A Joint Sparse And Laplacian Manifold Regularization Model For Fluorescence Molecular Tomography

Fluorescence molecular tomography?FMT?was a molecular level optical imaging technique.The emission light of the fluorescent molecular probes was stimulated by near-infrared light in FMT.And then,the emission light of biological surface was utilized to construct the inverse problem of FMT.Researchers obtained the fluorescent molecular probes distribution by solving the inverse problem.FMT had a great potential application value for drug development and early cancer detection.As the reconstruction of FMT was severely ill-posed problem,it was of great significance to study the method of solving the inverse problem of FMT.In this paper,the main work was focused on the reconstruction accuracy and efficiency of the inverse problem.The accuracy and efficiency were promoted by utilizing more prior information and adopting a warm-started strategy,respectively.The main work was as follows:1)In the inverse problem of FMT,to obtain meaningful solutions,a joint L₁-norm and Laplacian manifold regularization model was adopted for utilizing both the sparsity and the spatial aggregation of target fluorescence distribution.The joint L₁-norm and Laplacian manifold regularization model can't be directly solved by the optimization program in compressed sensing theory.In this paper,Sparse Reconstruction by Separable Approximation?Spa RSA?was deduced again to solve the joint model?Spa RSA-resolved Laplacian manifold regularization model,Spa RSALM?.The experimental results showed that compared with Spa RSA algorithm solving L₁-norm regularization model,the Spa RSALM algorithm solving the joint model improved the accuracy of reconstruction results.2)For the time-consuming problem of FMT reconstruction,this paper utilized the warm-started strategy to improve Spa RSALM algorithm efficiency.At first,the coefficient of L₁-norm regularization was set a large value by the warm-started strategy to accelerate Spa RSALM algorithm convergence speed.And then,the warm-started strategy gradually reduced the value of the L₁-norm regularization coefficient in the iterative process.However,in many iterations,the solution of previous regularization coefficient was directly used as the initial regularization coefficient of next iteration.So the solution speed of the Spa RSALM algorithm would be greatly improved.The experimental results showed the speed of Spa RSALM solving the joint model was significantly improved by adopting with the warm-started strategy.