Automatic tracking of biological cells and compartments using particle filters and active contours

In order to gain a detailed understanding of cell signalling and regulation processes, live cell imaging is often used. Such signalling can involve the nuclear-cytoplasmic translocation of signalling molecules, and to follow this, it is necessary to mark the cell and nuclear boundaries in an image sequence in individual cells. For complex processes, with both rapid cell motion and nuclear translocation, it is extremely time-consuming to mark such cell boundaries manually. The particle filter (PF) is a Monte Carlo method based on sequential importance sampling. It is robust to clutter and the occlusion of targets in object tracking. The active contour (AC) method is based on a deformable model and is capable of capturing minor cell shape variations, although a good initialisation is often required. Prior information from a particle filter provides a reasonable initialisation for the active contour method and prevents it from being trapped in local minima. Combining these two methods (PF-AC) thus makes it possible to track complex cellular processes automatically. The combined method is used to study the important NF-κB signalling pathway. Comparable results to those obtained manually by a biologist are obtained using the hyphenated approach, but in a small fraction of the time.