Alibaba’s research arm has released a generalist medical imaging model that reads a single contrast-enhanced abdominal CT scan for 146 clinical findings, then published the results in Science.
Damo Radar was trained on more than 420,000 abdominal CT exams and 15 million anatomy-focused image-text pairs. Across roughly 40,000 real-world scans it averaged an AUC of 0.913, and in a head-to-head study it outperformed 23 of 26 expert radiologists.
Accuracy gains carried into everyday reading. Radiologists working with the model missed 10% fewer findings, and moved through scans more than 30% faster.
Alibaba’s team bills Damo Radar as the first generalist imaging model to reach expert level. It expects the same recipe to carry over to other scan types.
Alibaba posted the weights, code and training framework on GitHub and Hugging Face. The code carries an Apache 2.0 license, while the weights are released for research use only.
That release matters commercially. Radiology AI has largely been sold as narrow per-scan detectors, and researchers project a shortfall of 750,000 radiologists by 2050. A free generalist that clears expert accuracy across more than a hundred conditions puts direct pressure on that pricing model, though the model has no US FDA clearance yet.
