Table 5 Network analysis result of 3D U-net layers with reference to Figs. 7 and 8
Sr | Name | Type | Activations | Learnable | Total learnable |
|---|---|---|---|---|---|
1 | Input 64 × 64 × 64 × 4 images | 3-D Image Input | 64 × 64 × 64 × 4 | – | 0 |
2 | en1_conv1 32 3 × 3 × 3 × 4 convolution with stride [1 1 1] and padding ‘same’ | Convolution | 64 × 64 × 64 × 32 | Weights 3 × 3 × 3 × 4 × 32 Bias 1 × 1 × 1 × 32 | 3488 |
3 | en1_bn1 Batch normalization with 32 channels | Batch Normalization | 64 × 64 × 64 × 32 | Offset 1 × 1 × 1 × 32 Scale 1 × 1 × 1 × 32 | 64 |
4 | en1_relu1 ReLU | ReLU | 64 × 64 × 64 × 32 | – | 0 |
5 | en1_conv2 64 3 × 3 × 3 × 32 convolution with stride [1 1 1] and padding ‘same’ | Convolution | 64 × 64 × 64 × 64 | Weights 3 × 3 × 3 × 32 × 64 Bias 1 × 1 × 1 × 64 | 55,360 |
6 | en1_relu2 ReLU | ReLU | 64 × 64 × 64 × 64 | – | 0 |
7 | en1_maxpool 2 × 2 × 2 max pooling with stride [2 2 2] and padding ‘same’ | 3-D Max Pooling | 32 × 32 × 32 × 64 | – | 0 |
8 | en2_conv1 64 3 × 3 × 3 × 64 convolution with stride [1 1 1] and padding ‘same’ | Convolution | 32 × 32 × 32 × 64 | Weights 3 × 3 × 3 × 64 × 64 Bias 1 × 1 × 1 × 64 | 110,656 |
9 | en2_bn1 Batch normalization with 64 channels | Batch Normalization | 32 × 32 × 32 × 64 | Offset 1 × 1 × 1 × 64 Scale 1 × 1 × 1 × 64 | 128 |
10 | en2_relu1 ReLU | ReLU | 32 × 32 × 32 × 64 | – | 0 |
11 | en2_conv2 128 3 × 3 × 3 × 64 convolution with stride [1 1 1] and padding ‘same’ | Convolution | 32 × 32 × 32 × 128 | Weights 3 × 3 × 3 × 64 × 128 Bias 1 × 1 × 1 × 128 | 221,312 |
12 | en2_relu2 ReLU | ReLU | 32 × 32 × 32 × 128 | – | 0 |
13 | en2_maxpool 2 × 2 × 2 max pooling with stride [2 2 2] and padding ‘same’ | 3-D Max Pooling | 16 × 16 × 16 × 128 | – | 0 |
14 | en3_conv1 128 3 × 3 × 3 × 128 convolution with stride [1 1 1] and padding ‘same’ | Convolution | 16 × 16 × 16 × 128 | Weights 3 × 3 × 3 × 128 × 128 Bias 1 × 1 × 1 × 128 | 442,496 |
15 | en3_bn1 Batch normalization with 128 channels | Batch Normalization | 16 × 16 × 16 × 128 | Offset 1 × 1 × 1 × 128 Scale 1 × 1 × 1 × 128 | 256 |
16 | en3_relu1 ReLU | ReLU | 16 × 16 × 16 × 128 | – | 0 |
17 | en3_conv2 256 3 × 3 × 3 × 128 convolution with stride [1 1 1] and padding ‘same’ | Convolution | 16 × 16 × 16 × 256 | Weights 3 × 3 × 3 × 128 × 256 Bias 1 × 1 × 1 × 256 | 884,992 |
18 | en3_relu2 ReLU | ReLU | 16 × 16 × 16 × 256 | – | 0 |
19 | en3_maxpool 2 × 2 × 2 max pooling with stride [2 2 2] and padding ‘same’ | 3-D Max Pooling | 8 × 8 × 8 × 256 | – | 0 |
20 | de4_conv1 256 3 × 3 × 3 × 256 convolution with stride [1 1 1] and padding ‘same’ | Convolution | 8 × 8 × 8 × 256 | Weights 3 × 3 × 3 × 256 × 256 Bias 1 × 1 × 1 × 256 | 1,769,728 |
21 | de4_relu1 ReLU | ReLU | 8 × 8 × 8 × 256 | – | 0 |
22 | de4_conv2 512 3 × 3 × 3 × 256 convolution with stride [1 1 1] and padding ‘same’ | Convolution | 8 × 8 × 8 × 512 | Weights 3 × 3 × 3 × 256 × 512 Bias 1 × 1 × 1 × 512 | 3,539,456 |
23 | de4_relu2 ReLU | ReLU | 8 × 8 × 8 × 512 | – | 0 |
24 | de4_transconv 512 2 × 2 × 2 × 512 transposed 3D convolutions with stride [2 2 2] and cropping [0 0 0; 0 0 0] | Transposed Convolution 3D | 16 × 16 × 16 × 512 | Weights 2 × 2 × 2 × 512 × 512 Bias 1 × 1 × 1 × 512 | 2,097,664 |
25 | concat3 Concatenation of 2 inputs along dimension 4 | Concatenation | 16 × 16 × 16 × 768 | – | 0 |
26 | de3_conv1 256 3 × 3 × 3 × 768 convolution with stride [1 1 1] and padding ‘same’ | Convolution | 16 × 16 × 16 × 256 | Weights 3 × 3 × 3 × 758 × 256 Bias 1 × 1 × 1 × 256 | 5,308,672 |
27 | de3_relu1 ReLU | ReLU | 16 × 16 × 16 × 256 | – | 0 |
28 | de3_conv2 256 3 × 3 × 3 × 256 convolution with stride [1 1 1] and padding ‘same’ | Convolution | 16 × 16 × 16 × 256 | Weights 3 × 3 × 3 × 256 × 256 Bias 1 × 1 × 1 × 256 | 1,769,728 |
29 | de3_relu2 ReLU | ReLU | 16 × 16 × 16 × 256 | – | 0 |
30 | de3_transconv 256 2 × 2 × 2 × 256 transposed 3D convolutions with stride [2 2 2] and cropping [0 0 0; 0 0 0] | Transposed Convolution 3D | 32 × 32 × 32 × 256 | Weights 2 × 2 × 2 × 256 × 256 Bias 1 × 1 × 1 × 256 | 524,544 |
31 | concat2 Concatenation of 2 inputs along dimension 4 | Concatenation | 32 × 32 × 32 × 384 | – | 0 |
32 | de2_conv1 128 3 × 3 × 3 × 384 convolution with stride [1 1 1] and padding ‘same’ | Convolution | 32 × 32 × 32 × 128 | Weights 3 × 3 × 3 × 384 × 128 Bias 1 × 1 × 1 × 128 | 1,327,232 |
33 | de2_relu1 ReLU | ReLU | 32 × 32 × 32 × 128 | – | 0 |
34 | de2_conv2 128 3 × 3 × 3 × 128 convolution with stride [1 1 1] and padding ‘same’ | Convolution | 32 × 32 × 32 × 128 | Weights 3 × 3 × 3 × 128 × 128 Bias 1 × 1 × 1 × 128 | 442,496 |
35 | de2_relu2 ReLU | ReLU | 32 × 32 × 32 × 128 | – | 0 |
36 | de2_transconv 128 2 × 2 × 2 × 128 transposed 3D convolutions with stride [2 2 2] and cropping [0 0 0; 0 0 0] | Transposed Convolution 3D | 64 × 64 × 64 × 128 | Weights 2 × 2 × 2 × 128 × 128 Bias 1 × 1 × 1 × 128 | 131,200 |
37 | concat1 Concatenation of 2 inputs along dimension 4 | Concatenation | 64 × 64 × 64 × 192 | – | 0 |
38 | de1_conv1 64 3 × 3 × 3 × 192 convolution with stride [1 1 1] and padding ‘same’ | Convolution | 64 × 64 × 64 × 64 | Weights 3 × 3 × 3 × 192 × 64 Bias 1 × 1 × 1 × 64 | 331,840 |
39 | de1_relu1 ReLU | ReLU | 64 × 64 × 64 × 64 | – | 0 |
40 | de1_conv2 64 3 × 3 × 3 × 64 convolution with stride [1 1 1] and padding ‘same’ | Convolution | 64 × 64 × 64 × 64 | Weights 3 × 3 × 3 × 64 × 64 Bias 1 × 1 × 1 × 64 | 110,656 |
41 | de1_relu2 ReLU | ReLU | 64 × 64 × 64 × 64 | – | 0 |
42 | convlast 2 1 × 1 × 1 × 64 convolution with stride [1 1 1] and padding ‘same’ | Convolution | 64 × 64 × 64 × 2 | Weights 1 × 1 × 1 × 64 × 2 Bias 1 × 1 × 1 × 2 | 130 |
43 | softmax softmax | Softmax | 64 × 64 × 64 × 2 | – | 0 |
44 | Output Dice loss | Classification Output | – | – | 0 |