CTOperators.jl

Matched forward and back projectors for X-ray CT on CPUs and GPUs.

CTOperators.jl provides projection operators for tomographic reconstruction, written once with KernelAbstractions.jl and run multi-threaded on the CPU and on GPUs (CUDA, AMD, Apple). An operator holds its geometry and workspace and is reused by every application, the setting of iterative solvers such as those of RegularizedLeastSquares.jl.

Warning

CTOperators.jl is in an early stage of development and not registered yet; its interface may still change.

Features

Projectors for iterative reconstruction: geometry and workspace are set up once; the back projection is the exact transpose of the forward projection.
Two ray-driven projectors: Siddon's method (in the incremental form of Jacobs et al.), exact intersection lengths of rays and voxels, and Joseph's method, linear interpolation between the voxel centres, see Siddon's method and Joseph's method. Footprint-based projectors are planned.
Generic geometries: arbitrary rays given by start and end points, and per-view descriptions of cone-beam (flat or curved detector) and parallel-beam scanners with any trajectory: circular, helical, or a moving object. Geometries of Sinograms.jl work directly.
One code base for CPU and GPU.
Differentiable with respect to image and projection data (ChainRulesCore rules).

Where to go next

Getting Started simulates a helical cone-beam scan and reconstructs it with RegularizedLeastSquares.jl.
The how-to guides show how to set up a geometry, including an object that moves during the scan, how to run on a GPU, how to build ordered-subset methods and how to differentiate through a projector and how to use geometries of Sinograms.jl.
The explanation pages describe the coordinates and conventions and the projection methods, Siddon's method and Joseph's method.

Installation

using Pkg
Pkg.add(url="https://github.com/JuliaImageRecon/CTOperators.jl")