Short-Axis PET Image Quality Improvement by Attention CycleGAN Using Total-Body PET

Table 3

Quantitative comparison of LQPET images from the scanner with 250 mm FOV.

Data	Methods	NRMSE	PSNR (dB)	SSIM
bed1	LQPET	0.257 ± 0.23	30.444 ± 4.22	0.886 ± 0.09
	CycleAGAN	0.225 ± 0.12	31.975 ± 2.76	0.927 ± 0.04
	CycleGAN	0.332 ± 0.21	28.726 ± 2.78	0.885 ± 0.05
	Pix2Pix	0.349 ± 0.29	29.387 ± 2.82	0.892 ± 0.04
	NLM	0.305 ± 0.29	31.063 ± 2.88	0.837 ± 0.06
	BM3D	0.242 ± 0.13	31.160 ± 3.07	0.862 ± 0.04
bed2	LQPET	0.268 ± 0.20	29.602 ± 4.27	0.874 ± 0.05
	CycleAGAN	0.205 ± 0.09	29.920 ± 2.87	0.919 ± 0.03
	CycleGAN	0.218 ± 0.10	30.919 ± 3.09	0.908 ± 0.03
	Pix2Pix	0.207 ± 0.08	30.119 ± 2.81	0.906 ± 0.03
	NLM	0.273 ± 0.21	29.708 ± 4.11	0.861 ± 0.04
	BM3D	0.211 ± 0.10	30.106 ± 4.36	0.875 ± 0.05
bed3	LQPET	0.278 ± 0.13	33.240 ± 5.40	0.899 ± 0.06
	CycleAGAN	0.258 ± 0.10	33.892 ± 4.67	0.933 ± 0.03
	CycleGAN	0.265 ± 0.10	34.365 ± 5.41	0.924 ± 0.03
	Pix2Pix	0.276 ± 0.10	32.422 ± 6.19	0.920 ± 0.04
	NLM	0.271 ± 0.14	33.872 ± 4.86	0.896 ± 0.04
	BM3D	0.256 ± 0.11	34.154 ± 4.63	0.910 ± 0.03

Best results and methods are highlighted.