Table 2.

Test set performance of U-Net baseline, QSMGAN and non-learning-based algorithms.

Methods	L1 error (1e-3)	PSNR	NMSE	HFEN	SSIM
TKD	2.826±0.178	38.82±1.69	0.496±0.076	99.84±4.86	0.806±0.023
MEDI	2.909±0.194	41.24±1.71	0.539±0.059	100.99±5.02	0.912±0.027
iLSQR	2.193±0.227	42.03±1.45	0.410±0.088	74.40±7.15	0.896±0.025
3D U-Net 64-->48	1.309±0.210	43.53±1.31	0.229±0.065	48.45±8.30	0.944±0.018
QSMGAN 64-->48	1.199±0.215	44.16±1.42	0.200±0.065	45.68±8.53	0.952±0.018