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

Table 2

Quantitative comparison on LQPET images from the scanner with 320 mm FOV.

Data	Methods	NRMSE	PSNR (dB)	SSIM
bed1	LQPET	0.248 ± 0.20	31.595 ± 4.67	0.915 ± 0.08
	CycleAGAN	0.219 ± 0.09	34.003 ± 2.11	0.950 ± 0.03
	CycleGAN	0.340 ± 0.19	31.263 ± 3.70	0.900 ± 0.07
	Pix2Pix	0.292 ± 0.16	33.171 ± 4.49	0.931 ± 0.05
	NLM	0.303 ± 0.24	34.309 ± 3.85	0.883 ± 0.06
	BM3D	0.244 ± 0.13	34.525 ± 3.67	0.883 ± 0.06
bed2	LQPET	0.258 ± 0.14	30.293 ± 5.28	0.910 ± 0.05
	CycleAGAN	0.222 ± 0.06	30.228 ± 2.79	0.927 ± 0.02
	CycleGAN	0.241 ± 0.09	31.284 ± 5.05	0.921 ± 0.03
	Pix2Pix	0.241 ± 0.11	30.685 ± 3.63	0.918 ± 0.03
	NLM	0.258 ± 0.14	31.454 ± 4.92	0.898 ± 0.03
	BM3D	0.241 ± 0.10	31.972 ± 5.35	0.912 ± 0.05
bed3	LQPET	0.263 ± 0.14	34.973 ± 8.13	0.948 ± 0.05
	CycleAGAN	0.257 ± 0.07	36.065 ± 6.87	0.958 ± 0.02
	CycleGAN	0.271 ± 0.14	36.286 ± 6.91	0.952 ± 0.03
	Pix2Pix	0.273 ± 0.12	35.526 ± 8.30	0.949 ± 0.04
	NLM	0.276 ± 0.14	35.319 ± 5.24	0.937 ± 0.04
	BM3D	0.300 ± 0.17	36.315 ± 7.27	0.934 ± 0.04

Best results and methods are highlighted.