SU‐G‐IeP4‐03: Cone Beam X‐Ray Luminescence Computed Tomography Based On Generalized Gaussian Markov Random Field

Abstract

Purpose:Cone beam X‐ray luminescence computed tomography (CB‐XLCT), which aims to achieve molecular and functional imaging by X‐rays, has recently been proposed as a new imaging modality. However, the inverse problem of CB‐XLCT is seriously ill‐conditioned, hindering us to achieve good image quality. In this work, a novel reconstruction method based on Bayesian theory is proposed to tackle this problemMethods:Bayesian theory provides a natural framework for utilizing various kinds of available prior information to improve the reconstruction image quality. A generalized Gaussian Markov random field (GGMRF) model is proposed here to construct the prior model of the Bayesian theory. The most important feature of GGMRF model is the adjustable shape parameter p, which can be continuously adjusted from 1 to 2. The reconstruction image tends to have more edge‐preserving property when p is slide to 1, while having more noise tolerance property when p is slide to 2, just like the behavior of L1 and L2 regularization methods, respectively. The proposed method provides a flexible regularization framework to adapt to a wide range of applications.Results:Numerical simulations were implemented to test the performance of the proposed method. The Digimouse atlas were employed to construct a three‐dimensional mouse model, and two small cylinders were placed inside to serve as the targets. Reconstruction results show that the proposed method tends to obtain better spatial resolution with a smaller shape parameter, while better signal‐to‐noise image with a larger shape parameter. Quantitative indexes, contrast‐to‐noise ratio (CNR) and full‐width at half‐maximum (FWHM), were used to assess the performance of the proposed method, and confirmed its effectiveness in CB‐XLCT reconstruction.Conclusion:A novel reconstruction method for CB‐XLCT is proposed based on GGMRF model, which enables an adjustable performance tradeoff between L1 and L2 regularization methods. Numerical simulations were conducted to demonstrate its performance.