDA) based on non-contrast CT, gastric cancer screening (DAMO GRAPE), and colorectal cancer screening (DAMO COCA). These achievements have been published three times in Nature Medicine, entered the innovation channel of China's National Medical Products Administration, and received Breakthrough Device designation from the U. S. Food and Drug Administration.