Model-based iterative reconstruction algorithms for computed tomography

Computed Tomography (CT) is a powerful tool for non-destructive imaging in which an object’s interior is visualized by reconstructing a set of projection images. The technique can be applied in various modalities, ranging from a typical X-ray CT scanner to electron microscopy and synchrotron beamlines. Often, only limited projection data is available, which makes the reconstruction process more dicult and results in reconstruction artifacts if standard techniques are employed. Limited data problems can arise in a variety of applications. In medical CT, the acquisition of only a limited number of projections is benecial to reduce the radiation dose delivered to the patient. In electron tomography, the sample can only be rotated over a limited tilt range due to mechanical constraints and the number of acquisition angles is often relatively small to avoid beam damage. In dynamic CT, the time to acquire sucient projection data over the full angular range is often long in comparison to the time interval in which substantial changes inside the scanned sample occur. This implies that, in order to avoid blurry reconstructions due to the time-varying nature of the sample, only a limited amount of projections can be acquired per time frame, resulting in a limited data problem. In this thesis, various improved reconstruction algorithms are proposed, all with the goal of achieving adequate image quality with only few projection images. The improvement is mainly due to the introduction of local models, specic to the problem at hand, into the reconstruction process. By more accurately modelling the sample, the reconstruction problem becomes more determined, which generally results in improved reconstruction quality.