Novel Approaches to Time-Lapsed Microscopy Image Analysis and Detection of Biological Agent

Abstract

This Thesis is comprised of four chapters that, respectively: 1) Examine the potential value of studying cellular dynamics and heterogeneity in the context of emerging biological warfare threats; 2) Review both the advantages and disadvantages of current methods that track cells using time-lapse live-cell microscopy; 3) Propose a novel algorithm to measure and track changes in cellular behavior dynamically; and 4) Highlight a novel semi-automated algorithm developed to identify and track cellular focal adhesions overtime in 3D. The ability of non-scientists to create and deploy a biological weapon (BW) highlights the emergence of a new threat, the biohacker. The advent of modern technologies enables the biohacker to employ one or a multitude of strategies to increase the tactical or strategic effectiveness of a biological agent. Preventing a successful large scale BW attack relies on the ability to rapidly detect the presence of and rapidly identify a biological weapon agent. The impact of BW agents on human health directly percolates from organ failure and tissue destruction, but is ultimately defined by the toxic effects on cellular functions, with the most severe being cell death. Understanding cellular dynamics and heterogeneity are therefore critical to immediately understanding BW effects. Time-lapsed live-cell microscopy is an ideal technique to study cellular changes. However, tracking methods are limited. Initial integer programming algorithms were successful in tracking cells but are limited to low cell number and density. Others handle large cell numbers but are limited in their capability to detect mitotic events. There exist three distinct and novel components to our approach: 1) user corrected segmentation, 2) a k-nearest neighbor algorithm to generate “high confidence tracks”, and 3) track assumptions automatically scrutinized within the program. Initial results indicates that the algorithm is accurate, repeatable, and applicable to a variety of datasets. The algorithm provides a surfeit of quantifiable data on cell morphological features. Though the results of the semi-automated focal adhesion (FA) algorithm are at an early stage, the potential for a fully automated algorithm is evident. The initial data indicate correct identification of FAs is being achieved.