Research On Image Restoration Of Fluorescence Microscopic Slice In Deep Tissue

Combined with fluorescence labeling technique, fluorescence microscopy is becoming a powerful tool to investigate activities and morphologies of specimens in life science. However, image stacks are always inevitably blurred due to the diffraction limit and distorted by noises. The degradation increases with depth when imaging in deep tissue due to strong and multi-scattering of specimens. As a post-processing approach, image restoration is a potential technique to reverse the degradation and restore the original fluorescence objects. Therefore, the algorithms of denoising and deconvolution for deep tissue are well investigated in this dissertation. In addition, attempts to realize the random access controlling of laser and obtain variable integral time of fluorescence signal through the design of random scanning two-photon fluorescence microscope are explored. The main subjects are as follows:The imaging model of fluorescence microscope is studied. First, different sources of noise are analyzed and the photon counting noise, which is naturally induced by Poisson counting procedure of detector and dependent on fluorescence signal, is considered as the main one and can not be avoided. So Poisson imaging model is more suitable for image restoration than Gaussian model because it coincide with the natural properties of photon detection. Second, point spread function and optical resolutions of three typical fluorescence microscopes are discussed. Last, almost all of the deconvolution techniques for fluorescence image are comparably introduced, and some typical deconvolution algorithms are studied either theoretically or experimentally. The conclusion can be drawn that noise amplifying will occur in restored results more or less, and some tedious ringing artifacts are reproduced too.More literatures are shown that it is a wise approach to pre-filter noise before performing the deconvolution. So the edge-persevering denoising algorithms based on the Perona-Malik (P-M) nonlinear anisotropic diffusion equation are studied thoroughly. First of all, the principles of edge-persevering and the drawback of P-M diffusion equation are discussed, and then a modified scheme that the local variation of pixel is incorporated in diffusion coefficient is proposed. The modified algorithm performs well in random noise removing and edge preserving. Furthermore, in order to enhance the performance of edge-preserving, another adaptive and robust P-M diffusion algorithm is proposed, which combines the robust estimator with two local statistic parameters of pixel and obtains better edge preserving. Experimental results of synthetic, standard image and real fluorescence image slices in deep tissue show the better performance of proposed scheme in edge-persevering and noise removing, and the fast convergence and stable solution is obtained, too.Regularization technique is more vital to moderate the degree of ill-posed of deconvolution and it is well studied based on Richardson-Lucy deconvolution in this dissertation. A novel scheme that pre-filters noise employed modified robust anisotropic P-M diffusion then performs maximum entropy regularized Richardson-Lucy deconvolution is proposed, which is based on maximum entropy regularization technique. Meanwhile, another hybrid regularization scheme is also proposed too, which integrate Tikhonov regularization with Total Variation regularization together on the consideration of their similar model and their principles of regularization and drawbacks. Experiments show the limited effects on restored objects.Last, it is an essential alternative to improve the signal-to-noise of fluorescence image by exactly controlling the laser scanning and prolonging the dwell-time of regions of interesting of specimens. However, the precise controlling of acousto-optic deflector is a significant puzzle. Therefore, a high-speed timing synchronization approach between the clock relationships of two sub-systems in only one multi-functions board is proposed in the design of custom-built random access scanning two-photon fluorescence microscopy. It makes possible that laser scanning can be controlled accurately and dwell-time of fluorescence signal integral can be set flexibly.