Table 4.

Overview of the participating algorithms. The training accuracy was computed on the 30 training subjects by training on the data from different sources only. As indicated below, three algorithms instead trained on all data using 5-fold or 10-fold cross-validation.

	Algorithm	Features	Classifier	Size training data	Training accuracy [%]
1	Abdulkadir	VBM	SVM	1492	60
2	Amoroso	Volume and intensity relations	Neural network	288	67 5-fold
3	Cárdenas-Peña	Raw intensities	SVM	451	83
4	Dolph	Volumes	SVM	30	80 10-fold
5	Eskildsen-ADNI1	Volume and intensity relations	Regression	794	77
6	Eskildsen-ADNI2	Volume and intensity relations	Regression	304	70
7	Eskildsen-Combined	Volume, thickness and intensity relations	Regression	1098	73
8	Eskildsen-FACEADNI1	Volume, thickness and intensity relations	Regression	794	70
9	Eskildsen-FACEADNI2	Volume, thickness and intensity relations	Regression	304	67
10	Franke	VBM	Regression	591	90
11	Ledig-ALL	Volume, thickness and intensity relations	Random forest	734	68
12	Ledig-CORT	Cortical thickness	Random forest	734	58
13	Ledig-GRAD	Intensity relations	Random forest	734	67
14	Ledig-MBL	Intensity relations	Random forest	734	66
15	Ledig-VOL	Volumes	Random forest	734	56
16	Moradi	VBM	SVM	835	77
17	Routier-adni	Shapes	Regression	539	50
18	Routier-train	Shapes	Regression	539	73
19	Sarica	Volume and thickness	SVM	210	70
20	Sensi	Intensity relations	Random forest, SVM	581	73
21	Smith	Volume and raw intensities	Regression	189	80
22	Sørensen-equal	Volume, thickness, shape, intensity relations	LDA	679	73
23	Sørensen-optimized	Volume, thickness, shape, intensity relations	LDA	679	80
24	Tangaro	Volume and thickness	SVM	190	73 5-fold
25	Wachinger-enetNorm	Volume, thickness and shape	Regression	781	73
26	Wachinger-man	Volume, thickness and shape	Regression	781	67
27	Wachinger-step1	Volume, thickness and shape	Regression	781	77
28	Wachinger-step1Norm	Volume, thickness and shape	Regression	781	77
29	Wachinger-step2	Volume, thickness and shape	Regression	781	80