Characterization of cutting-edge technologies in computed tomography
In this thesis, various of the above-mentioned developments in CT technology are technically evaluated, their performance on the resulting patient image quality characterized, and their effect on the dose to the patient determined. The first part of the thesis contains the technical evaluation of three developments in CT technology. In chapter 2, the first ultra-high-resolution CT (UHRCT) system is evaluated. The focus for this evaluation was mainly on the spatial resolution, noise magnitude, and dose used compared to a conventional state-of-the-art clinical CT system. Another technical evaluation is performed on a dual-energy tube-voltage-switching CT system that uses a deep-learning approach to fill in the missing sinograms. This evaluation is described in chapter 3. Currently, as described above, a CT scan starts with the acquisition of 2D localizer images. The effect on radiation dose to the patient of the replacement of these 2D localizer images by a heavily-filtered 3d localizer is determined in chapter 4. For this, a Monte Carlo simulation was developed, validated, and used to characterize the dose involved in these different localizer strategies. In the second part of this thesis, objective and subjective image quality comparison are performed on a DLR technique. In chapter 5, DLR for non-contrast cerebral CT imaging is evaluated. Since the contrast between white and grey matter in cerebral CT is very limited, the potential advantage of DLR, which results in images with lower noise magnitude compared to those generated by FBP or HIR, could have the clinical impact in the reconstruction of this type of acquisitions. Commonly, while reading CT acquisitions thick sections (3-5 mm) are used. High noise, thin section reconstructions (0.5-1 mm) are then used as problem solvers, if in doubt, when reading thick sections. Due to deep-learning based reconstructions having a low noise magnitude, this might allow for the interpretation of thin sections only for standard diagnostic purposes. Therefore, in chapter 6, the image quality of these thin, deep learning-based reconstructed sections is compared to that of both thin and thick sections reconstructed with HIR. One of the last applications in CT imaging in which FBP is still the default reconstruction technique is calcium scoring in the coronary arteries. Other reconstruction techniques, which could result in the use of lower doses, are not used since the calculation of the Agaston score is based on FBP. In chapter 7, the use of various reconstruction methods (FBP, HIR, MBIR, and DLR) is evaluated for cardiac risk scoring. Finally, in the last part, chapter 8, noise texture differences in CT images are researched. As mentioned above, iterative reconstruction can lead to images with “blotchy” or “plasticky” appearance. It is suggested in literature that the different noise texture present in reconstruction methods other than FBP can explain this appearance. As a first step into researching this phenomenon, the detectability of noise texture changes by the human visual system was investigated in a series of experiments. Furthermore, the results presented in this thesis are compared to results presented in published manuscripts and also the impact on clinical CT imaging is discussed.