Table 2.

SVM predictions using baseline MRI data.

	Total accuracy %	Balanced accuracy %	Balanced accuracy P value	Class accuracy %		Class accuracy P value		Class predictive value %		ROC
				Sensitivity	Specificity	Sensitivity	Specificity	Positive predictive value	Negative predictive value
Jacobian	83.19	63.33	0.01	31.82	94.85	0.01	0.82	58.33	85.98	0.79
GM	84.87	74.36	0.01	31.82	96.91	0.01	0.07	70	86.24	0.79
WM	79.83	52.48	0.1	9.09	95.88	0.24	0.12	33.33	82.3	0.62
WMH	79.83	54.24	0.01	13.64	94.85	0.01	0.95	37.5	82.88	0.65
Lacunar	84.03	56.82	0.01	13.64	100	0.02	0.06	100	83.62	0.64
Cortical Thickness	79.83	57.76	0.01	22.73	92.78	0.02	0.52	41.67	84.11	0.71
Putamen	82.35	68.09	0.01	45.45	90.72	0.01	0.04	52.63	88	0.78
Caudate	78.99	60.75	0.01	31.82	89.69	0.04	0.05	41.18	85.29	0.73
Striatum (Putamen + Caudate)	80.67	65.3	0.01	40.91	89.69	0.01	0.15	47.37	87	0.77
Hippocampus	83.19	65.09	0.01	36.36	93.81	0.02	0.01	57.14	86.87	0.8
Thalamus	80.67	61.79	0.01	31.82	91.75	0.01	0.01	46.67	85.58	0.72
Putamen + Hippocampus	84.87	69.93	0.01	45.45	93.81	0.01	0.01	62.5	88.35	0.84
Striatum + Hippocampus	84.87	73.17	0.01	54.55	91.75	0.01	0.05	60	89.9	0.84
Putamen + Hippocampus + Thalamus	83.19	70.36	0.01	50	90.72	0.01	0.15	55	88.89	0.81
Striatum + Hippocampus + Thalamus	82.35	66.33	0.01	40.91	91.75	0.01	0.15	52.94	87.25	0.81

The masked analysis used subcortical ROIs and the Jacobian determinant data. The best prediction (judged by balanced accuracy) was achieved using the combined striatum-hippocampus mask, which was significant at P < 0.01 using permutation testing (10,000 permutations)