Research On Image Fusion And Reconstruction Optimization Algorithm In Fluorescence Molecular Tomography

Fluorescence molecular tomography(FMT)can visualize the temporal and spatial distribution of specific fluorescent molecular probes in vivo.Taking advantages of high imaging sensitivity,no risk of ionizing radiation,and low cost,it has great potential in the field of visual detection in drug research and development screening,early diagnosis and treatment of diseases,surgical navigation.By using a high-sensitivity camera to collect nearinfrared optical signals on the surface of the organisms,and combining the physical model of light propagation in biological tissues,then FMT can use the reconstruction algorithm to obtain the three-dimensional distribution of fluorescent light sources.In recent years,FMT reconstruction algorithms and their applications have developed rapidly.However,due to the complexity of light transmission in biological tissues and the optical characteristics of tissues,the reconstruction accuracy,reconstruction speed,and stability of FMT still need to be improved.The dissertation will focus on the related issues of FMT image fusion and optimization of reconstruction algorithm in the following three main aspects:1.Multimodal fusion is a prerequisite for accurate FMT reconstruction,and threedimensional registration is a necessary step to achieve the fusion of different modal information in the FMT system.In order to improve the accuracy and speed of registration,this dissertation proposes an improved registration algorithm based on the bidirectional distance ratio.The algorithm uses principal component analysis(PCA)to achieve coarse registration of the point cloud model.In the fine registration stage,it firstly searches for point cloud matching point pairs from two directions,secondly calculates the bidirectional distance ratio between matching point pairs and converts it into a probability,and finally uses this probability as the weight of the algorithm for calculating registration parameters.The addition of weights limits the interference of mismatched point pairs,thereby improving the accuracy and speed of registration.Experiments show that when the point cloud has large position deviation,obvious shape difference and local incompleteness,then the registration accuracy of the algorithm in the dissertation is significantly improved compared to the classic ICP algorithm and Tr ICP algorithm,especially in the registration speed.At the same time,the registration algorithm is applied to the fluorescence mapping.Considering that white light and fluorescent image collection belong to the same optical system,the dissertation designs a method of using white light image as a mapping link to fuse fluorescent information and structural information.Firstly,a two-dimensional white light image is used to reconstruct a three-dimensional surface.Secondly,the registration algorithm proposed in the dissertation is used to realize the registration of white light 3D surface and CT 3D surface.Then,concerning that the white light and the fluorescence image have the same coordinate system,the quantized and corrected fluorescence information are mapped to the CT surface.In vivo experiment show the feasibility of the method to the fluorescence mapping and the effectiveness of the improved registration algorithm in this dissertation.2.Due to the coupling between the light source components to be reconstructed in FMT,it is difficult to solve the reconstruction problem directly using the l1-norm sparse regularization algorithm.If the primal problem of reconstruction is changed to its dual problem,the dual augmented Lagrangian algorithm(DALM)can reduce the difficulty of solution.However,when the full-angle projection data are used to solve the DALM algorithm,the convergence speed is slow and cannot match with rapid reconstruction.Therefore,I has proposed a DALM reconstruction algorithm combined with a linear regression approximation(LRA-DALM).Firstly,a linear regression approximation algorithm is used to reduce the original system matrix to an approximate sub-matrix,and then a DALM algorithm based on l1-norm sparse regularization is used to complete FMT reconstruction at the approximate sub-matrix data scale.The numerical simulation experiments of single and dual light sources show that the speed of light source reconstruction is significantly improved with better the reconstruction accuracy and stability comparing with Tikhonov regularization algorithm and DALM algorithm.When the number of excitation light sources increases from 3–36,the reconstruction speed of the LRA-DALM algorithm can maintain the fast level stably compared to the comparison algorithm.In vivo single light source reconstruction experiment verified the feasibility of the LRA-DALM algorithm applied to FMT reconstruction.3.To resolve the problems of poor uniform grid reconstruction performance and difficulty in setting adaptive grid parameters,a global reconstruction strategy based on nonuniform finite element grid is proposed herein.When the finite element method is used to numerically solve the optical transmission model,it is necessary to discretize the imaging region.When using a global uniform grid,a very sparse grid is difficult to accurately approximate the fluorescent target,and a very dense grid will affect the reconstruction speed.Although the adaptive grid can be continuously subdivided and reconstructed according to the feasible region,it is difficult to accurately determine the parameters of the feasible region.And when the feasible region deviates from the actual light source region,the unsubdivided region will fail to participate in the reconstruction,and subsequently will affect the final reconstruction accuracy.Therefore,the proposed algorithm first coarsely meshes the imaging region to complete the initial reconstruction,and then divides the tetrahedral node region with higher fluorescence yield again in the initial reconstruction.Unlike adaptive grids,this algorithm not only uses subdivision grids,but also can combine coarse subdivision grids and subdivision grids into a non-uniform grid covering the entire imaging region.Finally,the non-uniform grid is used to realize the global reconstruction of FMT.Considering the sparse characteristics of the FMT light source target,the IVTCG algorithm is applied in the reconstruction process to solve the sparse regularization problem.Numerical simulation and physical phantom experiments show that this method has good effects on reconstruction positioning accuracy,reconstruction speed and stability.