Robust Segmentation and Tracking of Generic Shapes of Neuro-stem Cells

Given the time lapse images of human Neuro Stem Cells (hNSC) marked by fluorescent proteins, that are obtained from a confocal laser microscope, we present algorithms to identify, segment, track, and estimate statistical parameters of the cells. The structure of these cells are quite complex and irregular, which makes segmentation and tracking even more challenging. We use a novel combination of Difference of Gaussians and a variant of the Watershed algorithm to segment cells accurately. Our tracking algorithm can identify not only the temporal path of the cells but also events like cell divisions and deaths. Our system is robust, efficient, completely automatic, and removes many drawbacks faced by earlier solutions. We also propose the first geometric algorithm that uses Delaunay triangulation, to find the number of the branches of the cells, which is an important biological feature.