Radiomics-based ML predicts RAS mutation status in colorectal cancer from 18F-FDG PET/CT

A radiomics-based ML model predicted RAS mutation status in colorectal cancer patients with AUC 0.675 (n=90, retrospective, no external validation). Prognostic stratification was comparable to using true RAS status (C-index 0.675 vs 0.685).

• Retrospective study included 90 colorectal cancer patients with pretreatment 18F-FDG PET/CT; ML model developed using 10-fold cross-validation.
• Risk scores from ML-predicted RAS status strongly correlated with true RAS-based scores (r=0.99) and provided similar prognostic separation (log-rank p<0.01).
• Limited by small sample size, potential single-center bias, and lack of external validation; moderate AUC of 0.675 may limit clinical utility.

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