Fig. 4From: Preselection of robust radiomic features does not improve outcome modelling in non-small cell lung cancer based on clinical routine FDG-PET imagingROC curves. Receiver operating characteristic (ROC) curves with 95%-CI for a 12-month EFS, b 12-month OS and c robust 12-month OS for training (dark grey, solid lines) and validation cohorts (light grey, interrupted lines) are shown. Only the 12-month EFS model trained on the entire feature set without robust preselection was successful in the validation cohortBack to article page