Research On Reconstruction Of Fluorescence Molecular Tomography Based On Matching Pursuit Method

Optical molecular imaging is a new type of molecular imaging technique.It has the advantages of non-invasiveness,non-radiation,high sensitivity,fast imaging speed,low cost,etc.In recent years,there have been great developments in theoretical and practical applications.However,two-dimensional(2D)surface fluorescence derived from the inner light sources of organism after scattering and absorption,only provide the 2D qualitative information,which does not reflect its 3D cluster morphology and distant effect.Therefore,many research groups have devoted to developing fluorescence tomography systems and algorithms in order to reconstruct the 3D distribution of fluorescent probes.Unlike traditional two-dimensional optical imaging methods,Fluorescent molecular tomography(FMT)could achieve non-invasive molecular activity monitoring in three-dimensional space.This low-cost real-time imaging technology has quickly become one of the hot research areas.However,due to the errors of the reconstruction models and the reconstruction methods,the robustness and accuracy of the reconstruction results are still insufficient in practical problems,and further exploration and research should be conducted in these two aspects.In this thesis,research on improving the robustness and accuracy of the FMT reconstruction algorithm with the sparse prior information of the fluorescent light source in the organization based on the matching tracking algorithm using the greedy strategy.A new Sparsity Adaptive Correntropy Matching Pursuit Method(SACMP)reconstruction algorithm was proposed to improve the robustness and accuracy of the reconstruction algorithm,and detailed verifications are performed using numerical simulation experiments and in vivo experiments.The main contributions of this thesis are listed as follows:1.According to the finite element method,we solve the forward problem and establish the inverse problem of the FMT reconstruction.Considering the transmission characteristics of light in vivo,the light transmission model is established based on the radiative transfer equation.Because of the high-scattering and low-absorption properties of near-infrared light in biological tissues,the first-order spherical harmonic approximation is used to simplify the radiative transfer equation into a diffusion equation.Then the diffusion equation is transformed into matrix-form equations using the finite element method,considering the boundary conditions.It is the target equations to be solved later.2.The application of orthogonal matching pursuit(OMP)algorithm based on greedy strategy in FMT reconstruction problem.The inverse problem is ill-conditioned and it is very difficult to obtain stable and accurate approximate solutions.One of the major solutions is proposed based on the assumption that fluorescent source is sparse,because the fluorescent probe distribution is relatively concentrated in some specific areas compared to the whole organ.Compared to the explicit methods of adding regularization terms,the greedy strategy is used to achieve the sparseness of the solution.Then we introduced the applications of some improved algorithms based on orthogonal matching pursuit in the field of FMT reconstruction.In order to alleviate the limitations of these methods,we proposed a new robust FMT reconstruction algorithm based on SACMP.3.Research on FMT reconstruction algorithm based on SACMP.In the iterative process,the traditional OMP algorithms usually use the mean square error criterion to estimate the error variation of each iteration which requires that the error conforms to the Gaussian distribution.Otherwise,the performance of the algorithm will reduce greatly.To solve this problem,we propose a FMT reconstruction method based on the SACMP which does not make any assumptions about the distribution of errors and can deal with Gaussian and non-Gaussian errors.4.The numerical simulation experiments and in vivo experiments are designed to evaluate the performance of proposed method.In order to assess the robustness and accuracy of the algorithm,we designed dual-source numerical simulation experiments.By adding different types and levels of noise to the simulation model,the robustness to noise of different algorithms were analyzed quantitatively.Simulation studies demonstrated that the reconstruction results of SACMP method had better performance than StOMP and SASP methods on dealing with data contaminated by non-Gaussian noise.To further evaluate the practical performance of the SACMP algorithm,we carried out in vivo mesenchymal stem cells(MSCs)distribution experiments in brain.The reconstructed result of SACMP had the biggest Dice index which means the reconstructed stem cell distribution shape match better with the actual morphology.