Alibaba open-sources Damo Radar, a CT-scan AI that spots nearly 150 abdominal conditions

Alibaba open-sources a generalist CT reader

Alibaba's research arm, Damo Academy, has released an open-source AI model that reads abdominal CT scans and identifies close to 150 different conditions, including cancers. According to the South China Morning Post, the model — called Damo Radar — is a vision-language system built to analyse contrast-enhanced CT scans across 18 abdominal organs, picking up a wide range of diseases and abnormalities, among them malignant tumours.

How the model was built

Rather than being engineered to detect a single disease, Damo Radar was trained on CT scans paired with the clinical reports written alongside them. Learning from that image-plus-text pairing is what allows one model to cover many findings instead of a narrow task, and the research team says the same training approach could eventually be extended to other types of medical imaging.

The reported performance

In nearly 40,000 real-world examinations, the model posted an average area under the curve (AUC) of 0.913 across 146 clinical findings, the institute told the South China Morning Post. An AUC of 1.0 represents perfect diagnostic accuracy, so the reported score is high but short of perfect — and because it is an average across many findings, performance on any individual condition may differ from that headline number. These figures come from the team itself; the report does not include an independent clinical evaluation.

The researchers describe Damo Radar as "the world's first expert-level generalist medical imaging model". The "first" and "expert-level" portions of that claim are difficult to verify, since they depend on which benchmarks were used and which human specialists the comparison was measured against.

Part of a broader medical AI push

The South China Morning Post frames the release as the latest step in Alibaba's expanding medical AI programme. Open-sourcing the model lowers the barrier for hospitals, universities and other labs to test it against their own patient data — the step that ultimately determines whether a diagnostic model holds up beyond the data it was trained on.

Why it matters

Most medical imaging AI to date has been narrow: one model per task, validated for one indication. A generalist model spanning 18 organs and 146 findings points in a different direction — a single system that could assist with routine reads, help prioritise worklists, or flag findings that are easy to miss. Open weights make those claims testable by outsiders rather than taken on trust.

The caveats are just as real. A self-reported AUC measured retrospectively is not the same as prospective clinical validation, and any use in actual patient care would face regulatory review before a scan-reading model could carry diagnostic weight. Abdominal CT also involves contrast material and varies across scanners, protocols and patient populations, all of which can shift real-world performance.

Even so, the combination of broad anatomical coverage, report-supervised training and an open licence makes Damo Radar a notable entry in medical AI, and part of a wider pattern of large research labs releasing healthcare models as open artefacts rather than keeping them locked inside commercial products.