Analysis of a deep learning-based method for generation of SPECT projections based on a large Monte Carlo simulated dataset

EJNMMI Physics

Table 3 Analysis of the u-net performance

Abbreviation	Synthetic versus noisy projections		Synthetic versus noise-free projections
Abbreviation	SSIM	NRMSE	SSIM	NRMSE
U1	0.979 (0.025)	2.45% (1.18%)	0.995 (0.011)	1.19% (0.75%)
U2	0.979 (0.026)	2.44% (1.19%)	0.995 (0.011)	1.19% (0.79%)
U3	0.973 (0.033)	2.79% (1.29%)	0.989 (0.022)	1.85% (1.17%)
U4	0.978 (0.024)	2.52% (1.25%)	0.994 (0.007)	1.29% (0.75%)
U5	0.978 (0.024)	2.53% (1.20%)	0.994(0.008)	1.32% (0.68%)
U6	0.978 (0.025)	2.56% (1.53%)	0.994 (0.009)	1.37% (0.75%)
U7	0.982 (0.018)	2.24% (1.10%)	0.997 (0.006)	1.04% (0.61%)

Mean SSIM and NRMSE values between synthetic projections and noisy/noise-free projections, respectively. Data in parentheses are standard deviations