Full Breakdown
Alibaba Open-Sources Damo Radar, a Generalist Medical Imaging AI
By Drooid · · How we work
New AI Model Targets Hundreds of Abdominal Conditions
Alibaba Group Holding’s research division, Damo Academy, has released an open-source artificial-intelligence system named Damo Radar that can examine contrast-enhanced computed tomography (CT) scans of the abdomen and flag nearly 150 distinct clinical findings. The model evaluates images of 18 abdominal organs and is designed to spot a wide spectrum of abnormalities, including malignant tumours. The release marks the latest milestone in the company’s expanding portfolio of medical-AI tools.
Technical Approach and Performance
Damo Radar is a vision-language model trained on a large dataset that pairs CT images with corresponding clinical reports. In testing on roughly 40,000 real-world examinations, the system achieved an average area-under-the-curve (AUC) of 0.913 across 146 identified conditions—a figure that approaches the perfect score of 1.0. The developers note that the training pipeline could be adapted to other imaging modalities, positioning the model as a potential “expert-level generalist” for medical imaging.
Context of Alibaba’s Medical-AI Initiatives
The launch follows a series of Alibaba-backed health-technology projects, including earlier AI models for disease screening and drug discovery. Damo Academy has positioned itself as a bridge between the company’s cloud-computing resources and the healthcare sector, aiming to accelerate research and clinical adoption through open-source collaboration.
Potential Impact on Healthcare Imaging
By making a high-performing, broadly applicable model publicly available, Alibaba hopes to lower barriers for hospitals and research institutions that lack proprietary AI tools. The open-source nature allows external developers to refine the system, extend it to other imaging types, and integrate it into existing diagnostic workflows, potentially improving early detection of abdominal cancers and other conditions.
Official Statements from Damo Academy
The team indicated that future work will explore applying the same training methodology to additional imaging domains.