Table 2.

Classification accuracies and standard deviations of the 5 model architectures on individual sites.

	SL	DH + SL	SSL	DH + SSL
	DNN	DNN	EDC	SHRED	SHRED-II	SHRED-III
COBRE	77.78  ± 2.26	78.52  ± 1.35	82.00  ± 0.78	75.78  ± 1.88	82.15  ± 1.23	76.89  ± 3.85
IMH	61.61  ± 4.07	57.96  ± 2.97	77.93  ± 4.67	73.71  ± 4.55	69.96  ± 2.96	79.18  ± 3.49
NMorphCH	66.01  ± 3.54	63.40  ± 4.07	75.58  ± 2.02	73.02  ± 1.96	75.35  ± 1.81	71.36  ± 4.97
UCLA	79.48  ± 1.34	81.05  ± 1.74	83.84  ± 1.39	83.56  ± 1.05	83.74  ± 1.23	83.92  ± 1.19

SL supervised learning, DH data harmonisation, SSL semi-supervised learning.