Advanced Image Deconvolution Techniques for Super-resolution Microscopy

Abstract

In last decades, super-resolution fluorescence microscopy has seen its great development. Many super-resolution techniques are developed: the non-linear super-resolution techniques such as STED, PALM/STROM have provided the power of imaging scale under 50nm; at the meantime, many linear super-resolution techniques, such as SIM and ISM are also developed for fast and low photon bleaching super-resolution imaging. In the work of this study, I focused on studying the super-resolution techniques, such as SOFI, SIM, ISM and Airy beams light-sheet microscopy, and deconvolution algorithms for these super-resolution microscopies. \n \nSpinning disk confocal microscopy is a widely-used fast confocal microscopy. It has been demonstrated that super-resolution can be obtained by applying ISM to spinning disk confocal microscopy. It is interesting to develop precise control system and easy-to-used image acquisition and reconstruction software for wide spread application of spinning disk confocal-ISM. For this, we develop the Spinning disk Confocal-ISM software package provides flexible functions and a friendly GUI for fast 3D confocal imaging, by which image acquisition can be done with a single click, as well as image reconstruction. The value of our work is that it will help any user to upgrade the SDC system quickly to SDC-ISM to obtain high-quality images. \n \nL1-norm regularization is one important technique for sparse signal representation. The involved reconstruction algorithm is the key for applications in compressed sensing and very important regularization for image storing. For the sake of easy-to-use requirement in application, I converted the very challenging L1-norm-regularized optimization problem to a normal non-linear optimization problem by approximating the L1-norm with a flexible smooth function. Then, the problem can be solved by existing and very powerful non-linear optimization methods, such as the LBFGS algorithm or the non-linear conjugate gradient methods. The simulation results show that the proposed method work quite well and outperform existing methods in computing time, with very close accuracy. \n \nDeconvolution algorithms based on efficient non-linear optimization methods with regularizations, such as Total Variation regularization, Hessian regularization, roughness regularization, were developed for SOFI, SIM, and ISM image enhancement and artifact removal. Furthermore, an accelerated regularized 3D Richardson-Lucy algorithm was developed for Airy light-sheet microscopy image reconstruction. The regularized 3D RL algorithm is very promising for 3D image data processing acquired with modern high-speed and high-resolution imaging systems, which also can be used to large aperture objective Airy light-sheet microscopy, to which the 1D deconvolution does not work well any more. The proposed methods are validated by simulations and experimental image deconvolution, which show promising results for application.