In this paper, we present a new method to segment the walls of coronary arteries in IVUS (Intravascular Ultrasound) images based on a deformable model, which integrates both edge and region information. The whole image is supposed to have three regions - lumen, vessel wall, and adventitia plus surroundings, which are separated by two closed contours - the inner and outer boundaries. Our method has two steps: firstly, the outer vessel wall boundary is detected by minimizing an energy function of the contrast along it; secondly, by minimizing another energy function that considers the different gray level distributions of the lumen and the vessel wall, and the contrast along the edge between these two regions, the inner vessel wall is located. Dynamic programming is adopted to implement this method. Experimental results show that contrast information is a good feature for boundary detection in IVUS images.