Fig. 3

Classification metrics for the a–c men and d–f women performed on the test datasets. The Matthews correlation data for a, d KDE, and b, e sigmoid-like KDE feature transformations at different polyp size thresholds and KDE exponents are shown. Receiver operating characteristic (ROC) curves on the test data for c the men and f the women applying the optimal models, the optimal models were those with the highest Matthews correlation: 0 mm polyp size threshold, sigmoid-like KDE-based feature transformation. The Matthews correlation obtained at distinct polyp size threshold without KDE transformation (No - KDE) for the datasets of g the men and h the women are shown