Comparison of dataset quality assessment measures (columns) evaluated on different datasets (rows)
| Dataset . | MOS . | FID . | BRISQUE . | SBR . | SSQA . | IoU(SSQA, SSQA0) . |
|---|---|---|---|---|---|---|
| Reference | 3.981 ± 1.184 | 1.170e−10 | 36.252 ± 5.997 | 1.767 | 0.026 ± 0.029 | 1 |
| Simulation | – | 122.147 | 41.096 ± 3.269 | 1.567 | 0.019 ± 0.023 | 0.4670 |
| SpCycleGAN | 2.967 ± 1.460 | 45.188 | 34.124 ± 6.796 | 0.926 | 0.030 ± 0.027 | 0.6173 |
| SpCycleGAN 3D | 3.083 ± 1.198 | 103.074 | 38.478 ± 2.187 | 1.234 | 0.018 ± 0.018 | 0.3652 |
| CycleGAN 3D | 3.008 ± 1.345 | 106.938 | 38.092 ± 5.439 | 1.364 | 0.021 ± 0.025 | 0.4371 |
| Ours: 3D Cyclic GAN () | 3.381 ± 1.241 | 80.798 | 34.238 ± 5.155 | 1.418 | 0.028 ± 0.029 | 0.6575 |
| Dataset . | MOS . | FID . | BRISQUE . | SBR . | SSQA . | IoU(SSQA, SSQA0) . |
|---|---|---|---|---|---|---|
| Reference | 3.981 ± 1.184 | 1.170e−10 | 36.252 ± 5.997 | 1.767 | 0.026 ± 0.029 | 1 |
| Simulation | – | 122.147 | 41.096 ± 3.269 | 1.567 | 0.019 ± 0.023 | 0.4670 |
| SpCycleGAN | 2.967 ± 1.460 | 45.188 | 34.124 ± 6.796 | 0.926 | 0.030 ± 0.027 | 0.6173 |
| SpCycleGAN 3D | 3.083 ± 1.198 | 103.074 | 38.478 ± 2.187 | 1.234 | 0.018 ± 0.018 | 0.3652 |
| CycleGAN 3D | 3.008 ± 1.345 | 106.938 | 38.092 ± 5.439 | 1.364 | 0.021 ± 0.025 | 0.4371 |
| Ours: 3D Cyclic GAN () | 3.381 ± 1.241 | 80.798 | 34.238 ± 5.155 | 1.418 | 0.028 ± 0.029 | 0.6575 |
Note: FID (Heusel et al., 2017) and BRISQUE are performed on 2D slices in 3D stacks, while the others directly evaluate 3D images. MOS, BRISQUE and SSQA are mean ± standard deviation values. The SBR scores are also averaged over all images in the dataset. Reference denotes the intra-real dataset quality statistics. The closer to the Reference scores, the better relative dataset image quality. The values highlighted in bold are the best results, and the underlined ones in italics are second best.
Comparison of dataset quality assessment measures (columns) evaluated on different datasets (rows)
| Dataset . | MOS . | FID . | BRISQUE . | SBR . | SSQA . | IoU(SSQA, SSQA0) . |
|---|---|---|---|---|---|---|
| Reference | 3.981 ± 1.184 | 1.170e−10 | 36.252 ± 5.997 | 1.767 | 0.026 ± 0.029 | 1 |
| Simulation | – | 122.147 | 41.096 ± 3.269 | 1.567 | 0.019 ± 0.023 | 0.4670 |
| SpCycleGAN | 2.967 ± 1.460 | 45.188 | 34.124 ± 6.796 | 0.926 | 0.030 ± 0.027 | 0.6173 |
| SpCycleGAN 3D | 3.083 ± 1.198 | 103.074 | 38.478 ± 2.187 | 1.234 | 0.018 ± 0.018 | 0.3652 |
| CycleGAN 3D | 3.008 ± 1.345 | 106.938 | 38.092 ± 5.439 | 1.364 | 0.021 ± 0.025 | 0.4371 |
| Ours: 3D Cyclic GAN () | 3.381 ± 1.241 | 80.798 | 34.238 ± 5.155 | 1.418 | 0.028 ± 0.029 | 0.6575 |
| Dataset . | MOS . | FID . | BRISQUE . | SBR . | SSQA . | IoU(SSQA, SSQA0) . |
|---|---|---|---|---|---|---|
| Reference | 3.981 ± 1.184 | 1.170e−10 | 36.252 ± 5.997 | 1.767 | 0.026 ± 0.029 | 1 |
| Simulation | – | 122.147 | 41.096 ± 3.269 | 1.567 | 0.019 ± 0.023 | 0.4670 |
| SpCycleGAN | 2.967 ± 1.460 | 45.188 | 34.124 ± 6.796 | 0.926 | 0.030 ± 0.027 | 0.6173 |
| SpCycleGAN 3D | 3.083 ± 1.198 | 103.074 | 38.478 ± 2.187 | 1.234 | 0.018 ± 0.018 | 0.3652 |
| CycleGAN 3D | 3.008 ± 1.345 | 106.938 | 38.092 ± 5.439 | 1.364 | 0.021 ± 0.025 | 0.4371 |
| Ours: 3D Cyclic GAN () | 3.381 ± 1.241 | 80.798 | 34.238 ± 5.155 | 1.418 | 0.028 ± 0.029 | 0.6575 |
Note: FID (Heusel et al., 2017) and BRISQUE are performed on 2D slices in 3D stacks, while the others directly evaluate 3D images. MOS, BRISQUE and SSQA are mean ± standard deviation values. The SBR scores are also averaged over all images in the dataset. Reference denotes the intra-real dataset quality statistics. The closer to the Reference scores, the better relative dataset image quality. The values highlighted in bold are the best results, and the underlined ones in italics are second best.
This PDF is available to Subscribers Only
View Article Abstract & Purchase OptionsFor full access to this pdf, sign in to an existing account, or purchase an annual subscription.
