Motion analysis and visualization of biological structures imaged via Nomarski differential interference contrast light microscopy.

摘要

Nomarski differential interference contrast (DIC) light microscopy is an important imaging technique used to study dynamic biological processes, such as cell motility. Advances in the data acquisition capabilities of modern automated microscopes have generated the requirement for similar advances in image processing capabilities to facilitate extraction of relevant information from collected images. For example, in the study of cell motility, specific processing requirements include automated image segmentation, enabling cell shape and motion measurements to be made, and reconstruction of a 3D view of an optically-sectioned specimen from stacks of 2D images. However, the characteristics of DIC images make these tasks difficult to perform. DIC images have a differential appearance which prevents simple image segmentation methods from producing satisfactory results. Additionally, these images are not compatible with existing volume and surface rendering methods, which prevents visualization of 3D data.; This thesis addresses the need for image analysis tools appropriate for use with DIC images. We develop a new active contour model (or snake) algorithm to segment and track multiple, deformable objects through a sequence of 2D images. Our algorithm modifications extend the capabilities of standard active contour models by defining new external forces which enable snakes to detect cells bounded by weak edges, by adapting the snake continuity control parameter values to achieve more flexible snakes, and by developing a new tracking method which handles large motion conditions robustly.; To enable visualization of 3D DIC data, we develop two methods for processing DIC images, generating data that can be rendered with existing software tools. The first method uses a variance filter to emphasize the features of interest in each DIC image. The second method inverts the differential effect of the DIC imaging process, producing data proportional to the local optical path length of a specimen. The data reconstruction is achieved with a directional integration algorithm developed using constrained optimization techniques. While both methods enable observation of a specimen's internal structures, the former algorithm is appropriate for rapid data visualization while the second algorithm is intended for more detailed data analysis.