Astra-toolbox is an advanced framework for general CT reconstruction. However, It only provides the inferences of Python and MATLAB. Actually, it still uses C++ for its lower layer, and CUDA is also used for GPU acceleration. In Python inference, for instance, it used Cython for secondary decoration. In this repository, we do sort of works like "deconstruction", where we use C++ and XML to describe the CT reconstruction parameters directly.
In this repository, we focus on the way in Conebeam reconstruction, which is the most widely used measure in medical service. In Astra-toolbox, there are 2 ways to describe the geometry: Cone and Cone vec, where the details can be seen in 3D Geometry Documentation. Here we use XML to describe geometry parameters, the rule of XML structure can be inferred in functions createConeXMLDocument and createConeVecXMLDocument in this file. As to algorithms, we provide 4 algorithms including forward/backward projection, FDK, and SIRT. Astra-toolbox provides more algorithms in its implementation and we can easily extend some new algorithms based on CUDA.
Before use, you are recommended to install some extensions of C++ including rapidxml and boost. If you want to use GPU, please install CUDA tools and modify nvcc path in configure file
For more, in main, we provide a demo of Cone geometry and its forward/backward algorithm based on CUDA. The parameters are listed as follows:
If you think this repository is usable, welcome to mark it then you can look for this on your own page. If you have any questions, you can ask in Issue or contact me at quinnywang@foxmail.com, China.
W. van Aarle, W. J. Palenstijn, J. Cant, E. Janssens, F. Bleichrodt, A. Dabravolski, J. De Beenhouwer, K. J. Batenburg, and J. Sijbers, “Fast and Flexible X-ray Tomography Using the ASTRA Toolbox”, Optics Express, 24(22), 25129-25147, (2016), http://dx.doi.org/10.1364/OE.24.025129