Skip to main content

Table 6 Network architecture

From: Segmentation of finger tendon and synovial sheath in ultrasound image using deep convolutional neural network

Layers

Output size

D2FC-DN

Convolution

\(384 \times 192 \times 16\)

\(7 \times 7 {\text{conv}}\)

Dilated dense block

\(384 \times 192 \times 32\)

\(\left[ {3 \times 3 \,{\text{dilated conv}}} \right] \times 4, {\text{k}} = 8\)

Transition down

\(192 \times 96 \times 32\)

\(1 \times 1 {\text{conv}}\)

\(2 \times 2 {\text{max pooling, stride 2}}\)

Dilated dense block

\(192 \times 96 \times 80\)

\(\left[ {3 \times 3 \,{\text{dilated conv}}} \right] \times 5, k = 16\)

Transition down

\(96 \times 48 \times 64\)

\(1 \times 1 {\text{conv}}\)

\(2 \times 2 {\text{max pooling, stride 2}}\)

Dilated dense block

\(96 \times 48 \times 192\)

\(\left[ {3 \times 3 \,{\text{dilated conv}}} \right] \times 6, k = 32\)

Transition down

\(48 \times 24 \times 128\)

\(1 \times 1 \,{\text{conv}}\)

\(2 \times 2 {\text{max pooling, stride 2}}\)

Dilated dense block

\(48 \times 24 \times 448\)

\(\left[ {3 \times 3 \,{\text{dilated conv}}} \right] \times 7, k = 64\)

Transition up

\(96 \times 48 \times 448\)

\(3 \times 3\, {\text{transposed conv, stride 2}}\)

Transition up

\(192 \times 96 \times 192\)

\(3 \times 3\, {\text{transposed conv, stride 2}}\)

Transition up

\(384 \times 192 \times 96\)

\(3 \times 3\, {\text{transposed conv, stride 2}}\)

Convolution

\(384 \times 192 \times 32\)

\(3 \times 3 \,{\text{conv}}\)

Convolution

\(384 \times 192 \times 1\)

\(3 \times 3 \,{\text{conv}}\)