Analysis of iterative reconstruction algorithms for digital tomosynthesis

OData support
Supervisor:
Dr. Horváth Gábor
Department of Measurement and Information Systems

I summarized the basic tomosynthesis algorithms in my thesis. These methods are capable to create a three dimensional volumetric model from multiply X-ray images taken from different angles. The volumetric model is presented in coronal slices. I focused on the iterative methods (ART), which improves the model step by step. I implemented the filtered back projection, shift and add and some basic ART algorithms on GPGPU with OpenCL API. The implemented SAA, FBP, Gordon ART, SART, SIRT, and tow variants of MART are compared by calculation time and by the error of the final model. I have created an optimized algorithm chain which finishes the reconstruction in tow minutes whith 30 iteration on the given hardware (GeForce GT 770 4GB, Intel Core i7). I also examined the post processing algorithms. The research shown the MART algorithm converges fastest, the second in speed is the Gordon ART according to learning curves. I have made some attempt to improve the model by using weighting on error images, but it doesn't help.

Downloads

Please sign in to download the files of this thesis.