Abstract
Breast ultrasound image segmentation is challenging task due to the low quality of ultrasound images and the complex breast structure. An accurate and automatic algorithm is presented to segment breast ultrasound images by combining image boundary and region information. The algorithm decomposes the image into a set of superpixels using the Normalized Cuts method along with texture analysis. An SVM classifier is employed to estimate the tumor likelihood of each superpixel based on five texture features. A seed superpixel is identified based on the tumor likelihoods and spatial locations of the superpixels. The seed superpixel is extended to accurately highlight the tumor region using a region growing approach that combines both the superpixels tumor likelihoods and edge-based analysis. The proposed algorithm and two popular segmentation algorithms are used to segment 50 breast ultrasound images The proposed algorithm achieved higher sensitivity and lower error rates compared to the two existing algorithms.