Table 3.

Performance comparison of DL algorithms used for nonhealing burn segmentation from MSI images

Algorithm	PAR-AUC	Sensitivity (%)	PPV (%)	Specificity (%)	NPV (%)
U-Net	0.167	6.6 (0.5, 51.2)	86.5 (70.9, 94.4)	100 (99.9, 100)	100 (99.9, 100)
SegNet	0.425	65.6 (30.4, 89.3)	34.7 (3.7, 88.1)	100 (99.9, 100)	100 (99.9, 100)
dFCN	0.561	43.5 (19.9, 70.5)	80.1 (42.3, 95.7)	100 (99.9, 100)	100 (99.9, 100)
Voting Ensemble	0.812	80.5 (60.6, 91.7)	96.7 (80.9, 99.5)	100 (99.9, 100)	100 (99.9, 100)

dFCN, Dilated fully connected neural network; NPV, negative predictive value; PAR-AUC, precision and recall curve; PPV, positive predictive value.