Fig. 5

ROC curves of the optimized machine learning models. In the first analysis (a) where the liver segments were randomly divided into equal size train and test sets, the random forest classifier (RFC) was able to differentiate between low-grade and high-grade fibrosis with excellent accuracy in the training set (AUC = 0.95, blue line). Its diagnostic ability was only slightly worse in the test set (AUC = 0.90, magenta line). The support vector machine classifier (SVM) achieved very good prediction accuracy in the training set (AUC = 0.88, teal line), and it performed acceptably in the classification of the test set (AUC = 0.76, orange line). In the second analysis (b) segments of 64-slice scans were used for training and segments of 16-slice scans for testing the models. The RFC model achieved very good prediction accuracy in both the training (AUC = 0.84, blue line) and test sets (AUC = 0.88, magenta line). The SVM’s accuracy for the prediction of high-grade fibrosis was excellent in both the training (AUC = 0.91, teal line) and the test set (AUC = 0.90, orange line) (b)