Automatic breast ultrasound image segmentation: A survey

Min Xian, Yingtao Zhang, H.D. Cheng, Fei Xu, Boyu Zhang, Jianrui Ding
2018 Pattern Recognition  
Breast cancer is one of the leading causes of cancer death among women worldwide. In clinical routine, automatic breast ultrasound (BUS) image segmentation is very challenging and essential for cancer diagnosis and treatment planning. Many BUS segmentation approaches have been studied in the last two decades, and have been proved to be effective on private datasets. Currently, the advancement of BUS image segmentation seems to meet its bottleneck. The improvement of the performance is
more » ... ly challenging, and only few new approaches were published in the last several years. It is the time to look at the field by reviewing previous approaches comprehensively and to investigate the future directions. In this paper, we study the basic ideas, theories,pros and cons of the approaches, group them into categories, and extensively review each category in depth by discussing the principles, application issues, and advantages/disadvantages. Keyword: breast ultrasound (BUS) images, breast cancer, segmentation, benchmark, early detection, computer-aided diagnosis (CAD) The rest of the paper is organized as follows: in sections 2 -5, we review automatic BUS image segmentation methods by presenting the principle of each category, discussing their advantages and disadvantages, and summarizing the most valuable strategies. In section 6, we discuss the approaches of three small sub-categories briefly. In section 7, the fundamental issues in BUS segmentation are discussed, e.g., denoising, interaction, biological priors modeling, validation, and the possible problem-solving strategies. Section 8 is the conclusion and the future directions. Graph-based approaches Graph-based approaches gain increasing popularity in BUS image segmentation because they offer several advantages: (1) they provide a simple way to organize task-related priors and image information in a unified framework; (2) they are flexible and suitable for expressing soft constraints between random variables; and (3) the computation based on graphical manipulation is very efficient [5]. Let = ( , ℰ) be a graph comprising a set of nodes (vertices) = { , , ⋯ , }, and each of them corresponds to an image pixel or superpixel; and a set of links (edges) ℰ = 〈 , 〉 , ∈ , and each of them connects two adjacent nodes according to a predefined neighborhood system = { | = 1, ⋯ , } where is the set of neighbors of node . Each link 〈 , 〉 is associated with a nonnegative weight ( , ). The weight is usually defined as the cost of separating the two connected nodes into different classes. The segmentation of an image is transferred to partition the graph into non-overlap subgraphs. The BUS segmentation is usually modeled as a bi-segmentation problem; therefore, the goal is to partition * = argmax ( | ) = argmax ( | ) ( )
doi:10.1016/j.patcog.2018.02.012 fatcat:mm6vwa7c3nhmvmpnmiegz6edue