Ertuğrul Umut Yıldırım, M.Sc.
Department of Scientific Computing
February 2022
Supervisor: Ömür Uğur (Institute of Applied Mathematics, Middle East Technical University, Ankara)
Co-Supervisor: Guenther Glatz (Petroleum Engineering, King Fahd University of Petroleum & Minerals (KFUPM), Saudi Arabia)
Abstract
Computed tomography has been widely used in clinical and industrial applications as a non-destructive visualization technology. The quality of computed tomography scans has a strong effect on the accuracy of the estimated physical properties of the investigated sample. X-ray exposure time is a crucial factor for scan quality. Ideally, long exposure time scans, yielding large signal-to-noise ratios, are available if physical properties are to be delineated. However, especially in micro-computed tomography applications, long exposure times constitute a problem for monitoring some physical processes that are happening quickly. To alleviate this problem, this thesis proposes a convolutional neural network approach for scan quality enhancement allowing for a reduction in X-ray exposure time while improving signal-to-noise ratio of the scanned image simultaneously. Moreover, the impact of using different loss functions, namely the mean squared error and the structural similarity index measure, on the performance of the network is analyzed. Both the visual and quantitative assessments show that the trained network greatly improves the quality of low-dose scans
Keywords: convolutional neural networks, computed tomography, denoising