This paper presents CTrex, a research-oriented software framework for micro-computed tomography (micro-CT). The authors built a system that gives researchers direct control over the low-level operators used to make tomographic images. The goal is to support imaging situations that break assumptions behind standard reconstruction methods, such as object motion, deformation, truncated fields of view, continuous rotation, or other unconventional scan protocols.

Existing reconstruction packages tend to treat projection and backprojection operators as fixed black boxes. CTrex exposes the full projection–correction–backprojection pipeline down to the GPU (graphics processing unit) kernel level. In plain terms, users can view and edit the building blocks that turn recorded X-ray views into a 3D image. That lets non-ideal acquisition effects be modeled inside the numerical operators rather than handled only as separate corrections.

The framework combines GPU acceleration with a flexible geometry description based on homogeneous-coordinate view matrices. These matrices let the code express rigid and affine transformations, detector misalignments, and time-varying geometries in a single consistent form. CTrex is also designed to be trajectory-agnostic on the kernel side: it supports circular, helical, offset, and conveyor-belt scan paths while keeping the low-level kernels independent of the specific trajectory definition.

The authors illustrate several kinds of applications that benefit from operator-level control. Examples in the abstract include motion- and deformation-aware reconstruction, event-based four-dimensional (4D) imaging, reconstruction in cylindrical coordinates, and extended-field-of-view CT. The paper argues that these strategies can be difficult to realize in conventional frameworks because of limited access to how projection operators are discretized and evaluated.