Biologists routinely use microscopy techniques that capture time-lapse image sequences showing the development of clones of stem and cancer cells in 3-D from intact tissue samples and in 2-D from in vitro cell cultures. These image sequences show not just proliferating cells, but also the dynamic behaviors of vasculature and sub-cellular organelles. High throughput microscopy makes it possible to capture hundreds of such image sequences rapidly and with minimal human effort, usually with expensive, dedicated systems.
The analysis of this image sequence data is a two-step process, and is the focus of our research. First, the dynamic phenotype that represents all of the observable and measurable aspects of the objects being imaged must be acquired and the tracking and lineaging must be completely corrected. Second, the corrected dynamic phenotype must be analyzed by computational sensing techniques that are capable of identifying the differences in dynamic behaviors corresponding to functionally or experimentally different populations.Current approaches to analyzing the dynamics of stem cell behavior focus on univariate probabilistic modeling of parameters of the lineage tree. My interest is in developing approaches that incorporate not just the parameters of population growth obtained from the lineage tree but also include spatiotemporal dynamics of motion, morphology and association of the cells, organelles and blood vessels.
We use computer gaming hardware for enhanced and affordable 3-D visualization, along with techniques from information theory and machine learning to combine the exquisite capabilities of the human visual system with computational sensing techniques for analyzing vast quantities of image sequence data.