Competences in
  • Artifact Removal
  • Dose Reduction
  • Image Reconstruction
  • Scanner Calibration
  • Scanner Design
Products and Solutions


Backprojection brings the projection data from rawdata domain into image domain. Our series of hyper-fast backprojectors combines high fidelity with high performance. We offer backprojection algorithms for the spiral trajectory, for the circular trajectory with and without voxel-specific weighting functions, for arbitrary detector layouts, and even for exotic customer-specific trajectories. Our backprojectors guarantee real-time reconstruction for any existing CT algorithm (exact reconstruction, approximate Feldkamp-type reconstruction, single-slice rebinning type reconstruction, iterative reconstruction, deep learning-based reconstruction, etc.) Our RayConStruct-BP backprojectors are key to all image reconstruction pipelines that we offer.


Forward projection converts the volume into rawdata. Forward projectors are either adjunct to the backprojectors or allow to model the beam more precisely. Our forward projectors are typically needed for sophisticated artifact correction techniques or for iterative image reconstruction algorithms. In analogy to our backprojectors our RayConStruct-FP algorithms are highly performant to enable real-time image reconstruction and artifact correction.


Image reconstruction is our key know-how. We offer 2D, 3D, and higher dimensional analytical and iterative reconstruction pipelines for numerous applications in clinical and industrial CT. For example in the field of circular cone-beam CT our Feldkamp-type image reconstruction algorithms enable the reconstruction of full scan and partial scan data as well as overscan data. It further offers the reconstruction of data from offset detectors. For spiral CT a variety of fast algorithms is available and can be selected depending on the customer requirements and boundary constraints. In clinical CT, for example, ASSR-type algorithms or approximate Feldkamp-type algorithms are used. They combine high flexibility, arbitrary pitch selection, full dose usage with high image quality for standard and cardiac CT. Certain scan protocols can also be supported with exact cone-beam image reconstruction at the same performance level. Besides the third generation CT geometry RayConStruct-IR also covers the first, second and fourth generation CT geometry and can be easily extended to very exotic scanners which becomes especially attractive for non-destructive material testing applications, such as needed for luggage scanning.

Data detruncation, first and higher order beam hardening correction, metal artifact suppression, ring artifact reduction and other important algorithms are available as well. Calibration of raw detector data, defect pixel correction, calibration of beam quality to suppress beam hardening artifacts or to compute the material decomposition in dual-energy CT or multi-energy CT can also be part of your RayConStruct-IR solution.


Monte Carlo simulation of x-ray photon histories as they interact with the object or with the detector are needed for accurate scatter and dose estimation. Our Monte Carlo software is dedicated to computing the typical interactions occurring in the CT energy range between 1 keV and 1000 keV. Several performance optimizations built into RayConStruct-MC help to reduce the computation time for typical problems from hours to minutes. Further regularization approaches bring the algorithm to performing within seconds with only minor compromises in accuracy. RayConStruct-MC is not only able to compute the interactions with the object, but also those with the CT system.


Projection simulation computes straight line integrals through mathematically well-defined objects. These can be defined using constructive solid geometry, by specifying a voxel volume, by providing a surface model (e.g. STL) or by a combination of those. RayConStruct-PS is useful during the design stage of image reconstruction algorithms since it provides rawdata for scanners that are not yet physically available. One can further use the projection simulator to test existing algorithms and to selectively switch on and off certain physical effects. Polychromatic and monochromatic beams with arbitrary lateral beam profile are supported as well as scatter and off-focal radiation and detector afterglow effects.


Artificial intelligence refers to data-driven algorithms that are not handcrafted but rather adapt themselves to training data. Typically, AI algorithms today are realized as deep neural networks. RayConStruct-AI refers to a number of solutions that either make physics-based calculations really fast, such as x-ray dose or scatter estimation or to those that make an existing, but typically ill-posed problem, solvable, such as the reconstruction from very few projections, for example. It also refers to methods to generate or to augment training data, for example if measured training data are either not available or come without labels.


Deformable (and affine) registration is key to several approaches, even beyond medical applications. RayConStruct-DR, for example, can be useful to register a simulated object onto a measured object in order to prepare the data for further processing. RayConStruct-DR comprises CPU and GPU solutions of classical registration approaches and of deep learning-based ones.


Compute Platforms, AI Frameworks and Operating Systems

Our algorithms are dedicated to run on CPU and on GPU platforms, in single instances, multiple instances or distributed in clusters. We use the compute unified device architecture (CUDA) of NVIDIA and are able to offload computationally highly demanding components to a GPU device or to a GPU cluster.

Our AI-based solutions use the TensorFlow and under the Pytorch framework. Our software operates under Windows and Linux. Customer-specific solutions may also use other frameworks and operating systems.