A review of substitute CT generation for MRI-only radiation therapy

Edmund, Jens M.; Nyholm, Tufve

doi:10.1186/s13014-016-0747-y

Radiation Oncology

Table 1 Table summarizing performance over different substitute CT approaches

From: A review of substitute CT generation for MRI-only radiation therapy

Approach category				MRI sequence/contrast		Number	Site	Performance metric				Note	References
main	sub 1	sub 2	sub 3					ΔDose [%]	MAE [HU]	DSC _bone	Other
Voxel	Semi-automatic			T1w		20	brain				1	Δ_ROI	[61]
	Threshold			dUTE		1	phantom	2		0.81	87	C_bone	[62–64]
				dUTE		5-19	brain	0.5	230	0.49-0.65	90	C_tissue	[22, 35, 65]
				Dixon		2	brain				94/77	SS_bone	[66]
				dUTE	Dixon	6-98	brain			0.75	88/88,81	SS_bone, C_tissue	[67–69]
	Probabilistic	Clustering	Fuzzy c-means	T2w	Dixon	2-5	brain				97-98/75-94	SS_bone	[70, 71]
		Clustering	Fuzzy c-means	T2w	T1w	10	brain				98/74	SS_air	[52]
				dUTE	Dixon	9	brain		130				[72]
		Bayesian	Markow RF	dUTE		5	brain	1	204-247	0.53-0.59			[22]
	Regression	Discriminant		dUTE	T2w	3	phantom	2.3	88				[73]
		Discriminant		dUTE	T2w	3	brain	1.2	153				[74]
		Gaussian		dUTE	T2w	5-9	brain	0.9-1.5	137-140	0.85	68-94,98	γ₁₁, γ₃₃	[23, 75–77]
		Gaussian		dUTE		5	brain	1	136-148	0.67-0.72			[22, 48]
		Random F		dUTE		5	brain	1	128	0.74			[22]
		Semi-automatic Dual		T1w	T2w	9	prostate	0.7	75	0.91	99.9	^a, γ₂₂	[78]
		Semi-automatic Dual		T1w	T2*w	10-15	prostate	0.3-2	135		93	γ₁₁	[21, 79, 80]
		PCA		T1w		10	torso				5	Δ_bone	[81]
	Sinogram			T1w		10	brain			0.85		^b	[82]
	Neural network			T1w		3	brain			0.78			[83]
	Neural network			dUTE		4	brain			0.83			[84]
	Pattern recognition			dUTE	Dixon	10	brain				76	C_air	[85]
	Hybrid	Neural Network	Template	dUTE		4	brain			0.77			[86]
		Gaussian	Spatial info	dUTE	T2w	9	brain		130				[75]
		Random F	Spatial info	T1w		9-10	brain			0.92-0.98		^c	[87, 88]
Atlas	Pattern recognition	Patch Patch Probabilistic		T1w		5 40	brain	0.5	85	0.84 0.75			[48] [89]
	Deformable			T1w		28	whole body				0.88	DSC_brain	[90]
				T1w	T2w	5-17	brain	1-2.5	97-114	0.63-0.83	1.7	Δ_ROI	[24, 48, 91–94]
				T2w		37	prostate	1.5		0.79			[25]
				T2w		10	cervix	0.3					[95]
	Hybrid	Deformable	Patch	T1w		17	brain		101				[96]
				T2w		39	Prostate	0.3	40.5	0.91	100	γ₂₂	[97]
Hybrid	Deformable	Probabilistic		T1w		9-27	brain		126	0.86	86/90	SS_bone	[26, 98]
				T2w		10	prostate	0.2	36.5		99.9	γ₂₁	[27]
		Regression	Random F	T2w		20	prostate				0.83	DSC_prostate	[99]
		Threshold		dUTE		154	brain			0.81			[100]

Column 1: Overall approach: Voxel, atlas or hybrid. Column 2: Sub categories within each main approach. Column 3: MRI sequences/contrasts applied in the studies. Column 4: Number of subjects included in the studies (patients, volunteers, phantoms). Column 5: Anatomical site investigated. Column 6: Performance metrics. Column 7: Specification of “Other” metric and comments. Column 8: References to included studies. SS_x = % specificity @ % sensitivity for tissue x, C_x = % correctly classified voxel for tissue x, Δ_x = mm distance bt. CT and sCT for tissue x, γ_xy = % of points with γ_x _%ymm< 1 where x and y are the dosimetric and geometric deviations, respectively, ^aMAE of whole FOV and DSC based on 2D DRR, ^b Overlap ratio similar to DSC_bone. ^c Bone > 600 HU. Method abbreviations: Bayesian Bayesian statistics, Markow RF Markow Random Fields, Discriminant Discriminant analysis, Random F Random Forest, PCA Principal Component Analysis, Patch = cluster/collection of MRI voxels, Deformable = deformable registration

Back to article page

ISSN: 1748-717X

Contact us

General enquiries: journalsubmissions@springernature.com