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

Table 1

Quantitative evaluation of the INbreast dataset (mean ± standard deviation). We compared the measurements of the different synthesis methods over the whole image domain and the tumor domain at a significance level of 0.05, and the underline indicates that Fp^roi-GAN is statistically significantly different from other methods.

Region	Methods	PSNR	SSIM	MS-SSIM
Whole image	DCGAN [11]	16.834 ± 3.28	0.769 ± 0.15	0.879 ± 0.21
	Pix2Pix [15]	17.398 ± 3.81	0.843 ± 0.13	0.923 ± 0.19
	cycleGAN [16]	17.815 ± 5.18	0.829 ± 0.17	0.919 ± 0.18
	Fp^roi-GAN	18.647 ± 3.25	0.882 ± 0.16	0.935 ± 0.15
Tumor region	DCGAN [11]	19.231 ± 7.43	0.872 ± 0.15	0.894 ± 0.23
	Pix2Pix [15]	21.811 ± 6.98	0.915 ± 0.11	0.902 ± 0.22
	cycleGAN [16]	20.485 ± 6.15	0.882 ± 0.07	0.904 ± 0.18
	Fp^roi-GAN	24.142 ± 6.70	0.967 ± 0.08	0.946 ± 0.18