An improved active contour model for segmentation of medical images

The application of active contour models (ACM) is a robust method for segmenting noisy images with sparse image features. This well-known technique adjusts an initially given rough approximation of a contour to the location of image features. The main shortcoming is that the result is largely dependent on the shape and location of the initial contour. The reason for this is the local property of the ACM's energy minimisation process. In this paper, we propose two globally working energy minimisation approaches to overcome this problem. In order to be able to selectively define the energy for the region of interest, we include a region oriented energy term that evaluates the homogeneousness of the area enclosed by the curve. This paper discusses the proposed scheme in detail and shows results of applying it to CT, MR, and ultrasonic images.