Mixed-effects radiomics model robustly predicts lung cancer across multicenter chest CT

In multicenter chest CT data from the LIBRA study, a mixed-effects radiomics model for lung nodule cancer prediction achieved AUC 0.91 (95% CI 0.90–0.93), significantly outperforming ComBat and providing net benefit over volume and Brock score alone.

• Mixed-effects radiomics model achieved AUC 0.91 (95% CI 0.90–0.93) in 10-fold cross-validation, significantly outperforming ComBat (0.75) and fixed-effects (0.88) approaches, p<0.0001 and p=0.003.
• The model was well calibrated (intercept -0.036, slope 0.89) and provided net benefit over volume and Brock score on decision-curve analysis.
• The study was retrospective; prospective validation is needed before clinical adoption.

Automated summary

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