LEAVS leverages large language models to automatically extract structured labels from radiology reports for abdominal CT scans, addressing a critical gap in medical image analysis.

• Creates high-quality annotations for 9 abdominal organs with 84.7% average accuracy
• Enables simultaneous detection of various abnormalities across complex abdominal anatomy
• Generates multi-dimensional labels including presence certainty and urgency assessments
• Significantly reduces the manual labeling burden for developing medical vision models

This innovation matters because it extends automated labeling beyond chest scans to the more complex abdominal region, potentially accelerating development of AI diagnostic tools for a wider range of pathologies.

LEAVS: An LLM-based Labeler for Abdominal CT Supervision