Bioluminescence Tomography Based On Non-convex Sparse Reconstruction

Bioluminescence tomography is a novel,non-invasive optical molecular imaging modality.It relys on the reaction between luciferase enzymes and luciferin substrate.By collecting the transmissible photons on the biological boby surface,the distribution information of internal light source can be reconstructed with sophisticated algorithms.BLT enable researchers to detect and observe the physiological or pathological changes at the molecular level,which make it a powerful tool in preclinical applications,including early diagnosis of tumor,gene therapy and drug evaluation,etc.BLT is a severe ill-posed inverse problem and its application relys on the development of reconstruction algorithm.In this work,based on the diffusion approximation model of radiative transfer equation,using the linear relationship between surface measurement data and light source distribution information derived by finite element method,we explore a series of sparse reconstruction methods based on non-convex regularization,which utilizes the sparsity of the light source.(1)An L1/2 norm regularization algorithm is introduced into BLT reconstruction.The non-convex L1/2 norm regularization problem is converted into a series of L1 norm minimization problem,and efficiently sovled by homotopy algorithm.The single-source simulation experiment based on digital mouse model show that the proposed algorithm has better performance in terms of location error than the other three comparison algorithms.The results of double-source simulation also show its superiority in ability of resolving multiple sources.The reoncstruciton from simulated measurements with noises and the reconstruction of light sources placed in different organs further validate the stability of the proposed algorithm.(2)A new non-convex reconstruction model with the smoothly clipped absolute deviation(SCAD)penalty function is proposed and the regularization functions and using the general iterative shrinkage and thresholding algorithm(GIST)is applied to solve this model.The iterative shrinkage feasible region and multi-spectral measurement strategy are combined to reduce the ill-posed of BLT reconstruction.The single-source and double-source simulation results show that the proposed reconstruction algorithm performs better than the comparison algorithms under different source depth.It has better location accuracy and multiple source resolving ability than the competitors.The experimental results of the phantom data collected on the CBCT/BLT imaging system further validate the advantages of the GIST algorithm in terms of light source positioning accuracy and stability.(3)A hybrid optimization algorithm for non-convex regularized problems(HONOR)is proposed for multispectral BLT.The single-source simulation results show that the proposed HONOR algorithm has better location accuracy compared to the GIST algorithm.The double-source simulation results show that the both the ability of multiple source resolving and the reconstruction image quality are better than the comparison algorithm.