Figure 1.

Radiomic pipeline process. (a) the first step consisted of PET/CT image acquisition (b) the tumor volume of interest is semi-automatically segmented (c) several first order and textural image features are extracted (d) a preprocessing step allowed to eliminate those highly correlated variables or with small variance (e) feature selection methods are applied to reduce the number of predictive variables (f) different machine-learning (ML) classifiers are used to construct the predictive models by using only the training cohort. Hyperparameter adjustment of these ML methods is performed by cross-validation (g) performance evaluation of predictive models by AUC-ROC analysis on the test cohort. PCA = principal component analysis, GLCM = gray-level co-occurrence matrix, GLRLM = gray-level run length matrix, NGLDM = neighboring gray-level dependence, GLSZM = gray-level size zone matrix, AUC = area under the curve, ROC = receiver-operating characteristic.