LLM-Filtered CT Reports Enable Whole-Body Reference Charts for Attenuation and Volume

A large language model (LLM) ensemble filtered pathological findings from >350,000 CT reports to generate whole-body reference charts for CT attenuation and volume of 106 structures, enabling standardized quantitative phenotyping.

• LLM ensemble cross-verification ground-truths findings to reduce pathology bias in the cohort.
• Reference charts account for age, sex, contrast enhancement, and acquisition parameters using GAMLSS.
• Longitudinal analysis revealed changes distinct from cross-sectional trends.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.