Table 3.

Mean±STD of our GAN method and the deep learning methods are summarized. CNN-Fuse, FusionGAN, and SESF-Fuse will be excluded from the quantitative comparisons as they did not generated fusion images that contain CT bone structures. Bold indicates the best results. Underline indicate a better result than ours that was excluded because it did not satisfy the fusion criteria

Method	ENT	STD	PSNR	$Q^{XY/F}$	MG	SF	NCC	MI	SSIM
CNN-Fuse	2.06± 0.5	0.24± 0.03	46.5± 6.49	0.51± 0.07	0.08± 0.02	0.36± 0.06	0.9± 0.09*	0.87± 0.56	0.66± 0.31
SESF-Fuse	2.06± 0.51	0.23± 0.03	43.04 ± 8.41	0.46± 0.04	0.08± 0.02	0.38± 0.06	0.7± 0.15	0.47± 0.33*	0.58± 0.25*
SwinFusion	2.05± 0.46	0.28± 0.03	22.65± 1.78*	0.6± 0.04	0.07± 0.02	0.34± 0.06	0.89± 0.04	0.23± 0.09	0.67± 0.16
IFCNN	2.11± 0.48	0.24± 0.03	21.54± 1.96	0.6± 0.05	0.08± 0.02	0.36± 0.06	0.9± 0.04	0.03± 0.05	0.78$\mathbf{\pm }$0.04
U2Fusion	2.14± 0.49	0.18± 0.03	18.11± 0.95	0.54± 0.07	0.06± 0.02	0.27± 0.05	0.91$\mathbf{\pm }$0.04*	0.24± 0.07	0.15± 0.06
DSAGAN	2.19± 0.39	0.27± 0.01	28.5$\mathbf{\pm }$3.33	0.4± 0.05	0.13± 0.02	0.58± 0.07	0.86± 0.07	0.66$\mathbf{\pm }$0.54	0.11± 0.05
CU-Net	2.83 ± 0.55	0.19 ± 0.02	21.07 ± 1.84	0.31 ± 0.04	0.04 ± 0.01	0.17 ± 0.03	0.88 ± 0.04	0.29 ± 0.07	0.4 ± 0.09
FusionGAN	1.64± 0.3	0.21± 0.03	25.89± 2.88	0.35± 0.03	0.1± 0.03	0.31± 0.03	0.69± 0.16	1.0± 0.14	0.59± 0.18
Ours	5.2$\mathbf{\pm }$0.38	0.44$\mathbf{\pm }$0.05	23.02$\pm$3.5	0.64$\mathbf{\pm }$0.1	0.20$\mathbf{\pm }$0.05	0.67$\mathbf{\pm }$0.14	0.91$\mathbf{\pm }$0.04	0.42$\pm$0.29	0.62$\pm$0.22

* is not statistically different (p-value $>0.05$) from our proposed MedFusionGAN method

Abbreviations: ENT entropy, STD standard deviation, PSNR peak signal-to-noise ratio, MG mean gradient, SF spatial frequency, NCC normalized cross-correlation, MI mutual information, SSIM structural similarity index