Segmentation Of Medical Ultrasonic Image Based On Active Contour Model

Medical ultrasound imaging is widely used in medical diagnosis and treatment due to its unique characteristic of noninvasiveness, real-time, repetition, cheapness and high sensitivity. Ultrasound medical image segmentation is the essential step of ultrasound image processing, and it plays a crucial role in both qualitative and quantitative ultrasound image analyses. Currently, the most widely used segmentation approach in a clinical ultrasound image system is based on manual delineation. To accomplish ultrasound image segmentation more efficiently and accurately, a computerized approach would be an ideal choice for clinical use. A computerized approach is expected to segment the object of interest automatically or semi-automatically with high reproducibility.Active contour model is widely used in segmentation of medical image due to it can combine the high-level vision knowledge and low-level image information. In this study, the active contour model is applied to segmentation of medical ultrasonic image. By analyzing the original active contour model and some main modified models, the internal energy and external energy of active contour model are modified as follows. 1. To avoid the convergence of neighbor control points, the average contour length term is added into the internal energy of the model. 2. To avoid the active contour find wrong boundary, the gradient directional energy is introduced to the external energy of the model. 3. Since the ultrasound image has a lot of speckle noise, the region energy based statistic characteristic of image is regarded as the external energy, so the model can overcome the influence of the speckle noise. Then, the greedy algorithm is used to find the minimum energy of the model. A fast algorithm is introduced to solve the minimum of region energy. Finally, multi-resolution optimization method is used to improve the convergence speed of the model further.The proposed model is tested by both synthetic image and real ultrasound image. Various levels of Gaussian noise (from 10% to 60%) is added into the synthetic image. Segmentation of the ultrasound breast tumor image and liver tumor image are studied in the experiments. Comparing to the traditional active contour model and the...