. Author manuscript; available in PMC: 2021 May 15.

Published in final edited form as: Predict Intell Med. 2020 Oct 1;12329:91–100. doi: 10.1007/978-3-030-59354-4_9

Table 1:

Model performance evaluated by various image quality metrics on both the entire image and only the cropped regions. Amount of cropping was fixed to 12.5%, similar to the training data. PSNR is measured in dB.

Model	Whole				Cropped
	SSIM	PSNR	MSE	Gradient	SSIM	PSNR	MSE	Gradient
U-VQVAE	0.9973	49.8733	1.089e-5	0.0209	0.9679	30.7809	8.535e-5	0.0041
VQVAE	0.8467	32.0512	5.437e-4	0.0064	0.9279	28.4808	1.449e-4	0.0034
U-Net [13]	0.9959	49.4796	1.152e-5	0.0210	0.9638	30.6117	8.723e-5	0.0039
VAE-GAN	0.8074	30.6311	2.919e-3	0.1206	0.8036	21.7169	2.348e-3	0.2033