Table 6.

AD/MCI classification comparison of our proposed method (iMSF) and missing value estimation methods (Zero, KNN, SVD and EM) in terms of accuracy, sensitivity and specificity when the training percentage varies from 1/2 to 3/4 as well as leave-one-out (LOO). In this experiment, we used the full multi-source data including MRI, PET, proteomics and CSF with 569 subjects in total.

Accuracy	Training Size	iMSF	Zero	EM	KNN	SVD
	50.0%	0.8278	0.804	0.8025	0.7963	0.8059
	66.7%	0.8335	0.812	0.812	0.8035	0.8149
	75.0%	0.8401	0.8242	0.8148	0.8091	0.8148
	LOO	0.8563	0.8209	0.813	0.8091	0.8071
Sensitivity	Training Size	iMSF	Zero	EM	KNN	SVD
	50.0%	0.4339	0.1406	0.1232	0.1021	0.159
	66.7%	0.4424	0.1663	0.1552	0.1192	0.1848
	75.0%	0.4514	0.1907	0.1467	0.1286	0.1649
	LOO	0.5	0.1964	0.1696	0.1607	0.1607
Specificity	Training Size	iMSF	Zero	EM	KNN	SVD
	50.0%	0.9628	0.9894	0.9924	0.9902	0.9867
	66.7%	0.9643	0.9898	0.9923	0.9913	0.9879
	75.0%	0.967	0.9946	0.9946	0.9923	0.9899
	LOO	0.9697	0.9975	0.9949	0.9924	0.9899