Table 5.

Prediction of MCI conversion.

	NA-ADNI			NA-ADNI*		JADNI*
	SVMst	SVMcog	VSRAD	SVMst	SVMcog	SVMst	SVMcog
AUC	0.9064	0.9019	0.7407	0.9400	0.9632	0.9240	0.9697
Accuracy (%)	83.0	85.0	62.1	87.2	90.6	86.4	91.2
Sensitivity (%)	81.7	83.8	54.2	81.7	83.8	81.0	86.0
Specificity (%)	85.9	87.5	79.7	90.1	94.1	90.8	95.4
PPV	92.8	93.7	85.6	81.1	88.1	87.5	93.7
NPV	67.9	70.9	44.0	90.4	91.8	85.7	89.5
F1	86.9	88.5	66.4	81.4	85.9	84.1	89.7
MCC	64.1	67.9	31.6	71.6	78.9	72.5	82.3

The subjects of the NA-ADNI were selected as validation set.

The cut-off values of SVMst and SVMcog were set as ADLS > 0.5.

^*, discrimination between pMCI and cognitively normal subject.

PPV, positive predictive value (=precision); NPV, negative predictive rate; AUC, area under the curve; MCC, Matthews correlation coefficient.