(Yicai) Sept. 18 -- Alibaba Group Holding’s DAMO Academy has unveiled an artificial intelligence model that can identify nearly 150 abdominal conditions, such as stomach and liver cancers, potentially replacing multiple disease-specific models with a single system.

The pioneering model, known as DAMO RADAR, has been open-sourced and covers a broad range of abdominal findings, including malignant tumors, the Hangzhou-based research institute said today. The findings were published in Science.

Founded by the Chinese e-commerce giant in 2017, DAMO Academy conducts research into AI and other emerging technologies. Its achievements include the Tongyi family of AI models, XuanTie processors and Hanguang 800 AI chip, as well as AI systems for detecting cancers.

DAMO RADAR could transform the detection of conditions such as pancreatic cancer, fatty liver disease, and acute appendicitis. The academy said that abdominal CT scans are among the most complex medical images, while existing AI models typically target individual diseases and struggle with broader clinical needs.

To overcome the challenge, the team used vision-language learning to link medical images with reports and identify a broad range of clinical findings. The team also converted CT images into three-dimensional anatomical units, precisely aligning them with medical reports to improve performance.

Researchers evaluated DAMO RADAR on 146 clinical findings involving 18 anatomical structures, with the model achieving a mean area under the curve (AUC) of 0.913 across nearly 40,000 real-world examinations. AUC measures how well a model distinguishes between positive and negative cases, with a score of one indicating perfect performance.

The research team also compared the model with 26 radiologists, with the model’s average performance exceeding that of 23 of them. The study also showed that DAMO RADAR could help radiologists increase disease-detection sensitivity by 10 percent while reducing the time needed by 30 percent.

DAMO RADAR has strong generalization capabilities and broad applicability, said Zhang Ling, a senior algorithm expert at DAMO Academy. The research approach could potentially be extended beyond identifying multiple findings in abdominal CT scans to other forms of medical imaging, accelerating the development of artificial general intelligence, he added.

Editor: Emmi Laine