Alibaba's research arm, Damo Academy, has open-sourced a vision-language model that reads contrast-enhanced abdominal CT scans and flags nearly 150 conditions, including cancers. The model, called Damo Radar, was built to analyse 18 abdominal organs and surface a wide range of diseases and abnormalities such as malignant tumours. It was trained on CT scans paired with real clinical reports, the approach that lets it describe findings in clinical language rather than just outputting a label.
The numbers are striking. Across nearly 40,000 real-world examinations, Damo Radar achieved an average area under the curve of 0.913 over 146 clinical findings, where 1.0 represents perfect diagnostic accuracy. In a comparative study involving 26 human radiologists from multiple hospitals, the model's average accuracy beat 23 of those participants. Working alongside it, the radiologists cut missed diagnoses by 10 per cent and reduced the time required by more than 30 per cent. The research was published on Thursday in the journal Science, with institutions including a Zhejiang University affiliated hospital taking part.
The team describes Damo Radar as the world's first expert-level generalist medical imaging model, and says the training method should extend to other imaging types such as ultrasound or MRI. For founders building in health tech, the release matters less as a benchmark and more as infrastructure: a strong abdominal imaging model is now available to build on rather than something each startup has to train from scratch. It follows Alibaba's Coca AI model, released in April, which the company said was more sensitive than radiologists at spotting early-stage colorectal cancer from CT scans.