斑点噪声分布拟合的乳腺超声病灶分割方法  

Breast ultrasound lesion segmentation by level set fitting with speckle noise distribution

作　　者：杨谊[1] 喻德旷[1] 申洪[1] 

机构地区：[1]南方医科大学,广东广州510515

出　　处：《中国体视学与图像分析》2014年第2期103-111,共9页Chinese Journal of Stereology and Image Analysis

基　　金：广东省科技计划项目(2010B060300001);广东省产学研项目2011B090400037

摘　　要：乳腺超声图像常存在斑点噪声,给自动分割带来了一定干扰。为了提升图像质量而进行的去噪操作可能导致病灶边界信息的损失。本文探索将噪声处理融合到分割中而提出新方法,采用统计特征概率分布形式描述乘性斑点噪声分布并融入水平集模型,从而将图像去噪与目标分割结合成一个统一体,实现此类型噪声分布图像的高效自动分割。本文方法对人工合成图像分割得到了较好的效果,对真实乳腺超声图像病灶的分割测试实现了速度提升,在分割精度方法则显示出分化结果,通过分析原因指出探索性的改进措施。Speckle noise in breast ultrasound image disturbs the automatic segmentation of lesion area. Traditional noise suppression methods often lead to loss of edge information. A novel proposal of incorpora- ting noise suppression and segmentation into level set model is proposed. With analysis of multiple speckle noise property and description in form of probability density distribution, a new level set model based on Chan-Vese model implements the fitting automatic segmentation and noise suppression, in which speckle noise is combined with object segmentation and obtained better results. Experiments on synthetic images showed good results, and experiments on breast ultrasound lesion segmentation showed enhanced speed, but with varying accuracy results, including both better and slightly worse, which direct the need of fur-ther exploration.

关 键 词：水平集方法 乘性斑点噪声 概率密度分布 乳腺超声病灶分割 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]