Table 7.

Performance of the combined fusion (${ℱ}_{SMVCCA}$) and modality selection (${\mathcal{K}}_{SEL}$) modules of CaMCCo for binary classification (${\mathscr{H}}_{BIN}$).

		ACC	BACC	AUC	SEN	SPEC	PPV
AD vs. HC	ADAS - Cog	0.93	0.92	0.98	0.97	0.88	0.93
	T1w MRI	0.63	0.55	0.65	0.86	0.24	0.66
	PET	0.80	0.78	0.87	0.86	0.71	0.83
	CSF	0.91	0.89	0.97	0.97	0.82	0.90
	PP	0.59	0.47	0.51	0.93	0.00	0.61
	APOE	0.76	0.70	0.86	0.93	0.47	0.75
	${\mathscr{H}}_{BIN}$	0.87	0.85	0.98	0.93	0.76	0.87
MCI vs. HC	ADAS-Cog	0.79	0.75	0.84	0.85	0.65	0.85
	T1w MRI	0.69	0.49	0.53	0.98	0.00	0.70
	PET	0.72	0.67	0.76	0.80	0.53	0.80
	CSF	0.83	0.77	0.92	0.90	0.65	0.86
	PP	0.71	0.52	0.54	0.98	0.06	0.71
	APOE	0.83	0.71	0.84	1.00	0.41	0.80
	${\mathscr{H}}_{BIN}$	0.86	0.82	0.93	0.93	0.71	0.88

${\mathscr{H}}_{BIN}$ shows improvement over individual modality classifiers for MCI vs. HC, particularly in terms of achieving the both high sensitivity and specificity. For AD vs. HC, several individual modalities have sufficiently high classification performance and thereby leaving no room for further improvement with ${\mathscr{H}}_{BIN}$.