Research On Reconstruction Algorithm Of Bioluminescence Tomography Based On Adaptive Sparse Representation

Bioluminescence tomography（BLT）,as a non-invasive medical imaging technique,can dynamically provide information on tumor distribution in living animals.However,BLT light source reconstruction is an ill-posed problem due to the high scattering properties of biological tissues,as well as the limited information and noise interference of optical measurements on biological surfaces.At present,the sparse representation theory represented by ₁L norm convex optimization and greedy algorithm provides a solution to ill-conditioned problems,and has been widely studied in BLT reconstruction,which can achieve effective reconstruction of light sources.Nevertheless,convex optimization algorithms based on ₁L norm usually face the problems of low time efficiency and difficult to select regularization parameters;the reconstruction algorithm based on the greedy strategy has a large influence on the reconstruction structure of the sparsity parameter,and it is easy to drop into the local optimum.Therefore,based on the sparse representation theory and combining the advantages of two types of algorithms,this thesis proposes adaptive sparse representation bioluminescence tomography reconstruction algorithms,which provide a new idea for designing an efficient,robust and accurate BLT reconstruction algorithm.The detailed research content are as follows:（1）In the research,to recover the 3D tumor distribution quickly and accurately,an adaptive Newton hard thresholding pursuit（ANHTP）algorithm was proposed to improve the performance of BLT.ANHTP adopts an adaptive sparsity adjustment strategy to obtain the support set,which simplifies the adjustment process of sparsity parameters.At the same time,based on the strong Wolfe line search criterion,an improved damped Newton algorithm is constructed to obtain the optimal light source distribution information.ANHTP algorithm fully combines the advantages of sparse constraint optimization and convex optimization,ensures global convergence,greatly shortens reconstruction time,and improves the low time efficiency of convex optimization algorithm and the local optimality of greedy algorithm.A series of numerical simulations,physical phantoms,and in vivo experiments show that ANHTP has the characteristics of accurate localization,short reconstruction time,and strong robustness,which can further improve the practicability of BLT in biomedical applications.（2）To solve the problem that the sparsity of the matching pursuit algorithm is difficult to estimate,an improved adaptive orthogonal matching pursuit（IAOMP）reconstruction algorithm is proposed based on the traditional sparsity adaptive matching pursuit algorithm and combined with the tetrahedral structure information.IAOMP selects multiple atoms in combination with spatial tetrahedral structure information in each iteration,which can speed up atom selection,reduce the number of iterations,and avoid excessive sparseness;then use nonlinear functions to adaptively estimate the sparsity,and prune to obtain the support set.The sparsity estimated by this function can quickly and accurately approximate the true sparsity,and it also reduces the tediousness of manual estimation.Experimental results confirm that IAOMP has better robustness and localization accuracy.