The diagnostic performance of the proposed statistical algorithm. (A-D) The probability of the predicted diagnostic class in the training set of each dataset studied. Gradient background indicates a continually increasing or decreasing likelihood of the diagnostic classes. The abscissa indicates the discriminant score generated using the proposed algorithm. (E-H) Evaluation of the diagnostic performance of the proposed algorithm. The plots are ROC curves for the entire dataset (that is training and test sets combined) since the diagnostic performance of the discriminant score was consistently high in the training and test subsets when assessed separately. Area under the ROC curve (AUC) was non-parametrically estimated using the Wilcoxon method. Insets show the strikingly bimodal distribution of the discriminant scores in the entire (that is training and test subsets combined) datasets. SE, standard error.