Learning rich features with hybrid loss for brain tumor segmentation

Table 1 Performance comparison between our proposed model and U-net

Method	Loss function	DSC			PPV			Sensitivity
Method	Loss function	Complete	Core	Enh	Complete	Core	Enh	Complete	Core	Enh
U-net	Cross entropy loss	0.84	0.66	0.58	0.87	0.80	0.60	0.84	0.61	0.61
proposed	Cross entropy loss	0.85	0.67	0.59	0.87	0.83	0.60	0.85	0.62	0.64
U-net	Sliced Dice loss	0.83	0.64	0.56	0.87	0.82	0.63	0.82	0.60	0.57
Proposed	Sliced Dice loss	0.84	0.66	0.56	0.86	0.83	0.56	0.85	0.62	0.54
U-net	Combined Dice loss	0.83	0.65	0.55	0.87	0.83	0.65	0.83	0.59	0.54
proposed	Combined Dice loss	0.85	0.69	0.59	0.86	0.82	0.64	0.87	0.65	0.60

ISSN: 1472-6947