Fig. 4

Manhattan plot shows AUC values of different classes of texture features. We calculated the area under the curve (AUC) estimate from 5-fold cross-validation to evaluate individual texture parameters (TP) as classifiers of low-grade vs. high-grade fibrosis. Among the different classes of texture parameters, the features calculated from a grey level co-occurrence matrix (GLCM) had the highest AUC (green dots). Meanwhile, the AUC of the shape-based features (blue dots) did not reach up to the accuracy of other classes. The solid line highlights AUC = 0.5, where features do not have discriminatory power, the best classifiers exceeded AUC = 0.7 (dotted line)