An expert-level generalist AI for abdominal CT diagnosis.

RADAR, a generalist AI trained on over 400,000 abdominal CTs, achieves expert-level diagnosis across 18 structures and 146 findings, with high accuracy and robustness.

• Why it matters: Current supervised AI methods in radiology are limited in scope, restricting their ability to support diverse clinical tasks and improve diagnostic accuracy in complex cases.
• What they did: The team developed RADAR, a vision-language model trained on clinical reports and millions of image-text pairs without manual annotation, evaluated across multiple centers and scenarios.
• The result: RADAR improved radiologists’ sensitivity by ~10%, demonstrating its potential as a scalable, versatile, and interpretable tool that can match human expertise in complex abdominal CT diagnosis.