Table 1.

MR-based DL approaches for AC correction in brain PET/MR

Ref	Input	Output	Network	No. in train and test	Dice in bone regions	Regional PET bias	PET surface error	Anatomic abnormalities
Gong et al. 2018 [14]	Dixon	pseudoCT	2.5D UNet	40 (cross validation)	0.76	< 3% (8 VOIs)	No	No
	Dixon + ZTE	pseudoCT	2.5D GroupUNet	14 (cross validation)	0.80	< 3% (8 VOIs)	No	No
Blanc-Durand et al. 2019 [4]	ZTE	PseudoCT	3D UNet	Train 23 Test 47	No	< 2% (70 VOIs)	No	No
Arabi et al. 2019 [12]	T1	PseudoCT	3D DL-AdvSS	40 (cross validation)	0.80	< 3.5% (63 VOIs)	No	No
Spuhler et al. 2019 [3]	T1	LRAM	2D UNet	Train 55 Test 11	No	< 3% (19 VOIs)	No	No
Tao et al. 2021 [13]	T1	pseudoCT	2D cGAN	11 (cross validation)	No	No	No	No
	Dixon	pseudoCT	2D cGAN	10 (cross validation)	No	No	No	No
Gong et al. 2021 [15]	T1	pseudoCT	2.5D GroupUNet	35 (cross validation)	0.84	< 2% (10 VOIs)	Yes	No
	Dixon	pseudoCT	2.5D GroupUNet	35 (cross validation)	0.84	< 2% (10 VOIs)	Yes	No
	mUTE	pseudoCT	2.5D GroupUNet	35 (cross validation)	0.87	< 2% (10 VOIs)	Yes	No