SGPNet: A Three-Dimensional Multitask Residual Framework for Segmentation and IDH Genotype Prediction of Gliomas

Table 1

Comparisons of the proposed and other deep learning-based models on cross-referenced TCGA and BRATS datasets.

Method	ACC	SN	SP	AUC	DSC
Liang et al. [12]	0.846	0.785	0.880	0.857	—
Chang et al. [11]	0.857	—	—	0.940	—
3D U-Net [27]	—	—	—	—	0.920
SGPNet (only segment)	—	—	—	—	0.937
SGPNet (multitasks)	0.895	0.907	0.883	0.949	0.935