Table 2.
Architecture of the 3D convolutional neural network model.
| Block and kernel inputs | Settings | |
| Convolution 1 | ||
|
|
Conv3Da | (3,3,3,64) |
|
|
MaxPooling3Db | (2,2,2) |
|
|
BatchNormalizationc |
|
| Convolution 2 | ||
|
|
Conv3D | (3,3,3,64) |
|
|
MaxPooling3D | (2,2,2) |
|
|
BatchNormalization |
|
| Convolution 3 | ||
|
|
Conv3D | (3,3,3,128) |
|
|
MaxPooling3D | (2,2,2) |
|
|
BatchNormalization |
|
| Convolution 4 | ||
|
|
Conv3D | (3,3,3,256) |
|
|
MaxPooling3D | (2,2,2) |
|
|
BatchNormalization |
|
|
|
GlobalAveragePooling3Dd |
|
| Dense 1 | ||
|
|
Fully connected | 64 |
|
|
Dropout | 0.3 |
| Output | ||
|
|
Fully connected | 2 |
aConv3D: 3D convolutional layer.
bMaxPooling3D: 3D max pooling layer.
cBatchNormalization: batch normalization layer.
dGlobalAveragePooling3D: layer performing global average pooling for 3D data.