Table 1 Quantitative comparison with representative learning-based methods for low-dose PET/CT reconstruction.

From: A generalized dual-domain generative framework with hierarchical consistency for medical image reconstruction and synthesis

Tasks

Low-dose PET reconstruction

Low-dose CT reconstruction

Method

SSIM

PSNR

SUVmean

SUVmax

SSIM

PSNR

RUNet45

0.8064 ± 0.1028

35.07 ± 1.03

8.70 ± 6.79

6.23 ± 5.94

0.9311 ± 0.0823

41.02 ± 2.31

p2pGAN46

0.9844 ± 0.0129

37.55 ± 1.20

5.99 ± 9.04

3.92 ± 5.02

0.9712 ± 0.0510

42.41 ± 1.79

CycleGAN40

0.9861 ± 0.0078

37.85 ± 1.18

7.79 ± 5.86

-6.63 ± 1.04

0.9811 ± 0.0086

41.90 ± 0.94

MedGAN38

0.9863 ± 0.0093

37.72 ± 1.11

4.01 ± 8.16

-8.31 ± 4.59

0.9613 ± 0.0577

41.53 ± 2.05

iBP-Net47

0.9859 ± 0.0154

38.47 ± 1.28

6.84 ± 3.97

5.96 ± 1.79

0.9812 ± 0.0143

42.71 ± 0.97

Ours

0.9939 ± 0.0124

39.57 ± 1.03

3.62 ± 5.83

3.54 ± 2.40

0.9910 ± 0.0032

43.57 ± 0.85

  1. Best performance is highlighted in bold font.
Back to article page