Robust topology-adaptive snakes for medical ultrasonic image segmentation

In this study, a modified topology-adaptive snake (T-snake) is proposed for the segmentation of ultrasound image. The algorithm was improved as follows. First, the image is decomposed in the place which has offset from pixel´s position while snake points are in pixel´s position. This will reduce the task during calculating intersections between contour and ACID grid. Second, the rule to process topology conflict is simplified and there is no need to judge triangle point in or out of the contour in our model. Third, since ultrasound image has a lot of speckle noise, our external energy is composed by three parts-the gradient-based image energy, the inflation energy and region-based image energy, which can push T-snake into the real edge. The proposed model is tested by both synthetic images and real ultrasound images. Experiments show that our algorithm has the advantage of topological adaptability, less sensitive to the initial contour and speckle noise.