Table 3.

Comparison of our proposed MPBG biomarkers with state-of-the-art methods based on s-MRI and d-MRI using a similar ADNI2 dataset.

Method	Subjects				Feature	Classifier	Classification ACC
	CN	eMCI	lMCI	AD			CN/AD	eMCI/lMCI
Nir et al.93	44	74	39	23	Tractography	SVM	84.9%	n/a
Prasad et al.60	50	74	38	38	Connectivity network	SVM	78.2%	63.4%
Zhan et al.94	n/a	73	39	n/a	Connectivity network	SLG	n/a	65.0%
Maggipinto et al.96	50	22	18	50	Voxel-based	RF	87.0%	n/a
La Rocca et al.95	52	85	38	47	Connectivity network	RF	83.0%	n/a
MPBG hippocampus	62	65	34	38	Patch-based	LDA	88.1%	68.8%
MPBG Subiculum	62	65	34	38	Patch-based	LDA	86.5%	70.8%

All results are expressed in percentage of accuracy.

LDA = Linear Discriminant Analysis,

SLG = Sparse Logistic Regression,

SVM = Support Vector Machine,

RF = Random Forest.