Table 1. Details and quantitative metrics of networks involved in the ablation experiment.

Network structure	Conditional input	Loss function	PSNR (dB)	NMSE (%)	SSIM	CNR (%)
Low dose	–	–	18.50	1.80	0.862	26.1
U-net	PET	lL1	25.44	0.37	0.976	6.13
U-net	MRI	lL1	21.23	0.96	0.924	3.95
Bi-U-net	PET* & MRI	lL1+Lbias	25.86	0.33	0.976	6.68
c-GAN	PET	lL1+LGAN	29.94	0.15	0.990	0.61
M-c-GAN	PET & MRI	lL1+LGAN	30.03	0.13	0.991	2.99
Bi-c-GAN	PET* & MRI	lL1+Lbias+LGAN	32.07	0.08	0.994	0.67
Bi-c-GAN	PET & MRI*	lL1+Lbias+LGAN	25.74	0.34	0.976	0.8

The best results are marked in bold. *, treated as the primary task’s conditional input. c-GAN, conditional generative adversarial network; M-c-GAN, multiple c-GAN; Bi-c-GAN, bi-task c-GAN; PSNR, peak signal-to-noise ratio; PET, positron emission tomography; MRI, magnetic resonance imaging; NMSE, normalized square error of the mean; SSIM, structural similarity; CNR, contrast noise ratio.