A Fully Automatic Method for Breast Lesions Segmentation in Ultrasound Images

Due to inherent speckle noise, poor quality data, low contrast, and large shape variations of ultrasound lesions, automatic segmentation in breast ultrasound images is still a challenging task. In this paper, we propose a novel algorithm to segment automatically breast lesion in ultrasound images. The images are first filtered with a Speckle Reducing Anisotropic Diffusion algorithm to remove speckle noise. After that we propose the use of some morphological operators namely the contrast enhancement, the adaptive thresholding, the suppression of small regions and regions connected to the image border, the erosion and dilation, and finally the edge detection technique for initial lesion localization. Finally, we apply the Muti-scale Vector Field Convolution snake for boundary lesion segmentation. A comparative study with previous state of the art algorithms of active contours using both qualitative and quantitative measures for real breast ultrasound images are presented in this study in order to evaluate the performance of our method. Experiments demonstrate the ability to detect breast lesions automatically, quickly, efficiently and with high accuracy.