针对肝包膜特征图的自动化提取方法  

Automated Extraction Method for Liver Capsule Feature Maps

作　　者：牛广利 刘翔[1] 宋家琳 汤显 NIU Guangli;LIU Xiang;SONG Jialin;TANG Xian(School of Electronic and Electrical Engineering,Shanghai University of Engineering Science,Shanghai 201620,China;Department of Ultrasound Diagnosis and Treatment,Changzheng Hospital,The Second Military Medical University of Chinese People′s Liberation Army,Shanghai 200003,China)

机构地区：[1]上海工程技术大学电子电气工程学院,上海201620 [2]中国人民解放军第二军医大学长征医院超声诊疗科,上海200003

出　　处：《电子科技》2022年第10期39-44,共6页Electronic Science and Technology

基　　金：国家自然科学基金(81101105);上海市自然科学基金(19ZR1421500)。

摘　　要：为了自动化提取肝包膜及其上下组织特征图,实现全自动特征学习,文中提出采用频域处理与图像形态学处理的方法对图像进行预处理,并借鉴移动平均法提出二路交叉感受野策略,由感受野映射区域进行特征筛选与分析。通过对数能量函数识别并定位目标区块,从而实现对肝实质病变特征、肝包膜、肌肉脂肪层纹理特征数据提取与分析,并根据数据分析获取肝包膜及其上下组织特征图。根据特征区域的相对位置,提出区块纠错机制对误检区块进行校正,使其更具鲁棒性。实验结果表明,在对肝硬化超声图像中的肝包膜及其上下组织特征图的提取过程中,该提取机制在正常、轻度、中度阶段特征提取均达到100%的准确率,对于重度病情阶段的特征提取准确率达到84.6%。In order to automatically extract the feature maps of the liver capsule and its upper and lower tissues,and realize automatic feature learning,the study proposes to use frequency domain processing and image morphology processing to preprocess the image,and proposes a two-way cross receptive field strategy based on the moving average method,and feature screening and analysis are carried out through the receptive field mapping area.The logarithmic energy function is used to identify and locate the target block,so as to realize the extraction and analysis of the liver parenchymal lesion features,liver capsule,muscle fat layer texture feature data,and obtain the liver capsule and its upper and lower tissue feature maps according to the data analysis.Based on the relative positions of the proposed feature regions,a block correction mechanism is proposed to correct the mischecked blocks to make them more robust.The experiments show that during the extraction of the liver envelope and its upper and lower tissue feature maps in the ultrasound images of cirrhosis,the present extraction mechanism achieves 100%accuracy in the normal,mild and moderate stages of feature extraction,and 84.6%accuracy is achieved in the severe disease stage.

关 键 词：高频超声图像 肝硬化 傅里叶变换 肝包膜 二路交叉感受野 对数能量收益函数 移动平均法 纠错机制 

分 类 号：TN911.73[电子电信—通信与信息系统]