Optimization of image quality and radiation dose in neuroradiological computed tomography

Abstract: Background: The goal of clinical computed tomography (CT) is to produce images of diagnostic quality using the lowest possible radiation exposure. Degradation of image quality, with increased image noise and reduced spatial resolution, is a major limitation for radiation dose reduction in CT. This can be counteracted with new post-processing image filters and iterative reconstruction (IR) algorithms that improve image quality and allow for reduced radiation doses. Implementation of new methods in clinical routine requires prior validation in phantoms and clinical feasibility studies including comprehensive evaluation of diagnostic image quality. Aims: The main objectives of this thesis were to assess new methods for improvement of image quality in CT, explore the associated potential for radiation dose reduction, and to outline a comprehensive approach for evaluation of image quality. Methods: Extensive phantom testing was performed and a total of 100 human subjects were included in the clinical studies. Image quality and diagnostic acceptability were assessed in brain CT acquired with 30% reduced radiation dose in combination with post-processing filter (Paper I) and IR (Paper II). The potential of IR for image quality improvement, without concomitant radiation dose reduction, was assessed in craniocervical CT angiography (CTA) (Paper III). The performance of six IR algorithms was evaluated in a brain CT phantom model (Paper IV), using different combinations of radiation dose levels and iterative image optimization levels. Throughout the studies, various approaches for subjective and objective evaluation of image quality were used and assessed. Results: Post-processing image filtering (Paper I) and IR (Paper II) compensated partly or entirely for the loss of image quality caused by 30% reduced radiation dose in brain CT. In both studies, considerable inter-observer variation was seen. In Paper II a discrepancy was seen between results of objective and subjective evaluation of image quality and also between grading and ranking, indicating observer bias. Statistical IR improved image quality in craniocervical CTA (Paper III) with fairly good inter-observer agreement. Despite having different strengths and weaknesses, the six iterative reconstruction algorithms evaluated in Paper IV all improved image quality. Best over-all improvement was seen for one of the model-based IR algorithms, especially at lower radiation doses. Conclusion: All evaluated methods improved image quality and showed potential for radiation dose reduction while maintaining diagnostic quality. Careful study design and comprehensive evaluation of image quality including objective and subjective evaluation steps may reduce observer bias and improve reliability of study results.