癌变组织偏振多参数识别与纹理特征分析  

Polarization Multi-parameter Recognition and Texture Feature Analysis of Cancerous Tissue

作　　者：张丽丽 黄丹飞[1,2] 高君朝[1,2] 宋东 洪景辉[3] 张勇[1,2] 唐鸿宇 张乐超[1,2] ZHANG Lili;HUANG Danfei;GAO Junzhao;SONG Dong;HONG Jinghui;ZHANG Yong;TANG Hongyu;ZHANG Lechao(College of Optoelectronic Engineering,Changchun University of Science and Technology,Changchun 130022,China;Zhongshan Research Institute of Changchun University of Science and Technology,Zhongshan 528437,China;The First Hospital of Jilin University,Changchun 130021,China)

机构地区：[1]长春理工大学光电工程学院,长春130022 [2]长春理工大学中山研究院,中山528437 [3]吉林大学第一医院,长春130021

出　　处：《光子学报》2024年第6期199-213,共15页Acta Photonica Sinica

基　　金：国家自然科学基金(No.61893096014)。

摘　　要：偏振成像技术中穆勒矩阵元素含义模糊、单一衍生参数信息解读受限,为了增强原有数据分析范式的可解释性和强关联性,提出一种癌变组织偏振多参数特征识别方法,并通过纹理分析来定量描述癌症组织的偏振差异。基于双波片旋转法穆勒矩阵测量方案升级搭建背向散射偏振成像系统,采集未染色肺癌与基底细胞癌切片。引入旋转不变量获取高维偏振参数集,生成偏振多参数特征曲线,实现病理区域多维特征提取与可视化,解决直接使用穆勒矩阵受方向影响的问题。计算了灰度共生矩阵和Tamura纹理属性,结果表明,当固定纹理维度时,不同偏振参数的癌变识别能力不同,单个纹理属性可成为多个偏振图像的共同量化指标;当固定偏振维度时,不同纹理属性有望成为单一偏振维度的多个辅助定量指标。该方法快速高效,可为辅助病理检测提供新思路,在临床上展现出良好的应用前景。At present,microscopic observation of pathological sections is still the gold standard for pathological diagnosis of cancer.The complex production process and manual detection process of pathological sections make pathological detection subjective and inefficient.Polarization imaging technology is sensitive to sub wavelength structures,and exploring an objective and efficient method for identifying cancerous tissue using polarization images has unique advantages in enhancing pathological diagnostic capabilities.In this article,based on the Muller matrix measurement scheme of double wave plate rotation method,a backscatter polarization imaging system is built and upgrated equipped with a tunable zoom microscope lens to meet the requirements of different resolution fields.The slice data of unstained lung cancer and basal cell cancerare collected,and the Muller matrix is obtained from 30 polarization intensity images based on the Fourier coefficient relationship.In order to enhance the interpretability and strong correlation of the original data analysis paradigm in which the meaning of the Mueller matrix elements is unclear and the interpretation of the information for a single polarization parameter is limited,we report a polarization multi-parameter feature recognition and texture feature analysis method for cancerous tissue.To overcome the limitation that a single Muller matrix image cannot accurately and comprehensively identify the structure of pathological tissue,we introduce rotation invariants to obtain a high-dimensional polarization parameter set,then select regions of interest randomly and generate polarization multi-parameter feature curves to achieve multi-dimensional feature extraction and visualization of pathological regions,solving the problem of direct use of Muller matrix affected by direction.At the same time,in order to further obtain organizational information from derived parameters,4 texture attributes from gray level co-occurrence matrix and 6 texture attributes from Tamura are calculated t

关 键 词：偏振成像 穆勒矩阵 偏振参数 纹理特征 癌变组织 

分 类 号：R730.4[医药卫生—肿瘤] TP391.41[医药卫生—临床医学] O436.3[自动化与计算机技术—计算机应用技术]