Table 2.

Summary of DL-based studies for prediction and classification of AD from MRI

Ref.	Reg.	DL Arch.	Pre-Proc.	Features	Dataset	Size	Accuracy
[33]	WB	SAE-3D, CNN	NM	CBF	ADNI	755 (AD, MCI, HC)	3-way 89.47%, AD vs. HC 95.39%, AD vs. MCI 86.84%, HC vs. MCI 92.11%
[31]		CNN	MC, STC, SS, HPF, SN, WMS, MD	SSIF	ADNI	52 AD${}^3$, 92 HC${}^3$, 211 AD${}^1$, 91 HC${}^1$	99.9%${}^3$, ${}^{5\alpha }$, 98.84%${}^1$, ${}^{5\alpha }$
[53]		CNN	MC, SST, HPF	SSIF	ADNI	28 AD, 15 NC	96.86%${}^{5\alpha }$
[43]		CNN	SN, BC, MD	CBF	ADNI	33 AD, 22 LMCI, 49 MCI, 45 HC	98.88%
[42]		CNN	INUC, DC, NM		ADNI	193 AD, 151 HC	Class Score 95%${}^{5\alpha }$
[69]		DNN		HPCV, CFV, LVV, ECT, MMSE	ADNI	60 AD, 60 HC, 60 cMCI 60 MCI	34.8%${}^{10\alpha }$
[36]		3D-CNN	CR, TE, IRE, IN	3D CBF	ADNI	199 AD, 141 NC; 3D MRI AD 600 NC 598	98.74%
[56]		3D-CNN	SST, NM	CBF	ADNI	50 AD, 43 LMCI, 77 EMCI, 61 NC
[8]		PNN	IR, WF	GLCM, SED	ADNI		85%
[70]		VAE, MLP	SG	Shape feature	ADNI	150 NC, 90 AD, 160 EMCI, 160 LMCI	NC-AD 84%, NC-EMCI 56%, NC-LMCI 59%. AD-EMCI 81%, AD-LMCI 57%, EMCI-LMCI 63%
[60]		DBN	VBM	VV 3611, MSD 24	OASIS	49 AD, 49 HC	MSD 0.7360${}^{10\alpha }$, VV 0.9176${}^{10\alpha }$
[47]		CNN	BE, MC, STC, IM, SS, THPF, NM, SN	CBF	ADNI	25 CN, 25 SMC, 25 EMCI, 25 LMCI, 13 MCI, 25 AD	CN 100%, SMC 96.85%, EMCI 97.38%, LMCI 97.43%, MCI 97.40%, AD 98.01%
[39]	BL	3D-CNN	NNM, BE, IR	4D features, clinical features	ADNI	192 AD, 184 HC, 181 pMCI, 228 sMCI	86%${}^{5\alpha }$
[38]	HPC	CNN	IR, SG	HPC shape, texture, CBF	ADNI-1, ADNI- GO&2, AIBL	ADNI: 1711, AIBL: 435
[71]		LSTM-RNN		LSTM-based features	ADNI-1, ADNI-GO&2	822 MCI
[40]	CSA	SAE-DNN	MC, NUC, IN, SST, VL	310 Vol., CorTh, SAF, 5000 FDCM	ADNI, CAD- Dementia	171 CN, 232 MCI, 101 AD	Model-1 ADNI 56.6%${}^{10\alpha }$, CAD-Dementia 51.4%${}^{10\alpha }$ Model-2 ADNI 58%${}^{10\alpha }$, CAD-Dementia 56.8%${}^{10\alpha }$
[41]	MCS	CNN	INUC	CBF	ADNI	47 AD 34 NC
[72]	SCS	CNN		CBF	OASIS	100 AD, 100 HC	VGG16: 92.3%${}^{5\alpha }$, Inception-V4: 96.25%${}^{5\alpha }$
[37]		CNN	NM, IR, MD	CBF	ADNI, Milan	ADNI: 294 PAD, 763 MCI, 352 HC Milan: 124 PAD, 50 MCI, 55 HC	ADNI: 99%${}^{10\alpha }$, MILAN: 98%${}^{10\alpha }$
[73]		CNN		CBF, 64	OASIS	416	80.25%
[46]	VB	CNN	SST, DA, CE, F	CBF	OASIS, MIRIAD	OASIS: 30 AD, 70 MCI, 316 HC MIRIAD: 46 MCI, 23 HC	0.8

Ref reference, Reg region, DL Arch deep learning architecture, Pre-Proc pre-processing technique used in the study, WB whole brain, BL-brain lobes HPC–hippocampus, CSA cortical surface area, MCS middle cross section, SCS single cross section, SSIF shift and scale-invariant features; Vol.-volume; CorTh-cortical thickness; SAF-surface area features; HPCV-hippocampal volumes; CFV-cerebrospinal fluid volume; LVV-lateral ventricle volume; ECT-entorhinal cortex thickness; MMSE-baseline scores of Mini-Mental State examination; $n\alpha$-n-fold cross-validation, 4DF 4D features, CF clinical features, GLCM gray-level co-occurrence matrix, SED Sobel edge detector, MSD-maximal self-dissimilarity, VV voxel values