TABLE 5.

Results of the ablation studies of BP module and metric learning loss for LMCI vs. AD classification (Mean ± Standard Deviation).

Method	ACC (%)	PPV (%)	NPV (%)	SEN (%)	SPE (%)	AUC	F1 (%)	p
Baseline	77.69 ± 2.67	74.16 ± 5.54	80.30 ± 6.73	74.57 ± 5.91	80.86 ± 2.54	0.7964 ± 0.0216	74.37 ± 2.67	–
Baseline + BP	80.06 ± 5.78	72.92 ± 11.12	85.37 ± 4.11	78.62 ± 5.95	81.29 ± 6.67	0.8108 ± 0.0581	75.66 ± 8.02	0.243
Tri-loss	80.60 ± 3.04	74.25 ± 9.76	85.31 ± 6.67	79.77 ± 6.85	82.23 ± 5.29	0.8040 ± 0.0422	76.91 ± 4.30	0.307
Tri-loss + BP	81.18 ± 2.72	78.90 ± 7.80	82.85 ± 4.39	77.53 ± 3.26	84.41 ± 4.42	0.8167 ± 0.0295	78.21 ± 3.97	0.022*
Con-loss	79.71 ± 0.97	74.87 ± 7.56	83.31 ± 5.89	77.53 ± 5.36	82.00 ± 3.37	0.8018 ± 0.0332	76.18 ± 2.19	0.327
Con-loss + BP	81.77 ± 4.50	77.54 ± 8.19	84.91 ± 6.27	79.64 ± 6.88	83.84 ± 4.97	0.8297 ± 0.0346	78.57 ± 5.53	0.028*

The bold values represent the highest number.

The asterisk represents the results have the statistical significance.