Alibaba’s DAMO Academy has open-sourced an artificial intelligence model capable of identifying 146 clinical findings, including several cancers, from abdominal computed tomography scans, reports a Qazinform News Agency correspondent.
Called DAMO RADAR, the vision-language model examines contrast-enhanced CT scans covering 18 abdominal organs, including the liver, pancreas, stomach and colon. It produces confidence scores for each condition, helping radiologists identify areas requiring closer examination.
The system was trained on more than 400,000 CT examinations and 15 million image-text pairs. It learned directly from existing clinical reports without requiring specialists to manually label every scan, according to the project’s official repository. Developers said the model demonstrated “expert-level performance across both routine and complex clinical tasks.”
In a study published in the journal Science, RADAR was evaluated using nearly 40,000 real-world examinations. It achieved an average area under the curve score of 0.913 across 146 findings, with 1.0 representing perfect discrimination between positive and negative cases.
During a separate reader study involving 26 radiologists, the model outperformed 23 participants on average. When used as an assistant, it helped reduce missed diagnoses by around 10% and cut interpretation time by more than 30%.
The model’s weights, training tools and inference code are publicly available, allowing researchers and medical institutions to run it locally instead of sending patient information to external cloud services. A single compatible graphics processor can operate the demonstration version.
However, DAMO Academy stressed that RADAR is “currently intended for research purposes only.” Its developers said further improvements and prospective clinical studies are needed before direct clinical deployment.
Earlier, Qazinform News Agency reported that China’s Spirit AI is developing robot software trained on human activity for wider use in manufacturing, including battery production.