Fig. 3: Improved performance on unseen test set when adding 3CB compositional information.

a Adding three-compartment breast (3CB) features to computer-aided detection (CAD) (orange) results in an area under the receiver operating characteristic (ROC) curve (AUC) of 0.81, standard deviation (SD) of 0.03, when compared to CAD alone (blue), AUC of 0.69, SD of 0.04. Mean curves (solid lines) and 95% confidence intervals (shaded regions) were computed via 1000 bootstrap samples. b The integrated sensitivity (IS), black shaded region between solid and dashed lines, indicates the change in sensitivity with the addition of 3CB. The integrated 1-specificity (IP), red shaded region between solid and dashed lines, indicates the change in specificity with the addition of 3CB. The integrated discrimination improvement (IDI) is the sum of the IS and IP (−1.06 + 13.17) which is 12.11 and a positive IDI indicates that predictive models benefit from the addition of 3CB. The borders of the Breast Imaging-Reporting and Data System (BI-RADS) assessment categories are indicated by the vertical dashed lines. Net reclassification improvement (NRI) for events or cancers (black) and non-events or benigns (red) are calculated at each BI-RADS border to demonstrate 3CBs effect on specificity with respect to each BI-RADS category.