Detection, segmentation, and tracking of cells in microscopy images

Update Item Information
Title Detection, segmentation, and tracking of cells in microscopy images
Publication Type dissertation
School or College College of Engineering
Department Electrical & Computer Engineering
Author Ramesh, Nisha
Date 2018
Description Robust cell detection and segmentation serves as a critical prerequisite to test hypotheses on cell morphology, development, and behavior. Tracking cells in time and studying their evolution is necessary to analyze cell growth, mobility, and lineages. In this dissertation, we propose methods for cell detection, segmentation, and tracking to address some of the limitations in the current literature. First, we develop an implicit parametric convex shape model to segment and track cells across time. The convex polytope is represented as the intersection of half-spaces in two dimensions (2D) or three dimensions (3D). Being an implicit representation allows the model to naturally change topology during its evolution if needed. We propagate the shape model of the cell embedded in a particle filter to track cells with varying shapes. We extend the convex shape model to a generic shape representation, namely, the disjunctive normal shape model (DNSM), a differentiable implicit and parametric model to segment and analyze regions of interest with arbitrary shapes in medical images. In DNSM, we approximate the characteristic function of a shape as a union of convex polytopes, which themselves are represented as intersections of half-spaces in 2D or 3D. Second, we present a hierarchical approach to detect cells based on nonoverlapping extremal region selection in a semisupervised framework using minimal user annotation. The set of extremal regions is generated using maximally stable extremal region (MSER) detector. Using the tree built from the MSER detector, we develop a differentiable unsupervised loss term that enforces the nonoverlap constraint with the learned function. The supervised and unsupervised losses are embedded in a Bayesian framework for probabilistic learning. The idea of semisupervised learning is also extended to predict cell division and motion in microscopy images. The supervised loss minimizes the error in predictions for the division and move classifiers. The unsupervised loss constrains the incoming links for every detection such that only one of the links is active. Similarly for the outgoing links, we enforce at most two links to be active.
Type Text
Publisher University of Utah
Dissertation Name Doctor of Philosophy
Language eng
Rights Management (c) Nisha Ramesh
Format application/pdf
Format Medium application/pdf
ARK ark:/87278/s6hky1x5
Setname ir_etd
ID 2528981
Reference URL https://collections.lib.utah.edu/ark:/87278/s6hky1x5
Back to Search Results