Table 8 Ablation study of the proposed framework for PET-CT synthesis on in-house dataset.

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

Setting

PET-to-CT

CT-to-PET

 

SSIM

PSNR

SSIM

PSNR

S1

0.9569 ± 0.0484

35.54 ± 1.02

0.9491 ± 0.0390

34.82 ± 1.62

S1 + S2

0.9810 ± 0.0319

37.30 ± 1.21

0.9612 ± 0.0312

36.02 ± 1.27

S1 + S3

0.9573 ± 0.0482

37.15 ± 1.76

0.9486 ± 0.0582

35.66 ± 1.89

S2 + S3

0.9828 ± 0.0327

38.01 ± 1.11

0.9384 ± 0.0598

36.21 ± 1.58

S1 + S2 + S3

0.9843 ± 0.0123

38.33 ± 0.55

0.9658 ± 0.0436

36.76 ± 1.25

Image domain

0.9783 ± 0.0155

36.24 ± 1.01

0.9379 ± 0.0759

34.84 ± 1.96

Dual-domain

0.9843 ± 0.0123

38.33 ± 0.55

0.9658 ± 0.0436

36.76 ± 1.25

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