Learning rich features with hybrid loss for brain tumor segmentation

Table 3 Performance comparison of different single models on the test set

Method	DSC			PPV			Sensitivity
Method	Complete	Core	Enh	Complete	Core	Enh	Complete	Core	Enh
DeepMedic [12]	0.836	0.674	0.629	0.823	0.846	0.64	0.885	0.616	0.656
DeepMedic + CRF [12]	0.847	0.67	0.629	0.85	0.848	0.634	0.876	0.607	0.662
FCNN + CRF (axial) [13]	0.78	0.64	0.54	0.78	0.76	0.48	0.81	0.62	0.71
FCNN + CRF (coronal) [13]	0.77	0.66	0.56	0.73	0.73	0.52	0.86	0.67	0.67
FCNN + CRF (sagittal) [13]	0.76	0.63	0.47	0.75	0.71	0.38	0.80	0.63	0.75
FCNN + 3D CRF (axial) [13]	0.84	0.72	0.62	0.88	0.75	0.62	0.82	0.76	0.67
FCNN + 3D CRF(coronal) [13]	0.84	0.72	0.62	0.88	0.75	0.62	0.82	0.75	0.66
FCNN + 3D CRF (sagittal) [13]	0.82	0.72	0.60	0.88	0.75	0.59	0.81	0.76	0.67
Proposed + \(\mathrm{H}{L}_{cb\_\_rl\_loss}\)	0.85	0.71	0.59	0.85	0.75	0.58	0.87	0.74	0.65
Proposed + \(\mathrm{H}{L}_{ce\_\_rl\_loss}\)	0.85	0.71	0.60	0.83	0.76	0.57	0.89	0.73	0.68

ISSN: 1472-6947