Intelligent Scissors for Medical Image Segmentation

Recently a new method for interactive image seg-mentation called Intelligent Scissors has been proposed by Mortensen and Barrett. Based on optimal path searching in a graph network, the method has a number of advantages over previous approaches. It is fast, robust, and very intuitively to use. Originally designed as a general-purpose tool for image manipulation , we have applied the method for medical image segmentation within an interactive contour editor. In this paper we describe some results and conclusions from our work. We also suggest a number of mod-iications of the original algorithm improving both, quality and performance of the method. A common requirement is to obtain smooth contours , i.e. contours free of sudden directional changes. Such smoothness constraints are quite diicult to implement using traditional edge costs, but may be achieved easily using a simple graph pruning strategy. This method not only avoids acute angles but also implies a 25% performance gain. Although the contours are no longer optimal in a global sense, we show that this is not of signiicance for practical applications .