Fluorescence Molecular Tomography (FMT) Reconstruction By Integration Of Prior Knowledge Of Structure And Sparsity

As a branch of optical molecular imaging technology,fluorescence molecular tomography(FMT)can non-invasively visualize 3D imaging targets in vivo by fluorescent probes,which can monitor the spatiotemporal distribution of cellular and subcellular molecular.Due to low cost,no radiation,fast imaging,it is a promising imaging technology in disease diagnosis and treatment,drug development,surgical navigation and other fields.However,due to the high scattering and nonlinear transmission of light in biological tissues,the reconstruction process of FMT is an ill-posedness problem.Based on the structure prior information and sparse prior information of fluorescence distribution,the reconstruction method is studied in this thesis.The main contents of this thesis are as follows:1.The joint model for fluorescence molecular tomography based on ₁L sparsity regularization and Laplacian Manifold RegularizationThe probe of FMT imaging is mainly fluorescent protein or fluorescent probe.For specific probes,the fluorescence distribution is locally correlated.According to the prior knowledge of the spatial distribution of fluorescent probes,the Laplacian manifold regularization is introduced to describe the continuous smoothness of specific probes in biological tissues better;according to the sparsity distribution of fluorescent probes,the sparse regularization is introduced to constrain the fluorescence distribution;a joint model is established based on Laplacian manifold regularization and sparsity regularization and gradient projection algorithm is used to solve the model.Through the digital mouse simulation experiment,it is demonstrated that the reconstruction method improves the location accuracy,energy aggregation and resolution of fluorescent distribution,and robust for noise interference and excitation light source interference.In this thesis,the performance of the reconstruction method is further evaluated in vivo experiments in nude mice implanted with fluorescent tube.2.The joint model of fluorescence molecular tomography based on L_1/2 sparse regularization and hybrid truncated Laplacian manifold regularizationOn the basis of work 1,the location accuracy and morphology of fluorescent probes'distribution are further improved.On the one hand,the hard truncation method is used to modify the Laplacian kernel function which represents the local correlation,and the anatomical structure information is weighted the solution space of fluorescent probes.The hybrid truncated Laplacian truncated manifold regularization is constructed through the modification.On the other hand,we use L_1/2 sparse regularization to enhance the sparsity.At the same time,in order to solve the nonconvex model simply and quickly,half threshold iterative algorithm is introduced to solve the model.The simulation results show that the reconstruction method proposed in this thesis can achieve higher reconstruction accuracy in FMT reconstruction,realize morphological reconstruction,and have certain morphological resolution.In this thesis,the performance of this method in FMT reconstruction is demonstrated by the in vivo experiment of glioma in nude mice.3.A sparse reconstruction method based on half thresholding matching algorithmThis work is for L_1/2-norm problem based on work 2.When solving this nonconvex problem,it is difficult to balance the implementation difficulty and the usage difficulty of the algorithm.Thus,a half thresholding iterative algorithm is introduced to deal with non-convex algorithm.At the same time,in order to speed up the reconstruction process,we introduce the idea of matching pursuit,which greatly reduces the number of iterations for the convergence,so as to obtain the reconstruction results more quickly and accurately.In this thesis,the simulation results show that the proposed half threshold pursuit algorithm can converge fast,be more robust to parameters and have higher reconstruction fluorescence yield than the half threshold algorithm.The performance of the half threshold pursuit algorithm is demonstrated by in vivo experiments of nude mice implanted with fluorescent tube.To be concluded,this thesis mainly studies how to introduce the structure prior and anatomical structure prior of fluorescent probes into reconstruction methods.The final purpose is to design a reconstruction method with higher accuracy,morphological distribution,spatial resolution,more robust to parameters and faster reconstruction speed.The joint model based on ₁L sparsity regularization and Laplacian Manifold Regularization is proposed to achieve high precision,high signal-to-noise ratio and high resolution.The joint model based on L_1/2 sparse regularization and hybrid truncated Laplacian manifold regularization is used to realize morphological reconstruction.The half thresholding matching algorithm is designed to solve inverse problem with high speed.