Table 3.

Comparison of normalized root-mean-squared error (NRMSE), peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), and perception-based image quality evaluator (PIQE) with other state-of-the-art methods for T1- and T2-weighted (mean ± standard deviation).

Sequence	Method	NRMSE$\downarrow$	PSNR$\uparrow$	SSIM$\uparrow$	PIQE$\downarrow$
T1	64mT	$0.432\pm 0.025$	$21.638\pm 1.148$ dB	$0.856\pm 0.015$	$71.820\pm 1.969$
	GANs	$0.140\pm 0.029$	$31.330\pm 1.381$ dB	$0.948\pm 0.004$	$60.896\pm 1.791$
	CycleGAN	$0.132\pm 0.022$	$31.820\pm 1.262$ dB	$0.948\pm 0.003$	$62.130\pm 1.851$
	U-Net	$0.126\pm 0.009$	$30.683\pm 1.210$ dB	$0.936\pm 0.005$	$61.423\pm 1.729$
	cGAN	$0.125\pm 0.023$	$32.302\pm 1.529$ dB	$0.946\pm 0.004$	$62.479\pm 1.731$
	LoHiResGAN	$\mathbf{0.104}\pm \mathbf{0.007}$	$\mathbf{33.842}\pm \mathbf{1.311}$ dB	$\mathbf{0.959}\pm \mathbf{0.004}$	$\mathbf{55.293}\pm \mathbf{1.476}$
T2	64mT	$0.379\pm 0.026$	$22.289\pm 0.406$ dB	$0.883\pm 0.012$	$69.366\pm 1.999$
	GANs	$0.115\pm 0.012$	$32.319\pm 1.019$ dB	$0.954\pm 0.003$	$54.532\pm 1.261$
	CycleGAN	$0.112\pm 0.011$	$32.457\pm 0.885$ dB	$0.948\pm 0.003$	$56.773\pm 1.242$
	U-Net	$0.131\pm 0.010$	$31.558\pm 0.542$ dB	$0.941\pm 0.004$	$57.185\pm 1.539$
	cGAN	$0.113\pm 0.011$	$32.977\pm 0.967$ dB	$0.954\pm 0.003$	$56.570\pm 1.058$
	LoHiResGAN	$\mathbf{0.110}\pm \mathbf{0.010}$	$\mathbf{33.127}\pm \mathbf{0.824}$ dB	$\mathbf{0.965}\pm \mathbf{0.003}$	$\mathbf{55.336}\pm \mathbf{1.193}$