Fp^roi-GAN with Fused Regional Features for the Synthesis of High-Quality Paired Medical Images

Table 4

Results of the quantitative evaluation of the ROI images of the BRATS 2017 dataset (mean ± standard deviation) with a significance level of 0.05; underline indicates that the Fp^roi-GAN is statistically significantly different from other methods.

Data	Methods	Dice coefficient	Hausdorff distance
HGG	DCGAN [11]	0.808 ± 0.29	8.36 ± 5.66
	Pix2Pix [15]	0.876 ± 0.23	7.54 ± 5.90
	cycleGAN [16]	0.931 ± 0.18	5.15 ± 3.03
	Fp^roi-GAN	0.936 ± 0.18	4.92 ± 3.22
LGG	DCGAN [11]	0.889 ± 0.26	7.83 ± 4.84
	Pix2Pix [15]	0.947 ± 0.23	6.25 ± 3.12
	cycleGAN [16]	0.984 ± 0.21	4.66 ± 2.33
	Fp^roi-GAN	0.990 ± 0.25	4.08 ± 2.79