Logo image
Development of reconstruction algorithms to process MARS datasets
Doctoral Thesis   Open access

Development of reconstruction algorithms to process MARS datasets

Sikiru Afolabi Adebileje
Doctor of Philosophy - PhD, University of Otago
2021
Handle:
https://hdl.handle.net/10523/10860

Abstract

Spectral computed tomography Photon counting detector Ray tracing Reconstruction algorithms Cartesian coordinate reconstruction Cylindrical coordinate reconstruction
This thesis reports the development of reconstruction algorithms to process data collected using the MARS spectral CT scanner. This work is divided into these phases: the design of multiple digital datasets to test the outcome of our designed algorithms; the simulation of of those digital datasets; the implementation of a Cartesian reconstruction algorithm to act as a baseline for comparison and the development of a few variations of cylindrical coordinate reconstruction algorithms. I contributed to developing the design of multiple datasets which was used along the MARS simulation. The MARS simulation was developed using the Cartesian coordinate system. By using the Bresenham raytracing algorithm in Cartesian coordinates, we modified the existing MARS reconstruction method. This method acts as a baseline for comparison to other reconstruction algorithms that were developed during my thesis. The development of the cylindrical coordinate reconstruction algorithm is based on modelling voxels as nearly cubic shape within concentric rings from the center of the volume. We adapted the Bresenham raytracing method into cylindrical coordinates to develop an equivalent reconstruction algorithm to the Cartesian coordinate one from before. A variation to the cylindrical coordinate reconstruction algorithm was designed. This aimed to improve speed by avoiding the use of trigonometric functions in the main loop of the algorithm. Instead, using a lookup table, a nearest neighbour approximation was used during the raytracing. The final variation of the reconstruction algorithms aimed to act as a middleground between the speed of the nearest neighbour approximation and the quality of the exact Bresenham method. In this case, a different lookup table permitted using a linear approximation to estimate the raytracing path lengths. Using the simulated digital phantoms that I designed and the various experimental datasets obtained from the MARS scanner, the various reconstruction algorithms developed were compared for both performance and quality.
pdf
SikiruAdebileje2020PhD.pdf21.04 MBDownloadView

Metrics

1 Record Views

Details

Logo image