Table 5.

Classification performance when directly using discriminative subnetworks and that of our proposed method.

Method	Task	ACC (%)	SEN (%)	SPE (%)	AUC
Discriminative subnetwork	MCI/NC	78.3	66.7	80	0.78
	AD/NC	88.2	82.3	94.1	0.97
	AD/MCI	69.1	71.2	67.4	0.76
Proposed method	MCI/NC	98.3	96.6	100	0.99
	AD/NC	91.3	82.5	100	1
	AD/MCI	77.3	54.1	100	0.97

Discriminative subnetwork denotes directly use discriminative subnetworks as features; AD, Alzheimer's disease; NC, Normal control; MCI, Mild cognitive impairment; ACC, Accuracy; SEN, Sensitivity; SPE, Specificity; AUC, The area under the receiver operating characteristic curve.