Brownian strings: image segmentation with stochastically deformable models

This paper describes an image segmentation technique in which an arbitrarily shaped contour is deformed stochastically until it fits around an object of interest. The evolution of the contour is controlled by a simulated annealing process which causes the contour to settle into the global minimum of an image-derived 'energy' function which is designed to be small when the contour is near the border of objects similar to the target. The nonparametric energy function is derived from the statistical properties of similar previously segmented images, thereby incorporating prior experience. Since the method is based on a state space search for the contour with the best global properties, it is stable in the presence of image errors which confound segmentation techniques based on local criteria such as connectivity. However, unlike 'snakes' and other active contour approaches, the new method can handle arbitrarily irregular contours in which each inter-pixel crack represents an independent degree of freedom. The method is illustrated by using it to find the brain surface in magnetic resonance head images, to identify the epicardial surface in magnetic resonance cardiac images, and to track blood vessels in angiograms.