Image processing for MARS spectral CT

This thesis reports on the developments made to give improvements in the MARS image reconstruction chain. The work described in this thesis helps all MARS users to produce high-quality material images that can offer an improved diagnosis in the medical imaging field. The thesis describes two main areas of research. Firstly, an existing algorithm was implemented in C++ to calculate the number of photons for each pixel position across the detector used in the MARS system. Secondly, speed and usability improvements were made to the MARS material decomposition (MD) software. The thesis describes the coding aspect of the x-ray source model behaviour of the MARS scanner according to a parameterised semi-analytic source model developed using a series of Monte Carlo simulations. Photon counts are calculated at each pixel using these parameters and implemented the source model in a form that is compatible with the existing image reconstruction software. However, some additional changes will need to be applied to the current reconstruction algorithm to support the high energy resolution provided by the model. This development will help to accurately provide the energy and position of incident photon counts, which is required for future polychromatic material reconstruction algorithms. The developed source model is now being used in simulation software that can produce a MARS-like data-set from an ideal set of material volumes. The thesis also details the improvements made in the current MARS MD software package that is used by the current MARS scanner users as a primary material analysis tool. Other members of the team are investigating hybrid material reconstruction, which combines the MD and reconstruction into a two-step process. In this case, the speed of the MD software is crucial as it would be run per-iteration and our current reconstruction has 900 iterations. The former MD algorithm uses non-negative least square equations to find the approximate solutions. The algorithm used in this thesis eliminates the need for iterations in the non-negative least square equations via direct analytic methods. The work completed in the MARS MD software led to some improvement in the performance of the algorithm and also a reduction in the computation times. The results of the developed version were also compared with the former MD to confirm the efficiency of the software