Skip to main content
. 2023 Oct 13;10:1167500. doi: 10.3389/fcvm.2023.1167500

Figure 1.

Figure 1

Schematic representation of the modified 3D U-Net (37,38) architecture used in this study for segmentation. Up-sampling is performed through nearest-neighbor interpolation. Each arrow denotes a 3 × 3 × 3 convolutional layer, subsequently followed by a rectified linear unit (ReLU) and batch normalization (BN). The channel numbers are mentioned on top of the feature maps. The concluding layer employs a 1 × 1 × 1 convolution to streamline to 9 output channels. These channels are then subjected to voxel-wise classification via a SoftMax layer.