Penalized-Likelihood Reconstruction with High-Fidelity Measurement Models for High-Resolution Cone-Beam Imaging

摘要

We present a novel reconstruction algorithm based on a general cone-beam CTforward model which is capable of incorporating the blur and noise correlationsthat are exhibited in flat-panel CBCT measurement data. Specifically, theproposed model may include scintillator blur, focal-spot blur, and noisecorrelations due to light spread in the scintillator. The proposed algorithm(GPL-BC) uses a Gaussian Penalized-Likelihood objective function whichincorporates models of Blur and Correlated noise. In a simulation study, GPL-BCwas able to achieve lower bias as compared to deblurring followed by FDK aswell as a model-based reconstruction method without integration of measurementblur. In the same study, GPL-BC was able to achieve better line-pairreconstructions (in terms of segmented-image accuracy) as compared todeblurring followed by FDK, a model based method without blur, and a modelbased method with blur but not noise correlations. A prototype extremitiesquantitative cone-beam CT test bench was used to image a physical sample ofhuman trabecular bone. These data were used to compare reconstructions usingthe proposed method and model based methods without blur and/or correlation toa registered {\mu}CT image of the same bone sample. The GPL-BC reconstructionsresulted in more accurate trabecular bone segmentation. Multiple trabecularbone metrics, including Trabecular Thickness (Tb.Th.) were computed for eachreconstruction approach as well as the {\mu}CT volume. The GPL-BCreconstruction provided the most accurate Tb.Th. measurement, 0.255 mm, ascompared to the {\mu}CT derived value of 0.193 mm, followed by the GPL-Breconstruction, the GPL-I reconstruction, and then the FDK reconstruction(0.271 mm, 0.309 mm, and 0.335 mm, respectively).