Table 3.

Performance metrics for segmentation with dermatoscopic datasets.

Method	Sensitivity	Specificity	Dice Coef	Hausdorff distance
ISIC2016
K-means method	0.5422	0.8249	0.5439	9.960
SVM method	0.7229	0.8602	0.6939	8.243
U-Net method without decomposition	0.9708	0.9175	0.9060	5.085
U-Net method with decomposition	0.9562	0.9422	0.9198	4.764
ISIC2017
K-means method	0.5709	0.7734	0.4926	10.567
SVM method	0.7650	0.7576	0.5967	9.388
U-Net method without decomposition	0.8971	0.8969	0.8188	5.392
U-Net method with decomposition	0.9043	0.9076	0.8199	5.338
HAM 10,000
K-means method	0.5500	0.9300	0.6381	6.807
SVM method	0.7256	0.8389	0.6674	8.381
U-Net method without decomposition	0.9542	0.9530	0.9121	4.683
U-Net method with decomposition	0.9569	0.9504	0.9166	4.621