Research On Fast Reconstruction Of Fluorescence Molecular Tomography For Multi Projection Measurements

Fluorescence molecular tomography(FMT),as an important optical molecular imaging,has been widely applied to cancer detection,disease treatment and drug discovery.However,the reconstruction problem is suffered from ill-conditioned due to the strong scattering of light propagation in biological tissues and the limited surface measurements.In order to reduce the ill-conditioned,multiple projection angles of fluorescence measurements are usually collected to further improve reconstruction results.However,such a large scale of data of reconstruction needs a large amount of computational memory and time.In this paper,we focus on the reconstruction accuracy and efficiency,research on FMT reconstruction algorithm from the matrix dimension reduction.The specific research work is as follows:Such a large scale of multi projection is very memory-costing and time-consuming.To address this problem,an accelerated reconstruction method based on dimension reduction is proposed.Larger dimensional data set can be reduced to a lower one by using principal component analyses which satisfies the orthogonal projection transformation.The experimental results demonstrate that the computing time can be reduced about 10 times after using lower dimensional data set for reconstruction.Based on manifold learning and compressive sensing theory,a fast reconstruction algorithm which combines with local preserving projection(LPP)and sparse regularization was proposed.This method utilized LPP to make a partial linear feature extraction on the surface fluorescence measurements,which nearly contains the overall characteristics of the measurements.Meanwhile,this strategy retains the nonlinear structure in the original data.The results show that the proposed method can not only speed up the reconstruction,but also it can improve the accuracy and the clustering of the target.In addition,an algorithm based on linear regression approximation scheme with dual augmented lagrangian has been presented in this work.First,sub-space set is obtained from original multi projection measurements using linear regression approximation method,which can effectively solve the large-scale matrix feature extraction problems.Then,the reconstruction problem is resolved by augmented lagrangian method.Compared with the traditional method,this method is more suitable for the case where the matrix condition is poor and the matrix is larger.The robustness and stability of the method were verified by numerical and in vivo experiments.And the reconstruction speed is further accelerated.