The Research Of X-ray Luminescence Computed Tomography Methods Based On Adaptive Sparse Regularization

X-ray luminescence computed tomography(XLCT),as a novel hybrid imaging technology,uses X-ray to deeply excite rare nanoparticles and emit near-infrared light,which can simultaneously collect biological body spatial structure information and surface fluorescence information,providing the possibility for deep tissue imaging.Therefore,XLCT has the advantages of high penetration and high spatial resolution,and has clinical application value in the fields of early tumor detection and drug metabolism tracking.However,due to the scattering and absorption of light in biological tissues,the reconstruction of XLCT is an ill-posed problem.Thus,efficient and robust reconstruction algorithms have always been the research focus of X-ray tomography.Based on the research of sparse regular light source reconstruction,this paper mainly studies the XLCT light source reconstruction method based on adaptive sparse regularization,aiming at the balance of sparsity and smoothness of the solution,and the selection of regularization parameters.The specific research content of this paper is as follows:(1)For different biological applications,the existing sparse regularization methods have been extensively studied in terms of solution sparsity and morphological restoration.However,the higher the sparsity of the solution,the greater the morphological loss.Therefore,it is necessary to design a general algorithm that takes both sparsity and smoothness into consideration.In this research,an Elastic net-l1l2 reconstruction method,which consists of a convex combination of l1and l2 regularization,is proposed aiming to enhance the sparsity and suppress the smoothness of specific solutions.Firstly,the reconstruction can be solved by soft threshold iteration algorithm.Secondly,multi-parameter K-fold cross validation strategy is adopted to find the optimal parameters adaptively within acertain parameter range under supervised learning.From the numerical simulation,phantom and in vivo experiments,it's domonstrated that the average location error of our proposed method is 0.47 ± 0.18 mm and the contrast-to-noise ratio is higher than other comparison algorithms.(2)In view of the limited prior knowledge of the XLCT light source,the reconstruction results of the traditional sparse Bayes algorithm deviate from the real position.By introducing the idea of approximate message passing,a bayesian learning of generalized approximate message passing(GAMP-SBL).is proposed In order to speed up the convergence of the algorithm,the Bayesian learning model and the minimum mean square error solution model are interactively iterated in the inverse problem solving,and the alternate optimization scheme is used to automatically calculate all the unknown hyperparameters in the message passing process.In order to verify the feasibility of the proposed method,the study was compared with the sparse Bayesian method and the generalized approximate message passing method.The research results show that the proposed GAMP-SBL method in this paper can improve the quality of XLCT reconstruction,and the average reconstruction time is increased by 2.5 times compared with the traditional Bayesian model.