Fig. 2.

ProST DeepNet framework for 2D/3D registration. A pelvis segmentation ${\mathit{V}}_{\text{seg}}$ is illustrated as 3D input. The rigid pose parameter ${\mathit{\theta }}_m$ is shown using RGB cross arrows. An example target X-ray image ${\mathit{I}}_t$ and moving image ${\mathit{I}}_m$ generated from ProST are presented. Forward pass follows the black arrows. Backward pass follows orange arrows, where the gradient $\frac{\partial S_{\text{net}}}{\partial {\theta }_m}$ is computed. Detailed structures of CrossViT and 3D CNN are illustrated in blocks on the right.