Research On Target And Background Separation Method In Dynamic Fluorescence Imaging

By adding the time dimension to fluorescent molecular imaging,dynamic fluorescence imaging can be used to observe the physiological processes of the organism dynamically and continuously.It involves the collection of circulation,distribution and accumulation of fluorescence probes in small animals.However,the target fluorescent signal is always contaminated by background florescence.The Background fluorescence is extensively present in the visible light range which is 400nm⁷00nm.Considering the difference of the target fluorescence signal and the background fluorescence signal with time,this paper intends to separate the target and background by using blind source separation?BSS?method,so as to improve the dynamic fluorescence imaging and provide high quality image series for the subsequent quantitative analysis.In this paper,the application of nonnegative matrix factorization?NMF?algorithm,Non-smooth nonnegative matrix factorization?nsNMF?algorithm and convex analysis of mixtures of non-negative sources?CAMNS?algorithm in dynamic fluorescence imaging are studied.Considering that the target distribution is somewhat sparse in the background fluorescence,a nonnegative matrix decomposition algorithm based on L1/2 norm constraint(L_1/2-NMF)is introduced in this paper.And the number of endmember elements,matrix initialization and iteration numbers are studied.First,the endmember number is estimated by the principal component analysis?PCA?algorithm which is simple and easy to implement.Second,the matrix is initialized by the vertex composition analysis?VCA?algorithm to improve the robustness of the algorithm.Finally,the number of iterations is evaluated for dynamic fluorescence imaging.Simulation is used to analyze the above methods,and the performances of the algorithms are compared in running time,the correlation coefficient of the unmixed dynamic curve and the signal to background ratio.In order to further testify the performance of the proposed methods,several in vivo experiments were carried out,including:multispectral experiment with subcutaneous injection of indocyanine green and cyanine cyanine Cy5.5 multi-fluorophore,dynamic experiment with subcutaneous injection of cyanine dye Cy5.5 and tail vein injection of cyanine dye Cy7.5.The experiment results show that the target fluorescence and background fluorescence based on blind source separation can be separated to a certain extent,and the separation result of L1/2-NMF algorithm has higher signal to background ratio.