Alibaba's damo lion AI model detects small liver tumors with high accuracy

Alibaba's DAMO Research Institute, in collaboration with Shengjing Hospital of China Medical University, has developed an AI model called DAMO LiON that can accurately identify small liver tumors. The model was tested in a two-month real-world prospective clinical trial, where it detected 15 previously missed malignant tumors, most of which were approximately 1 centimeter in size. These findings helped patients receive timely surgical or drug treatment. The results were published in the prestigious journal Nature Medicine. Liver cancer is a common malignancy, with both primary liver cancer and metastatic tumors from other organs such as the colon and pancreas. Early detection is crucial for better treatment outcomes, but it is challenging due to the small size of the tumors and interference from conditions like liver cirrhosis, fatty liver, and complex anatomical structures. Radiologists may also overlook small lesions, especially in cases of metastasis, where attention is focused on the primary tumor. To address these challenges, the DAMO Research Institute leveraged its long-term expertise in medical imaging AI to develop DAMO LiON. The model acts as an 'AI safety officer' to assist radiologists in interpreting CT scans, accurately identifying both primary liver cancer and often-overlooked metastases. The AI model demonstrated higher accuracy in detecting malignant tumors compared to radiologists. When used alongside AI, the time required for diagnosis was reduced by 27%, and the sensitivity for detecting malignant tumors increased by 11.5%, significantly reducing the risk of missed diagnoses. Junior doctors were also able to achieve the performance level of senior doctors with AI assistance. The model was deployed in hospitals for routine CT image analysis. When AI and radiologists' initial assessments differed, cases were reviewed by senior radiologists, and in complex cases, multidisciplinary teams (MDT) were involved. Over two months, the AI analyzed CT scans from over 10,000 patients, helping to identify 15 previously overlooked liver metastases, leading to changes in treatment plans. One notable case involved a 65-year-old patient with a history of bladder cancer. Although liver function and tumor markers were normal, the AI correctly identified liver metastasis, prompting the doctor to revise the report and recommend chemotherapy. Yan Ke, an algorithm expert at DAMO Research Institute, explained that the tumors detected by AI were typically small, faint, and located in less common anatomical positions. The DAMO LiON model uses an improved network architecture to capture the relationship between lesions and the entire liver while preserving local texture and boundaries, enhancing performance in challenging cases such as fatty liver, cirrhosis, and post-surgical livers. The AI also integrates images from different phases to detect subtle differences at the pixel level, enabling the identification of small tumors that may be missed in standard CT scans. Since 2017, DAMO Research Institute has been actively developing AI for medical imaging, pioneering the use of AI to detect subtle lesions invisible to the human eye. Previous models include DAMO PANDA for pancreatic cancer, DAMO GRAPE for gastric cancer, and DAMO COCA for colorectal cancer. These models have been published in Nature Medicine three times, entered the National Medical Products Administration's innovation channel, and received FDA's 'Breakthrough Device' designation. This latest model marks an expansion of DAMO's medical AI capabilities from screening to diagnosis, offering more patients timely and accurate treatment options.