Table 1.
AI-based approaches for attenuation correction
| Application | Approach | Advance | Quantitative Comparison vs. Acquired CT | Reference |
|---|---|---|---|---|
| Brain perfusion SPECT | CNN | Enable correction for SPECT only scanners | PSNR 62.2 dB, SSIM 0.9995 | Sakaguchi et al., 202112 |
| Brain PET (FDG) | CNN, UNet | Enable correction for PET only scanners | Dice coefficients of 0.80 ± 0.02 for air, 0.94 ± 0.01 for soft tissue, and 0.75 ± 0.03 for bone, MAE of 111 ± 16 HU, PET reconstruction bias <1% | Liu et al., 201813 |
| Brain PET/MR (FDG) | CNN | Synthesize CT from T1 MRI for PET/MRI | Dice coefficient of 0.936 ± 0.011 for soft tissue, and 0.803 ± 0.021 for bone, PET reconstruction bias −0.7% ± 1.1% | Liu et al., 201714 |
| Brain PET/MR (FDG) | CNN, GAN | Synthesize CT from T1 MRI for PET/MRI | Dice coefficient of 0.77 ± 0.07 for bone, PET reconstruction bias <4% | Arabi et al., 201915 |
| Brain PET/MR (FDG) | CNN, GAN | Synthesize CT from T1 MRI for PET/MRI | Dice coefficient of 0.74 ± 0.05 for bone, PET reconstruction bias <0.025 SUV | Gong et al., 202016 |
| Brain PET/MR (FDG) | CNN, UNet | Synthesize CT from T1 MRI for PET/MRI | Dice coefficient of 0.81 ± 0.03 for bone, PET RMSE 253.5, PET reconstruction bias <1% | Blanc-Durand et al., 201917 |
| Brain PET/MR (FDG) | CNN, UNet | Synthesize CT from T1 and ZTE MRI for PET/MRI | Dice coefficient of 0.86 ± 0.03 for bone, PET reconstruction bias <0.02 SUV | Gong et al., 201818 |
| Brain PET/MR (FDG) | CNN, UNet | Synthesize CT from UTE MRI for PET/MRI | Dice coefficient of 0.96 ± 0.006 for soft tissue, 0.88 ± 0.01 for bone, PET reconstruction bias <1% | Jang et al., 201819 |
| Brain PET/MR (FET) | CNN, UNet | Synthesize CT from UTE MRI for PET/MRI | PET reconstruction bias <1% | Ladefoged et al., 201920 |
| Myocardial perfusion SPECT | CNN | Enable correction for SPECT only scanners | MAE 3.60% ± 0.85% in whole body, 1.33% ± 3.80% in the left ventricle | Shi et al., 202021 |
| Myocardial perfusion SPECT | CNN | Enable correction for SPECT only scanners | Mean segmental errors 0.49 ± 4.35%, mean absolute segmental errors 3.31 ± 2.87% | Yang et al., 202122 |
| Myocardial perfusion SPECT | GAN | Enable correction for SPECT only scanners | RMSE 0.141±0.0768, PSNR 36.382±3.74 dB, SSIM 0.995±0.0043 | Torkaman et al., 202123 |
| Pelvis PET/MR (68Ga-PSMA) | CNN, GAN | Synthesize CT from T1 MRI for PET/MRI | PET reconstruction bias <2.2% | Pozaruk et al., 202124 |
| Pelvis PET/MR (FDG) | CNN | Synthesize CT from T1 and T2 MRI for PET/MRI | Dice coefficient of 0.98 ± 0.01 for soft tissue, 0.79 ± 0.03 for cortical bone, PET reconstruction bias 4.9% | Bradshaw et al., 201825 |
| Pelvis PET/MR (FDG) | CNN, UNet | Synthesize CT from T1 MRI for PET/MRI | PET reconstruction bias <4% | Torrado-Carvajal et al., 201926 |
| Whole body PET (FDG) | CNN, GAN | Enable correction for PET only scanners | PET reconstruction bias −0.8 ± 8.6% | Armanious et al., 202027 |