Date of Award
Doctor of Philosophy
Electrical and Computer Engr
This dissertation proposes a method to restore 3D fluorescence microscopy images obtained from samples with non-uniform refractive index (RI). Existing restoration methods assume the sample is thin and/or has uniform RI in order to keep the computations practical. This is an invalid assumption for most biological samples, which are optically thick because they are neither thin (>5 µm) nor have uniform RI. Specimen properties introduce spherical aberration (SA) in the gathered 3D image, which is dependent on the RI variability between the specimen and the imaging lens’s immersion medium as well as the thickness of the sample. Understanding the image formation process, and modelling it accurately, is essential in the development of a model-based algorithm that solves the inverse imaging problem thereby providing the true fluorescence intensity distribution of the underlying sample. The effect of sample variance on the point-spread function (PSF), i.e. the impulse response of the system, was investigated as part of this project, which led to the development of the N-interface 3D PSF model. Improvement in restoration accuracy in the range of 18%-35% was observed when PSFs from the proposed model were used with a depth-variant restoration algorithm, instead of PSFs from an existing model. Because the PSF predicted by this model is different at every point within the 3D imaging volume, as it should be under these imaging conditions and current imaging models do not address this space variant (SV) microscope response. As part of this dissertation, a block-based forward model was developed to approximate SV imaging. In the block-based model the object space is divided into a collection of non-overlapping 3D blocks where PSFs at block vertices are predicted by the N-interface 3D PSF model. An optimized combination of overlap-save and overlap-add imaging methods was developed to obtain the final SV image. Principal component analysis (PCA) was also used to represent the SV-PSFs, thereby further reducing the dimensionality of our block-based forward model and rendering it practical for use in restoration algorithms. The PCA block-based model produced images that show a 0.98 cross-correlation with images computed without the PCA representation, while achieving an 85% reduction in computational resources compared to those required by the block-based model without PCA. A block-based restoration (BBR) method that combines the SV model investigated here with the fast convergence of conjugate gradient type iteration was adapted to solve the inverse SV imaging problem. Results obtained with the BBR method show a two orders of magnitude improvement in accuracy (quantified using the I-divergence metric) of restoration from simulated SV images using only 20 iterations when compared to restoration of the same data using existing depth-variant or space-invariant approaches. A rigorous numerical phantom of lung tissue was modelled as a part of this study. The BBR was applied to simulated images of the lung phantom and the effect of SV imaging on image restoration was investigated. This dissertation provides a preliminary investigation into the problem of SV in fluorescence microscopy and a possible solution to it, contributing toward the eventual goal of restoring thick biological samples using computational optical sectioning microscopy (COSM).
dissertation or thesis originally submitted to the local University of Memphis Electronic Theses & dissertation (ETD) Repository.
Ghosh, Sreya, "3D Block-Based Restoration (BBR) Method Addressing Space-Variance in Fluorescence Microscopy due to Thick Samples with Non-Uniform Refractive Index" (2015). Electronic Theses and Dissertations. 1248.