Computer Vision, Computer Graphics, Modeling and Analysis

My research interests include image segmentation, visualization and analysis. Image segmentation has primarily focussed on the use of deformable models to find the plasma membrane of flourescently labeled smooth muscle cells. It has been used it to greatly reduce the labor required to analyze the relationship of Protein Kinase C to the cell membrane proteins.

Visualization research has focussed on methods to productively view and analyze 3D and 4D (3D+time) images of cells. This has resulted in development of the Data Visualization and Analysis Environment (DAVE) which can interactively visualize, in stereo, 3D single, double, or triple labeled images, simultaneously display volume data (fluorescently labeled plasma membrane pumps or channels) and surface data (e.g., the membrane itself ), automatically locate, size and sort discretely distributed sites, and automatically locate, count and highlight the colocalization sites of any two labels. DAVE has the ability to step through a time-series of 3D images and visually compare each with chosen "reference" time points in order to better understand temporal changes.

I have developed methods for measuring the degree of colocalization of two fluorophores in 3D. This includes designing Monte Carlo methods for estimating the statistical significance of that colocalization.

Another research interest focuses on modeling reaction-diffusion equations. Explicit time finite difference methods have been used to discretize the partial differential equations needed to model calcium influx into the cell and its interaction with fluorescent indicators and buffers within the cell. This software has been used to model caffeine activated plasma membrane channels as well as the relationship between STOCS (spontaneous transient outward currents) and sparks. The simulation has been parallelized to run on a Beowulf system (a network of Linux PCs with Alpha CPUs, connected via Myrinet). Variable sized voxels can be specified, thus enabling a locally high spatial resolution (e.g., 25 nm near a channel opening) without requiring similar resolution (and its concomitant increased simulation time) elsewhere (e.g. 100 nm voxels elsewhere).