Font Size: a A A

The Development Of Three-dimensional Super-resolution Microscopy System And Analysis Algorithm

Posted on:2016-03-22Degree:DoctorType:Dissertation
Country:ChinaCandidate:L S GuFull Text:PDF
GTID:1108330467498550Subject:Biophysics
Abstract/Summary:PDF Full Text Request
Optical microscopy especially fluorescence microscopy has become an indispensable tool for biological research, for its non-invasive specificity and multi-color nature. At the same time, optical microscopy is suffered from diffraction and the resolution is limited to about half of wavelength. Recently several super-resolution techniques have been invited to break the diffraction barrier, and push the resolution of fluorescence microscopy to several tens of nanometers. These technologies have greatly expanded the ability of microscopy in research on nano structure of cell, distribution of specific protein and protein-protein interactions.Recent years have witnessed the great development of super-resolution microscopy, but there are still requirements on higher spacial and temporal resolution. In this thesis, we introduced our work on super-resolution microscopy, to reach higher spacial and temporal resolution.In the first part, we introduced super-resolution microscopy that involves single molecule fluorescence interference to reach higher axial localization precision.For microscopy, the3D image of cell structure could provide more information than2D. Generally the lateral resolution of microscopy is better than axial resolution. For super-resolution microscopy, the axial localization technology includes astigmatic approach, bi-plane approach and double-helix PSF method. The axial resolution among these methods is lower than lateral resolution. There is also technology involves interferometry that reaches higher axial resolution, called iPALM and4pi-SMS, but the optical setup is complex and expensive for most labs. Here we describe our implementation of polarized interference photon activation localization microscopy, termed piPALM, which is more simple and cheap than iPALM. We described the optical setup, software as well as the building and adjustment process of our system. The single molecule fluorescence interference could be achieved in our system, and the3-axial resolution is20nm.In the second part, we discussed the theoretical image of restricted dipoles, and its affection for localization error. In order to decrease the impact of the asymmetric PSF, we applied the artificial neural network to predict the parameters of dipoles. Our algorithm has shown more robust and localization accuracy as well as fast computation speed.The resolution is largely depend on single molecule localization precision for single molecule localization based super-resolution microscopy. The non-linear least square (NLLS) and maximum likelihood estimation (MLE) are widely used estimation algorithms, while the2D Gaussian is commonly used model. When the molecule is fixed or restricted, the PSF will become asymmetry. Large localization error will be introduced when using a symmetry Gaussian model to estimate an asymmetry image. In the thesis, we derived the PSF formula of restricted dipole. The numerical calculation of the PSF of restricted dipoles is computation time demanding, which is not suitable for conventional estimation algorithms such as NLLS and MLE. We proposed an estimation method based on artificial neural network, which has proved to be of high localization accuracy, and is not affected by motion conditions of molecule and defocus. What’s more, our method is fast enough for real-time analysis.In the third part, we introduced our work on high molecule density analysis. Combined with compressed sensing (CS) and bi-plane imaging technology, we have achieved3D localization of high molecule density data.The single molecule fitting based method has achieved high resolution but the acquisition time is long, which limit its applications of research on live cell imaging. To decrease the acquisition time demanded, one way is to increase the molecule density to increase the molecule amount per frame, but at the same time increase the change where the points overlap with each other. There are already some algorithm dealing with high molecule density images such as DAO-STORM and compressed sensing, and the compressed sensing algorithm performs better. Here we combine the CS algorithm with bi-plane to extract the3D information of single molecules. The bi-plane approach doesn’t change the shape of PSF, so it doesn’t affect the performance of the reconstruction algorithm. The simulation and experimental analysis showed that our method reached a high molecule density of10μm-1while the resolution is about50nm. The reconstruction takes only300frames, the acquisition time is about6seconds.In this thesis, we proposed our work on increasing both spacial and temporal resolution with both hardware and software design.There’s always compromise between spacial resolution and temporal resolution, so it’s important to choose the balance between them according to the experiments. These three parts form a complete system, and expand the application field of super-resolution microscopy.
Keywords/Search Tags:Fluorescence Microscopy, Super-resolution Microscopy, Single MoleculeLocalization, Single Molecule fluorescence interference, Artificial NeuralNetwork, Compressed sensing
PDF Full Text Request
Related items