Cell Nuclear Morphology Analysis Using 3D Shape Modeling, Machine Learning and Visual Analytics

Quantitative analysis of morphological changes in a cell nucleus is important for the understanding of nuclear architecture and its relationship with cell differentiation, development, proliferation, and disease. Changes in the nuclear form are associated with reorganization of chromatin architecture related to altered functional properties such as gene regulation and expression. Understanding these processes through quantitative analysis of morphological changes is important not only for investigating nuclear organization, but also has clinical implications, for example, in detection and treatment of pathological conditions such as cancer. While efforts have been made to characterize nuclear shapes in two or pseudo-three dimensions, several studies have demonstrated that three dimensional (3D) representations provide better nuclear shape description, in part due to the high variability of nuclear morphologies. 3D shape descriptors that permit robust morphological analysis and facilitate human interpretation are still under active investigation. A few methods have been proposed to classify nuclear morphologies in 3D, however, there is a lack of publicly available 3D data for the evaluation and comparison of such algorithms. There is a compelling need for robust 3D nuclear morphometric techniques to carry out population-wide analyses. In this work, we address a number of these existing limitations. First, we present a largest publicly available, to-date, 3D microscopy imaging dataset for cell nuclear morphology analysis and classification. We provide a detailed description of the image analysis protocol, from segmentation to baseline evaluation of a number of popular classification algorithms using 2D and 3D voxel-based morphometric measures. We proposed a specific cross-validation scheme that accounts for possible batch effects in data. Second, we propose a new technique that combines mathematical modeling, machine learning, and interpretation of morphometric characteristics of cell nuclei and nucleoli in 3D. Employing robust and smooth surface reconstruction methods to accurately approximate 3D object boundary enables the establishment of homologies between different biological shapes. Then, we compute geometric morphological measures characterizing the form of cell nuclei and nucleoli. We combine these methods into a highly parallel computational pipeline workflow for automated morphological analysis of thousands of nuclei and nucleoli in 3D. We also describe the use of visual analytics and deep learning techniques for the analysis of nuclear morphology data. Third, we evaluate proposed methods for 3D surface morphometric analysis of our data. We improved the performance of morphological classification between epithelial vs mesenchymal human prostate cancer cells compared to the previously reported results due to the more accurate shape representation and the use of combined nuclear and nucleolar morphometry. We confirmed previously reported relevant morphological characteristics, and also reported new features that can provide insight in the underlying biological mechanisms of pathology of prostate cancer. We also assessed nuclear morphology changes associated with chromatin remodeling in drug-induced cellular reprogramming. We computed temporal trajectories reflecting morphological differences in astroglial cell sub-populations administered with 2 different treatments vs controls. We described specific changes in nuclear morphology that are characteristic of chromatin re-organization under each treatment, which previously has been only tentatively hypothesized in literature. Our approach demonstrated high classification performance on each of 3 different cell lines and reported the most salient morphometric characteristics. We conclude with the discussion of the potential impact of method development in nuclear morphology analysis on clinical decision-making and fundamental investigation of 3D nuclear architecture. We consider some open problems and future trends in this field.PhDBioinformaticsUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttps://deepblue.lib.umich.edu/bitstream/2027.42/147598/1/akalinin_1.pd