Table 2.
Comparison of the results obtained from analysis of image quality of LD PET images and images predicted using ResNet (RNET) and CycleGAN (CGAN) for the test and validation datasets. SSIM, structural similarity index metrics; PSNR, peak signal to noise ratio; RMSE, root mean squared error
| MSE | SSIM | PSNR | |
|---|---|---|---|
| Validation dataset | |||
| CGAN | 0.03 ± 0.04 | 0.98 ± 0.08 | 41.08 ± 3.90 |
| RNET | 0.12 ± 0.10 | 0.94 ± 0.10 | 35.41 ± 5.56 |
| LD | 0.15 ± 0.09 | 0.89 ± 0.11 | 31.21 ± 3.08 |
| P value (CGAN vs. RNET) | 0.012 | 0.018 | 0.035 |
| P value (CGAN vs. LD) | 0.022 | 0.011 | 0.017 |
| P value (RNET vs. LD) | 0.051 | 0.025 | 0.041 |
| Test dataset | |||
| CGAN | 0.03 ± 0.07 | 0.97 ± 0.02 | 39.08 ± 3.56 |
| RNET | 0.13 ± 0.10 | 0.93 ± 0.04 | 34.91 ± 1.50 |
| LD | 0.17 ± 0.04 | 0.9 ± 0.03 | 29.21 ± 2.43 |
| P value (CGAN vs. RNET) | 0.021 | 0.015 | 0.015 |
| P value (CGAN vs. LD) | 0.013 | 0.03 | 0.022 |
| P value (RNET vs. LD) | 0.021 | 0.011 | 0.021 |