TABLE 4.

A brief description of the state-of-the-art studies using baseline sMRI data of ADNI-1 for AD classification (i.e., AD vs. NC) and MCI conversion prediction (i.e., pMCI vs. sMCI).

Reference	Methodology	Subject	AD vs. NC				pMCI vs. sMCI
			ACC	SEN	SPE	AUC	ACC	SEN	SPE	AUC
Hinrichs et al. [9]	Conventional classifiers (i.e., LPboosting, SVM) + voxel-level engineered features	183 (AD+NC)	0.82	0.85	0.80	0.88	-	-	-	-
Salvatore et al. [41]		162 NC + 76 sMCI + 134 pMCI + 137 AD	0.76	-	-	-	0.66	-	-	-
Koikkalainen et al. [45]	Conventional classifiers (i.e., linear regression, ensemble SVM) + region-level engineered features	115 NC + 115 sMCI + 54 pMCI + 88 AD	0.86	0.81	0.91	-	0.72	0.77	0.71	-
Liu et al. [17]		128 NC + 117 sMCI + 117 pMCI + 97 AD	0.93	0.95	0.90	0.96	0.79	0.88	0.76	0.83
Coupé et al. [19]	Conventional classifiers (i.e., linear discriminant analysis, hierarchical SVM, MIL model) + patch-level engineered features	231 NC + 238 sMCI + 167 pMCI + 198 AD	0.91	0.87	0.94	-	0.74	0.73	0.74	-
Liu et al. [21]		229 NC + 198 AD	0.92	0.91	0.93	0.95	-	-	-	-
Tong et al. [22]		231 NC + 238 sMCI + 167 pMCI + 198 AD	0.90	0.86	0.93	-	0.72	0.69	0.74	-
Suk et al. [50]	Deep Boltzmann machine [49] + patch-level engineered features	101 NC + 128 sMCI + 76 pMCI + 93 AD	0.92	0.92	0.95	0.97	0.72	0.37	0.91	0.73
Liu et al. [51]	Stacked auto-encoders [52] + region-level engineered features	204 NC + 180 AD	0.79	0.83	0.87	0.78	-	-	-	-
Shi et al. [66]	Deep polynomial network [67] + region-level engineered features	52 NC + 56 sMCI + 43 pMCI + 51 AD	0.95	0.94	0.96	0.96	0.75	0.63	0.85	0.72
Korolev et al. [68]	CNN + whole brain sMRI	61 NC + 77 sMCI + 43 pMCI + 50 AD	0.80	-	-	0.87	0.52	-	-	0.52
Khvostikov et al. [33]	CNN + hippocampal sMRI	58 NC + 48 AD	0.85	0.88	0.90	-	-	-	-	-
Our wH-FCN mehtod	Hierarchical FCN + automatic discriminative localizatoin	429 NC + 465 sMCI + 205 pMCI + 358 AD	0.90	0.82	0.97	0.95	0.81	0.53	0.85	0.78