Scale-adaptive Category-ComBat harmonization boosts multicenter CT radiomics performance

Category-ComBat-ω, a scale-adaptive harmonization for multicenter CT radiomics, raised mean AUC from 0.84 (original) and 0.82 (ComBat) to 0.89 for nodule classification. Tunable ω compressed feature spread without affecting means.

• Category-ComBat separated systematic (scanner, protocol, kernel) from random variability and introduced a distribution coefficient ω to adjust scale orthogonally.
• Best feature reproducibility and classification performance occurred with ω between 0.5 and 0.9.
• The method preserved benign-malignant discriminative separation while reducing inter-center variation.

Automated summary

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