helical_weighted_backproject

diffct.reconstruction.helical_weighted_backproject(sinogram, trajectory, volume_shape, view_weights, *, isocenter, detector_spacing=1.0, voxel_spacing=1.0)

Gather filtered cone data using replaceable (D,views) weights.

sinogram is a filtered CUDA (views,U,V) tensor. view_weights includes angular quadrature and redundancy and is applied AFTER filtering. Gradients flow to both tensors through the discrete cone-gather transpose. Fixed axial geometry, constant source radius and radial flat detectors are required. Caller-supplied weights own their angular coverage and truncation. One slice is processed per launch, avoiding a views-by-volume weight array; a fused slice-weight kernel can replace this if launch overhead dominates.