CT Image Quality Characterization

Abstract

This chapter presents mathematical and physics foundations of quantitative CT image quality assessment. The linear systems theory was applied to derive quantitative relations between noise, spatial resolution, and CT number of an idealized linear CT system and the associated hardware specifications, acquisition, and reconstruction parameters. The derivations not only highlighted the physical and mathematical foundations of several widely recognized image quality “laws” (e.g., the inverse scaling law between CT noise variance and radiation exposure level) but also the underlying assumptions behind these laws. Next, the chapter showed how these assumptions can be violated in realistic modern CT systems, even for those employing the linear filtered backprojection (FBP) algorithm. It demonstrated how the classical image quality models can be revised for modern systems. One revision is the introduction of a virtual image object with negative attenuation to compensate for the spatially and directionally varying impacts of the bowtie filter to CT noise power spectrum (NPS). After its revision, linear systems theory can still be used to model CT image quality of FBP-based realistic CT systems. Finally, the chapter demonstrated severe violations of the classical linear systems theory in CT systems with nonlinear model-based iterative reconstruction (MBIR) algorithms and the new challenges in estimating image quality properties for these systems. The chapter presented strategies toward addressing these challenges, including ensemble statistics- and task-based image quality assessment methods as well as empirical modeling of the relationship between image quality performance and system parameters. It also clarified the concept of NPS and justified the applicability of this objective image quality metric in nonlinear MBIR-based systems with nonstationary noise. Throughout this chapter, discussion on the topic of CT image quality was accompanied by illustrations and plots of experimental data, some of which are published for the first time.