Automatic Diagnosis of Alzheimer’s Disease and Mild Cognitive Impairment Based on CNN + SVM Networks with End-to-End Training

Table 1

The architecture of 3DCNN designed in the paper.

Layer ID	Layer	Kernel number	Kernel size/stride	Output size
0	Input			1 × 80 × 100 × 76
1	Conv1	32	(1, 1, 1)/1	32 × 80 × 100 × 76
2	Conv2	64	(3, 3, 3)/1	64 × 80 × 100 × 76
3	MaxPool3D		(2, 2, 2)/2	64 × 40 × 50 × 38
4	Conv3	128	(3, 3, 3)/1	128 × 40 × 50 × 38
5	Attention			128 × 40 × 50 × 38
6	Maxpool3D		(2, 2, 2)/2	128 × 20 × 25 × 19
7	Conv4	256	(3, 3, 3)/1	256 × 20 × 25 × 19
8	Attention			256 × 20 × 25 × 19
9	Maxpool3D		(2, 2, 2)/2	256 × 10 × 12 × 9
10	Conv5	512	(3, 3, 3)/1	512 × 10 × 12 × 9
11	Attention			512 × 10 × 12 × 9
12	Maxpool3D		(2, 2, 2)/2	512 × 5 × 6 × 4
13	Conv6	512	(3 × 3 × 3)/1	512 × 3 × 4 × 2
14	GAP			512 × 1 × 1 × 1
15	Flatten			512
16	FC			2