Study Of The Reconstruction Algorithm And Compensation Techniques For Fluorescence Molecular Tomography

Fluorescence molecular tomography?FMT?is a medical imaging technology with low cost,high sensitivity and non-radiation.It has been successfully applied in many fields such as cancer detection,drug discovery,and gene expression visualization.In the FMT,a fluorescent biochemical marker used as contrast agent is first injected into the biological system and consequently accumulates in diseased tissues.Then,a laser in the near-infrared region is used to irradiate the fluorophore which then emits fluorescence light with longer wavelength than that of the excitation light.From the boundary measurements of the emission photons,one can reconstruct the images of the optical parameters such as the fluorescent lifetimes and the fluorescent yield,from which diseases can be diagnosed.Image reconstruction is crucial to the FMT,therefore our thesis focuses on developing efficient and effective reconstruction algorithms for it.The main work in this thesis is as follows:First,in order to alleviate the ill-posedness of the inverse problem in the FMT,An adaptive monotone fast iterative shrinkage thresholding algorithm?AMFISTA?is proposed.To resolve the problem of the over-smoothing,the edge blurring and the low spatial resolution of the traditional ₂l-norm based Tikhonov regularization and considering the sparsity of the fluorescent source,we propose to use the ₁l-norm regularization term in the objective function instead.To obtain a solution to such an optimization problem,an innovative version of the traditional over-relaxation algorithm was proposed by including additional procedures for updating the step size and the regularization parameter adaptively.Simulation results demonstrate that our proposed algorithm can improve the reconstruction accuracy and convergence speed effectively as compared with existed algorithms such as the perturbation algorithm and the over-relaxation algorithm.Second,to consider the error of the diffusion equation in describing the propagation procedure of the photons in tissues,a reconstruction algorithm is proposed in the Bayesian framework where the modelling error of the diffusion approximation can be considered during the reconstruction procedure.As known to all,we usually use the diffusion approximation as a physical model in the FMT reconstruction because this model has superiorities in its simplicity and easiness in computing.However,this model will result in larger error especially for low scattering media and subsequently will result in larger reconstruction error.On the other hand,although the Monte-Carlo method can accurately describe the photon propagating in media,it is computing intensive and hence is not suitable for iterative reconstruction of the FMT.In order to perform accurate reconstruction based on the diffusion approximation,we proposed to reconstruct the FMT image in the Bayesian framework where the unknown optical parameters are regarded as random variables,the modelling errors between the diffusion approximation and the Monte-Carlo simulation are used to construct the noisy term of the projection data along with the measurement noises.In this framework,the unknown parameters are reconstructed upon maximizing the a posteriori probability.In addition,a novel image reconstruction method based on the alternative light source group excitation strategy is proposed to reduce the computation burden of the large scale matrix in the forward problem.The strategy can save operation time effectively.Extensive experiments are performed to validate both the efficiency and effectiveness of the proposed algorithms.In the reconstruction of fluorescence molecular tomographic images,fluorescent targets even with the same optical parameters usually appear with quite different intensities in the reconstructed image if their volume and position are different from each other.To tackle such a problem,a novel reconstruction algorithm with a volume-depth conjoint compensation procedure is proposed.In this algorithm,an improved iterative self-organized data analysis technology?ISODATA?is first used to cluster the reconstructed image to obtain the size and depth of different fluorophores.Then,the reconstructed intensities of different fluorophores are compensated nonlinearly according to their sizes and positions.Extensive simulation results show that this compensation method could amend the reconstruction error resulted from the volume and depth difference of the fluorophore.