3D U-Net Architecture Global Interpretability
===================================================

*  Interpretability is achieved using 3D Gradient-Weighted Class Activation Maps (3D Grad-CAMs) for all layers within the architecture. Outputs for various functional elements are shown below.


----------------


*  Object shape error voxelization



.. figure:: voxel.png
   :align:   center

   *Fig 1: Object shape error voxelization*


----------------


*  Encoder with down-sampling kernels



.. figure:: encoder.png
   :align:   center

   *Fig 1: Encoder with down-sampling kernels*


----------------


*  Decoder with up-sampling kernels



.. figure:: decoder_interpret.png
   :align:   center

   *Fig 1: Decoder with up-sampling kernels*



----------------


*  Attention gate



.. figure:: attention.png
   :align:   center

   *Fig 1: Attention gate*


----------------


*  Residual connections



.. figure:: residuals.png
   :align:   center

   *Fig 1: Residual connections*