基于先验概率和统计形状的前列腺超声图像自动分割方法  

Automatic Segmentation Method based on Probability Priors and Statistical Shape for Prostate TRUS Images

作　　者：黄建波[1] 倪东[1] 汪天富[1] 

机构地区：[1]医学超声关键技术国家地方联合工程实验室广东省生物医学信息检测与超声成像重点实验室深圳大学医学院生物医学工程系,深圳518060

出　　处：《生物医学工程研究》2015年第1期15-19,共5页Journal Of Biomedical Engineering Research

基　　金：国家自然科学基金资助项目(61101026)

摘　　要：从经直肠超声图像中自动精确地提取前列腺边界。采用基于先验概率和统计形状的前列腺超声图像自动分割新方法。首先,利用致密尺度不变特征变换,从超声图像中快速定位前列腺;其次,从多个统计形状模型中选择最优模型,在分割过程中,前列腺伪影区域缺失的边界信息可通过形状模型估计;最后,在最优形状模型指导下,采用多分辨率分割方式,利用局部灰度模型和局部高斯分布函数能量的最小化,实现前列腺的自动分割。用30幅超声图像测试得到平均Dice相似系数(DSC)为0.9552,平均绝对距离(MAD)的均值为0.5016 mm。该方法相比传统的形状模型的分割精度有较大提高。To automatically and accurately extract prostate boundary from 2D TRUS images. A novel method of utilizing probability priors and statistical shape for automatic prostate segmentation was presented. First,DENSE SIFT features of image were used to find the location of prostate in image quickly. Next,the optimal model from the multiple mean shape models by using the location was selected. During the segmentation process,missing boundaries in shadow areas were estimated by using the shape model. Last,with the guidance of this shape,the segmentation of an image was executed in a multi- resolution fashion,and an optimal search was performed by minimizing local gray model and local Gaussian distribution energy function for image segmentation. The result showed that the value of average dice similarity coefficient was 0. 9552 and the error of average mean absolute distance was 0. 5016 mm for 30 images. The result demonstrates that the accuracy of this method is obviously improved compared with the traditional Active Shape Model method.

关 键 词：经直肠超声图像 点分布模型 前列腺分割 多个平均模型 局部高斯分布 致密尺度不变特征变换 

分 类 号：R318[医药卫生—生物医学工程]