基于线性判别分析与超声图像特征的组织损伤检测方法  

Detection methods of tissue lesion based on linear discriminant analysis and ultrasonic image features

作　　者：陈华[1] 赵新民[1] 谭乔来[1] 邹孝[1] 钱盛友[1] 江剑晖 

机构地区：[1]湖南师范大学物理与信息科学学院,湖南长沙410081 [2]深圳普罗惠仁医学科技有限公司,广东深圳518067

出　　处：《中国医学影像技术》2016年第11期1757-1760,共4页Chinese Journal of Medical Imaging Technology

基　　金：国家自然科学基金项目(11174077;11474090);湖南省自然科学基金项目(11JJ3079)

摘　　要：目的探讨根据超声图像特征对高强度聚焦超声(HIFU)治疗中组织损伤情况进行无损检测的方法。方法对HIFU辐照新鲜离体猪肉组织前后获取的超声图像进行特征参量提取,并对其用线性判别分析(LDA)方法进行处理,再用处理后的参量结合LIBSVM进行识别。结果用LDA方法对减影图像的灰度均值和标准差处理后的参量进行识别,总识别率为86.07%;用LDA方法对HIFU辐照前后超声图像间的条件熵和相关系数处理后的参量进行识别,总识别率为92.62%。这两种方法所得到的识别率高于未用LDA方法。结论基于LDA方法的识别率要高于仅依据单个图像特征参量或将两个不同特征参量直接结合的识别方法,单参量识别时相关系数具有较明显优势。Objective To explore noninvasive method to detect tissue lesion during high intensity focused ultrasound（HIFU）treatment with ultrasonic image features.Methods Characteristic parameters were extracted from ultrasonic images obtained before and after HIFU radiation of fresh pork tissue in vitro,which were processed by linear discriminant analysis（LDA）method,then these processed parameters were used for identification with LIBSVM.Results The gray mean and standard deviation of the subtraction images were processed by the LDA method for identification,and its recognition rate was 86.07%;The conditional entropy and correlation coefficient from two ultrasonic images obtained before and after HIFU radiation were processed by the LDA method for identification,and its recognition rate was 92.62%.The recognition rate in these two situations was higher than that without using LDA method.Conclusion The recognition rate based on the LDA method is higher than that of others,such as the method only based on single characteristic parameter or direct combination of two different parameters,and the correlation coefficient had obvious advantages while single parameter is used for identification.

关 键 词：高强度聚焦超声 超声图像 组织损伤 线性判别分析 

分 类 号：R445.1[医药卫生—影像医学与核医学] R730.5[医药卫生—诊断学]