Fully automatic detection and segmentation algorithm for ultrasound breast images using SVM and level set

Ultrasonic technology is one of the most important diagnostic tools for breast cancer detection. In this paper, we propose a fully automatic detection and segmentation algorithm of masses on breast ultrasound images by using support vector machine and level sets, which can be summarized in four distinct steps: 1)a SVM classifier for tumor detection is trained based on both gray and textural features; 2)all suspicious tumor regions, possibly including false positive regions, are detected with the trained SVM classifier; 3)an improved Chan-Vese (CV)active contour model, which adds a Bhattacharyya distance item to CV level set energy functional and performs better on ultrasound images, is used to segment regions accurately; 4)all regions are ranked by a score formula which combines area, position, gray level and textural information, and the highest score region is identified as the tumor region. The whole algorithm is completely automatic with no manual intervention. Experimental results demonstrate the high efficiency of the proposed method.