TABLE 3.

Results of the ablation studies of BP module and metric learning losses for NC VS. AD classification (Mean ± Standard Deviation).

Method	ACC (%)	PPV (%)	NPV (%)	SEN (%)	SPE (%)	AUC	F1 (%)	p
Baseline	85.25 ± 2.50	92.01 ± 2.82	76.33 ± 6.77	83.91 ± 3.96	87.99 ± 3.52	0.9074 ± 0.0215	87.77 ± 1.96	–
Baseline + BP	88.94 ± 1.20	93.52 ± 3.48	82.79 ± 3.50	87.92 ± 1.77	90.93 ± 4.33	0.9286 ± 0.0218	90.63 ± 1.22	0.051
Tri-loss	88.29 ± 0.86	93.54 ± 0.99	81.33 ± 2.04	86.92 ± 1.44	90.47 ± 1.31	0.9279 ± 0.0127	90.35 ± 0.74	0.059
Tri-loss + BP	89.80 ± 0.62	93.53 ± 3.70	84.81 ± 4.88	89.28 ± 2.79	91.20 ± 4.28	0.9281 ± 0.0192	91.11 ± 0.70	0.032*
Con-loss	89.14 ± 1.75	93.53 ± 2.56	83.35 ± 2.78	88.02 ± 1.59	90.76 ± 3.54	0.9281 ± 0.0253	90.69 ± 1.48	0.088
Con-loss + BP	89.80 ± 0.99	93.90 ± 2.44	84.32 ± 3.40	88.91 ± 2.00	91.40 ± 2.97	0.9334 ± 0.0141	91.35 ± 0.80	0.029*

The bold values represent the highest number.

The asterisk represents the results have the statistical significance.