Model-Based Reconstruction for High-Spatial Resolution Cone-Beam Computed Tomography

点击翻译按钮获取中文摘要

Project Summary The main goal of this work is to improve image quality, particularly resolution, of flat-panel cone-beam CT (fpCBCT) systems. This modality is promising for its high resolution, adaptability to different geometries, and portability. In particular, fpCBCT is being investigated for use in bone-morphology quantification and microcal- cification detection, which are important in the study of osteoarthritis and detection breast cancer, respectively. These tasks require high resolution to visualize trabecular bone and microcalcifications, and while fpCBCT resolu- tion is superior to that of multi-detector CT, it is often unable to resolve these structures with sufficient detail. This work aims to improve resolution by modeling different system blur and noise properties, and incorporating these models in a model-based iterative reconstruction (MBIR) algorithm. MBIR methods have been increasingly pop- ular in tomography due to their ability to generate higher quality images than traditional analytical methods. This is largely due to the fact that MBIR methods include a noise model, a feature lacking in analytical methods such as filtered backprojection. A MBIR method with an accurate flat-panel-specific mathematical model will result in higher spatial resolution reconstructions. The following sources of blur will be measured and modeled on multiple fpCBCT test benches: the extended X-Ray focal spot, the detector scintillator, detector lag, and gantry motion. Noise correlation resulting from these blurs or readout electronics will also be measured and modeled. Particular attention will be paid to the shift-variant nature of these blurs which, along with noise correlation, has been tradi- tionally overlooked in current CT and fpCBCT reconstruction methods. These models will balance accuracy with computationally efficient so they may be used in iterative methods. Novel MBIR methods will be developed to in- corporate a wide range of system models, including those developed in this work. These algorithms will be used to reconstruct data acquired in simulation and on multiple fpCBCT test benches. The importance of each blur/noise model will be evaluated with a range of system properties and acquisition settings. Data will be reconstructed using the new, more accurate methods/models and traditional methods/models for comparison. Image quality will be assessed using a variety of metrics, including resolution, spatial noise, segmentation quality, modulation transfer functions, and noise power spectra. For example, the ability to segment trabecular bone will provide a clinically relevant image quality metric. Thus, this work will result in novel MBIR methods, detailed models of blur and noise correlation applicable to many fpCBCT systems, and a detailed analysis the image quality improve- ments resulting from these methods. While this work will focus on high-resolution imaging of bone morphology, it will improve the ability of fpCBCT to accomplish other high-resolution clinical tasks by improving resolution in current systems. Additionally, this work will provide a software solution to hardware induced image quality limita- tions, providing avenues not only to extend imaging performance in currently available system, but also to provide improved trade-offs and a way to relax hardware constraints in the design of future fpCBCT systems.