Fig. 3.

Conceptual visualization illustrating how predictions are evaluated. The reference standard contains two definite and one possible CMB. The model predicts four lesion locations and confidences. The panel labeled ‘Outcome’ details how individual predictions are scored. Only the highest prediction within the mask of a definite CMB counts towards the TPs. Additional predictions as well as predictions that fall within the mask of possible CMBs are disregarded for evaluation purposes. D: definite CMB, P: Possible CMB, TP: true positive, FP: false positive, FN: false negative, ign: disregarded during evaluation.