Fine-Tuning Parameters in CT

Abstract: Computed tomography (CT) is a medical imaging technique that usesX-rays to obtain a reconstruction of an object. The term acquisition ge-ometry refers to the arrangement of imaging sensors and the X-ray sourceas well as the procedure used for data collection. The quality of the re-construction is often limited by the acquisition geometry and parametervalues. In this thesis, we present a procedure for fine-tuning acquisitiongeometry parameters in CT by minimizing the difference between theforward projection of a known phantom and measured data, i.e. data dis-crepancies. We extend the ODL library in Python and create acquisitiongeometries where different parameters have been distorted. We utilizegradient descent in an attempt recover the true parameters of the acquisi-tion geometries. Our results show that the recovery of the true geometryis successful when one or, in some cases, two parameters are perturbed.The objective function becomes very sensitive when more parameters areperturbed, requiring a low learning rate and making convergence slow.Nevertheless, we are able to minimize the objective function in the for-ward projection for all perturbations. Although our algorithm performswell in some aspects relating to parameter recovery, there is potential forfurther research by implementing other optimization methods.