High-quality computed tomography using advanced model-based iterative reconstruction

Computed Tomography (CT) is an essential technology for the treatment, diagnosis, and study of disease, providing detailed three-dimensional images of patient anatomy. While CT image quality and resolution has improved in recent years, many clinical tasks require visualization and study of structures beyond current system capabilities. Model-Based Iterative Reconstruction (MBIR) techniques offer improved image quality over traditional methods by incorporating more accurate models of the imaging physics. In this work, we seek to improve image quality by including high-fidelity models of CT physics in a MBIR framework. Specifically, we measure and model spectral effects, scintillator blur, focal-spot blur, and gantry motion blur, paying particular attention to shift-variant blur properties and noise correlations. We derive a novel MBIR framework that is capable of modeling a wide range of physical effects, and use this framework with the physical models to reconstruct data from various systems. Physical models of varying degrees of accuracy are compared with each other and more traditional techniques. Image quality is assessed with a variety of metrics, including bias, noise, and edge-response, as well as task specific metrics such as segmentation quality and material density accuracy. These results show that improving the model accuracy generally improves image quality, as the measured data is used more efficiently. For example, modeling focal-spot blur, scintillator blur, and noise iicorrelations enables more accurate trabecular bone visualization and trabecular thickness calculation as compared to methods that ignore blur or model blur but ignore noise correlations. Additionally, MBIR with advanced modeling typically outperforms traditional methods, either with more accurate reconstructions or by including physical effects that cannot otherwise be modeled, such as shift-variant focal-spot blur. This work provides a means to produce high-quality and high-resolution CT reconstructions for a wide variety of systems with different hardware and geometries, providing new tradeoffs in system design, enabling new applications in CT, and ultimately improving patient care