Fig. 2

General overview of the MRI protocol and of the methods used for region-of-interest segmentation on the conventional anatomical whole-body MRI, spinal dynamic contrast-enhanced MRI, and spinal diffusion-weighted MRI sequences, for feature extraction, for feature selection, for statistical model building and for testing the models’ performances. In general, models are tested using receiver operating characteristic curve analysis including all MRI features and separate models are retested on the dataset using only the top three most predictive MRI features (in the final model with the three most prevalent features, generalizability can be reduced due to lack of external testing). AUC area-under-the-curve, b0-b1000 diffusion sensitizing gradients, DCE dynamic contrast-enhanced, DWI diffusion-weighted imaging, LASSO least absolute shrinkage and selection operator, MRI magnetic resonance imaging, NPV negative predictive value, PPV positive predictive value, PR-AUC precision-recall area-under-the-curve, ROC receiver operating characteristic, ROI region-of-interest, sens. sensitivity, spec. specificity, WB whole-body