Fig. 4: Performance plots for ALK prediction on the test sets using the train-finetune model.
From: Predicting ROS1 and ALK fusions in NSCLC from H&E slides with a two-step vision transformer approach
A Receiver Operating Characteristic (ROC) Curves: This panel displays the ROC curves for each cross-validation fold, with individual ROC AUC values listed in the legend. The average ROC AUC (0.85) ± standard deviation (0.02) across all folds is shown at the top of the plot. B Precision-Recall Curves: This panel presents the precision-recall curves for each fold. Precision corresponds to Positive Predictive Value (PPV), and recall corresponds to Positive Percent Agreement (PPA). C Normalized Probability Histogram: This panel shows the normalized histogram of predicted probabilities for the test set, distinguishing between positive and negative predictions. D–H Confusion Matrices for Individual Folds: These panels display the confusion matrices for different cross-validation folds (Fold 0 through Fold 4, respectively). In each matrix, true labels are compared to predicted labels. The Positive Percent Agreement (PPA) is indicated at the top of each confusion matrix.
