Combining global probability density difference and local gray level fitting for ultrasound image segmentation

Because of low signal-noise ratio (SNR), low contrast and blurry boundaries, the segmentation of ultrasound image becomes a difficult problem in the digital image processing field. In this paper, a novel active contour model combining global probability density difference and local gray level fitting is proposed for the segmentation of ultrasound image. In the proposed model, global information and local information are extracted in original ultrasound image and pre-processed image, respectively. In the original ultrasound image, by combining the background knowledge the distributions of gray levels of different regions are utilized for modeling the global information. For considering the local information, the ultrasound image is pre-processed, and in the pre-processed image, the local gray level fitting model is utilized for modeling the local information. By modeling the global and the local information in different images, the proposed method combines both approaches that utilize and remove speckle noise. With both simulated and clinical ultrasound images, the experimental results demonstrate that the proposed method is adaptive to the noise and robust to the initial conditions, and that it can segment ultrasound image accurately.