DRR

Module for computing digitally reconstructed radiographs

DRR

DRR is a PyTorch module that compues differentiable digitally reconstructed radiographs. The viewing angle for the DRR (known generally in computer graphics as the camera pose) is parameterized by the following parameters:

  • SDD : source-to-detector distance (i.e., the focal length of the C-arm)
  • \(\mathbf R \in \mathrm{SO}(3)\) : a rotation
  • \(\mathbf t \in \mathbb R^3\) : a translation
Tip

DiffDRR can take a rotation parameterized in any of the following forms to move the detector plane:

If using Euler angles, the parameters are

  • alpha : Azimuthal angle
  • beta : Polar angle
  • gamma : Plane rotation angle
  • bx : X-dir translation
  • by : Y-dir translation
  • bz : Z-dir translation
  • convention : Order of angles (e.g., ZYX)

(bx, by, bz) are translational parameters and (alpha, beta, gamma) are rotational parameters.

source

DRR


def DRR(
    subject:Subject, # TorchIO wrapper for the CT volume
    sdd:float, # Source-to-detector distance (i.e., the C-arm's focal length)
    height:int, # Height of the rendered DRR
    delx:float, # X-axis pixel size
    width:int | None=None, # Width of the rendered DRR (default to `height`)
    dely:float | None=None, # Y-axis pixel size (if not provided, set to `delx`)
    x0:float=0.0, # Principal point X-offset
    y0:float=0.0, # Principal point Y-offset
    p_subsample:float | None=None, # Proportion of pixels to randomly subsample
    reshape:bool=True, # Return DRR with shape (b, 1, h, w)
    reverse_x_axis:bool=True, # If True, obey radiologic convention (e.g., heart on right)
    patch_size:int | None=None, # Render patches of the DRR in series
    renderer:str='siddon', # Rendering backend, either "siddon" or "trilinear"
    voxel_shift:float=0.5, # 0 or 0.5, depending if the voxel is at the top left corner or the center
    persistent:bool=True, # Set persistent value in `torch.nn.Module.register_buffer`
    compile_renderer:bool=False, # Compile the renderer for performance boost
    checkpoint_gradients:bool=False, # Checkpoint gradients to improve memory usage
    renderer_kwargs:VAR_KEYWORD
):

PyTorch module that computes differentiable digitally reconstructed radiographs.

The forward pass of the DRR module generated DRRs from the input CT volume. The pose parameters (i.e., viewing angles) from which the DRRs are generated are passed to the forward call.

source

DRR.render


def render(
    density:torch.tensor, # Volume from which to render DRRs
    source:torch.tensor, # World coordinates of X-ray source
    target:torch.tensor, # World coordinates of X-ray target
    mask_to_channels:bool=False, # If True, structures from the CT mask are rendered in separate channels
    kwargs:VAR_KEYWORD
):

source

DRR.forward


def forward(
    args:VAR_POSITIONAL, # Some batched representation of SE(3)
    parameterization:str=None, # Specifies the representation of the rotation
    convention:str=None, # If parameterization is Euler angles, specify convention
    calibration:RigidTransform=None, # Optional calibration matrix with the detector's intrinsic parameters
    mask_to_channels:bool=False, # If True, structures from the CT mask are rendered in separate channels
    degrees:bool=False, # If parameterization is Euler angles, use degrees instead of radians
    kwargs:VAR_KEYWORD
):

Generate DRR with rotational and translational parameters.

source

DRR.rescale_detector_


def rescale_detector_(
    scale:float
):

Rescale the detector plane (inplace).

source

DRR.perspective_projection


def perspective_projection(
    pose:RigidTransform, pts:torch.Tensor
):

Project points in world coordinates (3D) onto the pixel plane (2D).

source

DRR.inverse_projection


def inverse_projection(
    pose:RigidTransform, pts:torch.Tensor
):

Backproject points in pixel plane (2D) onto the image plane in world coordinates (3D).