Expert-committee AI framework boosts CT classification of four lung nodule subtypes across multicenter cohorts

An expert-committee AI framework using quantitative CT features achieved AUCs of 0.918 for AAH/AIS, 0.860 for minimally invasive adenocarcinoma, and 0.913 for invasive adenocarcinoma and inflammatory nodules in a 491-nodule internal test; external validation showed consistent pe…

• The adaptive framework improved sensitivity for heterogeneous subtypes (MIA and inflammatory nodules) versus conventional single-model approaches in external multicenter validation.
• Candidate classifiers were pooled and an arbitration mechanism selected models based on subtype-specific validation performance.

Automated summary

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