Historically, the application of AI in medical imaging has progressed at a sluggish pace, predominantly constrained to specialized disease models that lack efficiency. Alibaba DAMO Academy, in partnership with the First Affiliated Hospital of Zhejiang University and other esteemed institutions, has successfully developed DAMO RADAR, the world’s first expert-level general imaging AI model. This groundbreaking model is capable of identifying 146 distinct conditions across 18 abdominal anatomical structures. Following rigorous validation processes, DAMO RADAR has exhibited outstanding AUC (Area Under the Curve) performance in diverse scenarios, outperforming the accuracy of most radiologists. Moreover, it significantly enhances doctors’ sensitivity in interpreting images and reduces the time required for diagnosis.

DAMO RADAR leverages visual-language contrastive learning technology, achieving pivotal breakthroughs such as organ-level fine-grained alignment and adaptive contrastive modeling. This innovation eliminates the necessity for additional slice-by-slice annotation, streamlining the diagnostic process. The model not only addresses diagnostic blind spots for doctors but also functions as an educational tool, facilitating knowledge transfer and skill enhancement.

Currently, the model, along with its code and framework, has been made open-source. This move underscores the feasibility of the general medical imaging AI approach and hints at its vast potential for future expansion into other high-precision medical scenarios.