Name: Csaba Olasz
Affiliation: Department of Image Processing and Computer Graphics, University of Szeged
Primary research interest: transmission X-ray computed tomography
Title of the lecture: Artifact reduction in X-ray computed tomography
Keywords: computed tomography, artifact reduction, discretization, beam-hardening
Summary: transmission X-ray computed tomography (CT) is an imaging technique with which one can determine the inner structure of a three-dimensional unknown object from the projections of the object. CT scanners work in a non-destructive manner which makes them well suited for medical exemption and industrial quality control.
In the first part, I will introduce transmission X-ray computed tomography. I am going to define the projection of an object and show various settings to acquire multiple projections. I will detail two reconstruction algorithms using parallel beam geometry.
In the second part, I will present our results regarding the interpolation error caused by the discrete computer representation of the reconstruction problem.
In the third part, I will show our results in reducing the beam-hardening artifacts. We used U-net based fully convolutional neural networks to reduce beam hardening artifacts and electrical noise. Most of the research in this field proposed the usage of the neural network before (preprocessing) or after (postprocessing) the reconstruction. Our original neural network contained the reconstruction step as a non-trainable layer. As a consequence, reconstruction took place in the neural network.