. Author manuscript; available in PMC: 2023 Oct 9.

Published in final edited form as: PET Clin. 2021 Aug 5;16(4):543–552. doi: 10.1016/j.cpet.2021.06.010

Table 3.

AI-based approaches for simultaneous attenuation and scatter correction

Application	Approach	Quantitative Comparison vs. Acquired CT	Reference
Brain PET (FDG, FDOPA, Flortaucipir, Flutemetamol)	CNN	PET reconstruction bias <9%	Arabi et al., 2020³⁴
Brain PET (FDG)	CNN, UNet	PET reconstruction bias −4.2% ± 4.3%	Yang et al., 2019³⁵
Whole body PET (68Ga-PSMA)	CNN	MAE 0.91 ± 0.29 SUV, SSIM 0.973 ± 0.034, PSNR 48.17 ± 2.96 dB, PET reconstruction bias −2.46% ± 10.10%	Mostafapour et al., 2021³⁶
Whole body PET (FDG)	CNN	PET reconstruction bias −1.72 ± 4.22%	Shiri et al., 2020³⁷
Whole body PET (FDG)	CNN	NRMSE 0.21 ± 0.05 SUV, SSIM 0.98 ± 0.01, PSNR 36.3 ± 3.0 dB	Yang et al., 2020³⁸
Whole body PET (FDG)	CNN, GAN	MAE of < 110 HU, PET reconstruction bias <1%, PSNR >42 dB	Dong et al., 2019¹⁵