Cancer Area Extraction for Uterine Cervical Image Based on Spectral Matting
基于谱抠图的宫颈图像癌症区域的提取 Cancer Area Extraction for Uterine Cervical Image Based on Spectral Matting
张子倩 温铁祥 辜嘉 王磊
摘 要 准确的提取病变区域对于机器人辅助外科手术系统和计算机辅助诊断系统是非常重要的.我们以谱抠图为基础提取宫 颈图像中的癌症区域(醋酸白上皮组织).谱抠图引入了基本抠图组成部分的概念.实验结果表明对于大量的清晰的醋酸白上 皮组织,几乎所有的组织细节都可以通过用户指导下的抠图提取出来.除此之外,一种扩展的用户交互式模式减少了用户的工 作量并能避免不完整的提取结果:在这种模式中,用户得以提供一些抠图的选项来进行选择.另外,将宫颈图像从RGB颜色模 型转换为XYZ颜色模型可以大大提高基于谱抠图的癌症区域提取的性能. 关键词 宫颈图像;癌症区域提取;谱抠图;颜色空间模型 ABSTRACT Extracting pathology area accurately is of great importance for robot assisted surgery system and computer aided diagnosis system. We extracted cancer area (acetowhite epithelium) in uterine cervical image based on spectral matting, which introduced the concept of fundamental matting components. Experiment results showed that for large and clear acetowhite epithelium area, almost all of details of tissue could be detected by user-guided matting. Besides, an extended user interaction mode reduces user effort and may avoid incomplete extraction result: in this mode the user is presented with several matting alternatives to choose from. In addition, transforming uterine cervical images from RGB model to XYZ color model could greatly improve the performance of cancer area extraction based on spectral matting. KEYWORDS Uterine Cervix Image; Cancer Area Extraction; Spectral Matting; Color Space Model
1 Introduction
M
inimal invasive procedure offers advantages of reduced pain, shorter hospital stay, and quicker convalescence for the patients. However, due to the fatigue arising from long operation, there may be errors in operator's vision and operation. Robot assisted surgery system have been developed to provide more precise positioning and stable images. Extracting region of interest accurately has an important significance in image navigation system, which guides robot to position accurately. Based on accurate extraction of pathology area, robot can avoid injury on normal tissue, and then ensuring sufferer safety. Computer assisted diagnosis system can screen people for cancer and provides patients and doctors with early warning by identifies suspicious cancer tissue. Accurate extraction of cancer tissue is the base upon which computer assisted diagnosis system can provide accurate diagnosis. Natural image matting is the process of extracting a foreground object from an image along with an opacity estimate for each pixel covered by the object [1]. All of existing methods require the users to provide additional constrains in the form of a trimap [2,3,4] or a set of brush strokes [5,6,7] ,since matting problem is inherently under-

下一页