Event Location: NSH 3002

Abstract: Computer vision analysis of live cells in time-lapse microscopy images enables long-term continuous monitoring of a large number of intact cells with minimal human intervention, which has not been feasible by existing image processing methods with cellular staining images. Of critical importance in time-lapse microscopy image analysis is to understand the effect of biochemical factors on cell proliferation and differentiation, which can be measured by the detection of cellular events, particularly, mitosis (cell birth), apoptosis (cell death), and differentiation. This thesis proposes a framework to address the detection of these cellular events in time-lapse live cell microscopy images. The framework consists of three stages that (1) detects candidates of cellular events at each frame, (2) incorporates temporal information from neighboring frames, and (3) adopts temporal probabilistic models that not only capture the visual change around an event, but also utilize additional useful information, such as the timing of an event. In our preliminary work, experiments on several types of stem cell populations under different culture conditions show that the proposed methods reliably detect cellular events and the detection results can significantly improve cell tracking systems, which is a comprehensive tool for the analysis of cell behavior.

Committee:Takeo Kanade, Co-chair

Stephen E. Fienberg, Co-chair

Robert F. Murphy

Fernando De la Torre

Alan J. Russell, University of Pittsburgh