TABLE 2.

Performance comparison with different baseline models

Models	Train			Test
	SSIM	PSNR	MSE	SSIM	PSNR	MSE
AE	0.897 ± 0.006	33.87 ± 0.87	0.0008 ± 0.0001	0.889 ± 0.004	34.88 ± 0.38	0.0006 ± 0.0001
VAE	0.886 ± 0.01	35.45 ± 0.13	0.0008 ± 0.0001	0.894 ± 0.003	34.57 ± 0.48	0.0009 ± 0.0001
FAnoGAN	0.804 ± 0.02	25.10 ± 0.14	0.0034 ± 0.0001	0.797 ± 0.006	24.45 ± 0.06	0.0037 ± 0.0001
DC‐CNN	0.879 ± 0.002	36.99 ± 0.81	0.0004 ± 0.0001	0.884 ± 0.01	36.45 ± 0.73	0.0004 ± 0.0001
AAE	0.931 ± 0.002	30.75 ± 0.11	0.0016 ± 0.0001	0.927 ± 0.005	30.93 ± 0.12	0.0015 ± 0.0007
GANCMLAE	0.934 ± 0.006	31.66 ± 0.07	0.0015 ± 0.0006	0.929 ± 0.003	31.04 ± 0.09	0.0014 ± 0.0001
L1 + L2	0.929 ± 0.009	35.49 ± 0.62	0.0006 ± 0.0003	0.929 ± 0.009	35.27 ± 0.78	0.0007 ± 0.0004
L1 + L3	0.889 ± 0.016	33.35 ± 0.10	0.0009 ± 0.0005	0.895 ± 0.014	33.73 ± 0.96	0.0008 ± 0.0003
L2 + L3	0.911 ± 0.015	35.10 ± 0.45	0.0006 ± 0.0003	0.914 ± 0.011	35.21 ± 0.50	0.0006 ± 0.0003

Note: The methods are conducted with cross‐validation and their results are given as mean ± standard deviation. The best performing models are highlighted in bold.

Abbreviations: AE, autoencoder; GANCMLAE, generative adversarial networks constrained multiple loss autoencoder; MSE, mean squared error; PSNR, peak signal‐to‐noise ratio; SSIM, structural similarity index measure; VAE, variations encoder.