Research on Classification Method of Medical Ultrasound Image Processing Based on Neural Network

Table 1

Structure of CNN15 used in this study.

Layer	Input	Filter	Padding	Stride	Output
Conv1	353 × 353, 2	13 × 13, 96	6 × 6	2 × 2	177 × 177, 96
Max-pooling1	177 × 177, 96	3 × 3	1 × 1	2 × 2	89 × 89, 96
Conv2a	89 × 89, 96	5 × 5, 256	2 × 2	2 × 2	45 × 45, 256
Conv2b	45 × 45, 256	5 × 5, 256	2 × 2	1 × 1	45 × 45, 256
Max-pooling2	45 × 45, 256	3 × 3	0 × 0	2 × 2	22 × 22, 256
Conv3	22 × 22, 256	3 × 3, 384	1 × 1	1 × 1	22 × 22, 384
Conv4	22 × 22, 384	3 × 3, 384	1 × 1	1 × 1	22 × 22, 384
Conv5a	22 × 22, 384	3 × 3, 384	1 × 1	1 × 1	22 × 22, 384
Conv5b	22 × 22, 384	3 × 3, 384	1 × 1	1 × 1	22 × 22, 384
Conv5c	22 × 22, 384	3 × 3, 384	1 × 1	1 × 1	22 × 22, 384
Conv5d	22 × 22, 384	3 × 3, 384	1 × 1	1 × 1	22 × 22, 384
Conv5e	22 × 22, 384	3 × 3, 384	1 × 1	1 × 1	22 × 22, 384
Conv5f	22 × 22, 384	3 × 3, 384	1 × 1	1 × 1	22 × 22, 384
Conv5g	22 × 22, 384	3 × 3, 384	1 × 1	1 × 1	22 × 22, 384
Conv5h	22 × 22, 384	3 × 3, 384	1 × 1	1 × 1	22 × 22, 384
Conv5i	22 × 22, 384	3 × 3, 256	1 × 1	1 × 1	22 × 22, 256
DS	22 × 22, 256	—	—	—	44 × 44, 64
Conv6	44 × 44, 64	3 × 3, 1	1 × 1	1 × 1	44 × 44, 1

conv denotes convolutional layer; DS denotes double size layer; the format of the data in the table is size × number.