TL;DR
Alibaba's Damo Academy has open-sourced Damo Radar, a vision-language model that detects 146 abdominal conditions from CT scans with near-expert accuracy, validated on 40,000 real-world exams and published in Science.
What happened
- Damo Academy, Alibaba's research arm, released Damo Radar as an open-source medical AI model on September 18, 2026.
- The model reads contrast-enhanced CT scans across 18 abdominal organs and flags 146 clinical findings, including malignant tumours.
- Validated on nearly 40,000 real-world examinations, it achieved an average AUC of 0.913 (where 1.0 is perfect diagnostic accuracy).
- At that performance level, the model outperformed most radiologists, according to the peer-reviewed study published in Science.
- Damo Academy calls it "the world's first expert-level generalist medical imaging model" and says the training method can extend to other imaging modalities.
Why it matters
- Open-sourcing at this capability level puts near-radiologist CT analysis in the hands of any hospital, startup, or researcher with compute, dramatically lowering the barrier to clinical AI deployment.
- An AUC of 0.913 across 146 conditions simultaneously is a generalist benchmark: most prior models target one disease or one organ, not 18 organs at once.
- Cancer detection in the mix raises the clinical stakes: early abdominal cancer identification is one of radiology's highest-value and most shortage-constrained tasks globally.
- The Science publication adds independent credibility, making regulatory and procurement conversations easier for health systems considering adoption.
- Alibaba's move pressures Western medical AI incumbents (and closed-model rivals) by commoditizing a capability that has taken years and large datasets to build.
What to watch next
- Whether regulatory bodies (FDA, CE, China NMPA) move to approve or fast-track Damo Radar for clinical use, which would be the real unlock for hospital deployment at scale.
- How quickly the open-source community extends the model to chest, brain, or musculoskeletal imaging, testing Damo Academy's claim that the training method generalises.
- Competitive responses from Google Health, Microsoft, and medical AI pure-plays like Rad AI or Viz.ai, who now face a free, Science-validated baseline eating into their moat.
Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.