Artifact estimation network for MR images: effectiveness of batch normalization and dropout layers

BMC Medical Imaging

Table 3 Layer structure of Network 3

Layer no.	Structure	Layer no.	Structure	Layer no.	Structure
1–1	Conv 1	13	Dropout 2 (0.5)	24–2	Conv 8
13	Conv 1	13	Upsampling 1	25	Upsampling 4
2	Max pooling 1	14	Batch normalization1	26	Batch normalization 4
3–1	Conv 2	15	Dropout 1 (0.5)	27	Dropout 4 (0.5)
3–2	Conv 2	16–1	Conv 6	28–1	Conv 9
4	Max pooling 2	16–2	Conv 6	28–2	Conv 9
5–1	Conv 3	17	Upsampling 2	29	Conv 10
5–2	Conv 3	18	Batch normalization 2	30	Regression (output)
6	Max pooling 3	19	Dropout 2(0.5)
7–1	Conv 4	20–1	Conv 7
7–2	Conv 4	20–2	Conv 7
8	Dropout 1 (0.5)	21	Upsampling 3
9	Max pooling 4	22	Batch normalization 3
10	Conv 5	23	Dropout 3 (0.5)
11	Conv 5	24–1	Conv 8

Batch normalization and dropout layers with a dropout rate of 0.5 were added after the ReLU activation function, which is the final layer of each decoder in U-Net

ISSN: 1471-2342