Table 1 Comparison with previous studies.

Method	Dataset	Input	Period	Accuracy	AUC	Sensitivity	Specificity
DNN [7]	NA-ADNI	MRI, FDG-PET	3 years	0.83	–	0.80	0.84
DNN [8]	NA-ADNI^a	MRI	3 years	0.75	–	0.75	0.75
DNN [9]	NA-ADNI	MRI	1.5 years^b	0.79	–	0.75	0.82
DNN [10]	NA-ADNI	MRI	3 years	0.83^c	0.88^c	0.76^c	0.87^c
SVM [11]	NA-ADNI	MRI, Cognitive scores	3 years	0.85	–	0.47	0.97
Random forest[12]	NA-ADNI	MRI, gender, age	1 year	0.79	–	0.82	0.74
SVM [13]	NA-ADNI	MRI, MMSE	3 years^b	0.85^d	0.90^d	0.84^d	0.88^d
DNN [14]	NA-ADNI^a	MRI, Cognitive scores, age	Series	–	0.76	–	–
DNN [15]	NA-ADNI	MRI, Cognitive scores, CSF, demographics	1 year	0.81	–	0.84	0.80
DNN [16]	NA-ADNI	MRI, Cognitive scores, APOE, gender, age	1 year	0.86^e	–	0.82^e	0.88^e
Random forest[17]	NA-ADNI	MRI, PET, Cognitive scores, APOE	3 years	0.87 ^f	0.87 ^f	0.86 ^f	–
Proposed (M2)	NA-ADNI	MRI	2 years	0.78	0.85	0.78	0.78
Proposed (M5)	NA-ADNI	MRI, Cognitive scores, APOE, age	2 years	0.88	0.95	0.88	0.88

^aRelatively small datasets compared with the NA-ADNI were also used for both training and evaluation.
^bFollow-up periods used for sMCI are longer than that of pMCI (5 years in DNN [9] and 4 years in SVM [13]).
^cTrained with mixed groups of NC+sMCI and pSMI+AD.
^dtwofold validation accuracy.
^eTenfold cross-validation accuracy.
^fTenfold cross-validation accuracy, much higher accuracy of 91% on a small hold out test set.

Quick links

Search