Is (N)PRI suitable for evaluating automated segmentation of cutaneous lesions?

► Simplified PRI expression for image segmentation into 2 classes (lesion/healthy skin). ► Misclassified pixels worsen (decrease) PRI if <50%, but improve (increase) it if >50%. ► NPRI (linear rescaling of PRI) suffers from same shortcoming. ► (N)PRI not better than XOR metric if misclassification is low, strictly worse if high.