Table 3. Quantitative evaluation of the results of the different methods from Figure 9 for 50 K, 500 K, and 5 M counts.

Methods	50 K				500 K				5 M
	PSNR	SSIM	rRMSE		PSNR	SSIM	rRMSE		PSNR	SSIM	rRMSE
EM + Gaussian filter	21.028	0.593	0.276		23	0.882	0.231		26.309	0.954	0.15
EM + NLM	20.165	0.651	0.304		28	0.895	0.127		27.131	0.932	0.137
CNN	21.924	0.786	0.249		30	0.926	0.104		30.595	0.938	0.092
GAN	22.233	0.818	0.258		28	0.914	0.118		32.945	0.959	0.07
LCPR-Net	25.873	0.819	0.234		30	0.934	0.099		33.303	0.963	0.067

PSNR, peak signal-to-noise ratio; SSIM, structural similarity index; rRMSE, relative root mean square error; EM, expectation-maximization; NLM, nonlocal means; CNN, convolutional neural network; GAN, generative adversarial network; LCPR-Net, low-count positron emission tomography image reconstruction network.