Table 1.
Parameter setting of each module of 3D DenseUNet.
| Block | Feature size | Convolution layer |
|---|---|---|
| Input | 224 × 224 × 8 | — |
| Convolution 1 | 112 × 112 × 4 | 7 × 7 × 7 × 96 conv |
| Pooling | 56 × 56 × 2 | 3 × 3 × 3 max pooling |
| Dense block 1 | 56 × 56 × 2 | (1 × 1 × 1 × 128 conv) × 3 (3 × 3 × 3 × 32 conv) × 3 |
| Transition layer 1 | 28 × 28 × 2 | 1 × 1 × 1 conv 2 × 2 × 1 average pooling |
| Dense block 2 | 28 × 28 × 2 | (1 × 1 × 1 × 128 conv) × 4 (3 × 3 × 3 × 32 conv) × 4 |
| Transition layer 2 | 14 × 14 × 2 | 1 × 1 × 1 conv 2 × 2 × 1 average pooling |
| Dense block 3 | 14 × 14 × 2 | (1 × 1 × 1 × 128 conv) × 12 (3 × 3 × 3 × 32 conv) × 12 |
| Transition layer 3 | 7 × 7 × 2 | 1 × 1 × 1 conv 2 × 2 × 1 average pooling |
| Dense block 4 | 7 × 7 × 2 | (1 × 1 × 1 × 128 conv) × 8 (3 × 3 × 3 × 32 conv) × 8 |
| Upsampling layer 1 | 7 × 7 × 2 | 2 × 2 × 1 × 504 upconv |
| Sum with dense block 3 | 14 × 14 × 2 | — |
| Upsampling layer 2 | 14 × 14 × 2 | 2 × 2 × 1 × 224 upconv |
| Sum with dense block 2 | 28 × 28 × 2 | — |
| Upsampling layer 3 | 28 × 28 × 2 | 2 × 2 × 1 × 192 upconv |
| Sum with dense block 1 | 56 × 56 × 2 | — |
| Upsampling layer 4 | 56 × 56 × 2 | 2 × 2 × 2 × 96 upconv |
| Sum with convolution 1 | 112 × 112 × 4 | — |
| Upsampling layer 5 | 224 × 224 × 8 | 2 × 2 × 2 × 64 upconv |
| Output | 224 × 224 × 8 | 1 × 1 × 1 × 3 conv |