Performance analysis of regularization algorithms used for image reconstruction in computed tomography

Computed tomography is one of the most significant diagnostic techniques in medicine. This work is focused on hard-field imaging, where signals take a form of straight rays and the reconstructed image can be presented as a matrix with unknown pixels. Algebraic methods for direct computation of the image have not been used in practice because of the scale of the problem and numerical errors appearing in the solution. The aim of this work was to analyse the performance of direct algebraic algorithms for tomographic image reconstruction including regularisation mechanism such as: generalised regularisation, Tikhonov regularisation, Twomey regularisation and ridge regression (RR), as well as comparing the results with the filtered backprojection (FBP) as the reference method. The performed analyses demonstrated that the regularised algebraic methods are more accurate than the commonly used FBP, and RR appeared the most precise among them. Additionally it was shown that the invariant system matrix (inverted during calculations) can be easily determined by solving the forward problem. Finally, potential directions of further research have been pointed out.