Research On Reconstruction Method Based On Hybrid Optical Propagation Model And Sparse Learning In XLCT

Because the optical information and anatomical structure information can be obtained simultaneously in a single excitation,X-ray luminescence computed tomography（XLCT）becoming a promising molecular imaging technique and having been widely concerned in the preclinical study.Accurate optical propagation model and suitable reconstruction method are two keys for XLCT.Rapid and accurate reconstruction of XLCT is an urgent need for practical application.Based on the theory of optical transmission and biological tissue optics,the hybrid optical transmission model and reconstruction algorithm are studied in this research.To achieve the accuracy and efficiency of XLCT imaging,the hybrid model of third order Simplified Spherical harmonic Approximation（SP₃）and Diffusion Equation（DE）is constructed for solving the mismatch problem of optical transmission model in XLCT,and a more accurate and efficient reconstruction algorithm is used for alleviating the ill-condition problem in the reconstruction process.The research contents and innovations of this paper are as follows:（1）In order to overcome the limitation of single optical transmission model,a hybrid DE-SP₃ model was proposed as the optical transmission model for XLCT.We first simulated the light propagation in various kinds of organs under DE model and SP₃ model respectively.By comparison with the Monte Carlo,these tissues can be categorized into two types,namely DE-fitted tissue that includes muscle and lung,and SP₃-fitted tissue including heart,kidney,liver and stomach.According to the classification results above,we built a hybrid DE-SP₃model to more accurately describe light transport.The experiments illustrated that hybrid DE-SP₃ model achieves higher spatial resolution than DE,and less computational cost than SP₃.（2）Due to the high scattering characteristics of biological organs,the reverse source reconstruction of XLCT is a highly ill-posed problem.In order to realize fast iteration and accurate sparse reconstruction,Lasso algorithm based on Least Square QR（LSQR）is proposed.LSQR method is used to optimize the solution of Lasso algorithm.In the iterative process,the sparse signal is reconstructed by the basis tracing method and L₁ regularization method,which has the advantages of fast iteration and accurate sparse reconstruction.Experimental results show that Lasso-LSQR algorithm can achieve better results in the positioning accuracy,morphological similarity and computational efficiency.（3）The model based traditional reconstruction algorithm needs to manually set the regularization parameters,and the improper selection of parameters will greatly affect the reconstruction effect.In order to avoid the error caused by regularization parameters,a depth framework based on ADMM algorithm（ADMM-Net）is proposed to learn parameters automatically.Different from the existing end-to-end deep learning networks,ADMM-Net extends the iterative process of solving three sub-problems by ADMM method into three sub-layers of the network,and its internal solving process is more visible and interpretable.The experimental results show that the ADMM-Net method has good results in both spatial localization and morphological similarity.