This thesis focuses on using Computer-Assisted Design (CAD) data priors to infer missing X-Ray measurements instead of sampling them. X-Ray Computed Tomography (XCT) is a non-destructive imaging technique that produces cross-sectional, volumetric images of bodies sensitive to X-Ray. It relies on the repeated sampling of the body from different points of view and on numerical methods to obtain an image from the measurements. Due to the models on which the computations rely, that necessarily approximate the actual object and sampling process, reconstructing the causal factor that produced the sequence of measurements, the sinogram, is a non-trivial task. Indeed, it involves solving an ill-posed inverse problem, which becomes under-determined under certain circumstances that depends on the object’s material or else the sampling constraints. Several approaches to this problem have been proposed, but they primarily focus on reconstruction algorithms and overlook certain contingencies of the sampling process. Specifically, reconstruction schemes need to account for the opacity of specific object’s components to X-Ray, objects of significant size that cannot fit fully in CT-scanner’s gantries and scanning time in non-destructive testing. Reconstruction schemes designed for data acquired in such scenarios are yet to be proposed and implemented at scale, and crucially, no pre-processing of the raw data from the detector has been thoroughly investigated. In this thesis, we investigate various techniques to provide an initial estimate of missing and impaired measurements, given prior knowledge about the sampled object. This goal stems from the availability of CAD models in engineering, as well as the recent developments of machine-learning tools allowing to learn cross-modality dependencies between CAD priors and actual measurements.