分形理论在数字病理图像分析中应用的可行性研究  

Study on feasibility of apply typing theory to digital pathological image analysis

作　　者：曹入双 何坤 徐容 杨茂 甘仲霖 Cao Rushuang;He Kun;Xu Rong;Yang Mao;Gan Zhonglin(School of Public Health,Southwest Medical University,Luzhou,Sichuan,646000,China;Institute for Cancer Medicine,School of Basic Medical Science,Southwest Medical University,Luzhou,Sichuan,646000,China)

机构地区：[1]西南医科大学公共卫生学院,四川泸州646000 [2]西南医科大学基础医学院肿瘤医学研究所

出　　处：《齐齐哈尔医学院学报》2023年第16期1501-1505,共5页Journal of Qiqihar Medical University

摘　　要：目的 以SD大鼠各脏器组织为材料,分析其分形特征并计算其分形维数,利用分形维数进行判别分析,探讨分形理论在数字病理图像分析中应用的可行性,促进数字病理图像分析技术发展,提高临床病理诊断水平。方法 选取洁净动物房饲养的SD大鼠作为实验对象,对其心脏、肝脏、脑、胰腺、脾脏、肺、肾脏取材后进行石蜡包埋切片、苏木精-伊红染色及组织病理学检查与图像拍摄。将各脏器组织显微彩色图裁剪为1 024×1 024像素大小并转换为二位图像,导入Fractal3e软件进行分形分析,并应用SPSS软件进行判别分析。结果 对各样本二位图像进行分形分析,相关系数r在-0.9925~-0.9998之间,P<0.001,lgN与lgδ有明显线性关系,各组脏器具备明显分形特征。各组脏器显微图像分形维数均不同,表明各组脏器显微图像在粗糙程度上存在差异,其分形维数分别为:心脏:1.308±0.152,肝脏:1.632±0.105,脑:1.285±0.109,脾脏:1.790±0.054,肺:1.643±0.050,肾脏:1.557±0.101,胰腺:1.743±0.057;不同类型组织间可分性判断值G以脾脏与心脏、脑与脾脏、脑与胰腺较为显著,分别为:1.129,1.558,1.262;应用SPSS进行判别分析,各脏器Fisher线性判别函数表达式分别为:心脏=140.733-93.95X1,肝脏=175.661-145.285X1,脑=138.290-90.783X1,脾脏=192.656-174.364X1,肺=176.828-147.197X1,肾脏=167.674-162.547X1,胰腺=187.640-162.503X1;应用Bayes判别分析判别法正判率为49.3%。结论 大鼠各个脏器组织具有典型的分形特征,不同脏器组织分形维数不同,故分形维数可在脏器组织判别分析时作为判别参量。某组织脏器分形维数的改变可提示该组织脏器出现病变,分形理论可作为临床病理诊断的重要参考。Objective To analyze the fractal characteristics and calculate the fractal dimension of various organs of SD rats,and to use the fractal dimension for discriminant analysis,to explore the feasibility of applying fractal theory in digital pathology image analysis,so as to promote the development of digital pathology image analysis technology,and improve the clinical pathology diagnosis.Methods SD rats housed in clean animal rooms were selected as the experimental subjects,and their hearts,livers,brains,pancreases,spleens,lungs and kidneys were extracted and then subjected to paraffin-embedded sections,hematoxylin-eosin staining and histopathological examination and image capture.The microscopic colour images of each organ were cropped to a size of 1024×1024 pixels and converted to two-bit images,imported into Fractal3e software for fractal analysis,and discriminant analysis was performed using SPSS software.Results The fractal analysis was performed on each sample.The correlation coefficients ranged from-0.9925 to-0.9998,P<0.001.The correlation coefficients between lgN and lgδwere linear,and each group of organs had obvious fractal characteristics.The fractal dimensions of the microscopic images of the organs in each group were different,indicating that there were differences in the degree of roughness of the microscopic images of the organs in each group,and their fractal dimensions were as follows:heart:1.308±0.152;liver:1.632±0.105;brain:1.285±0.109;spleen:1.790±0.054;lung:1.643±0.050,kidney:1.557±0.101,pancreas:1.743±0.057;the differentiability judgement values G between different types of tissues were more significant for spleen and heart,brain and spleen,and brain and pancreas,they were 1.129,1.558,and 1.262,respectively.Using SPSS for discriminant analysis,the expressions of Fisher's linear discriminant function for each organ were:heart=140.733-93.95 X1,liver=175.661-145.285X1,brain=138.290-90.783X1,spleen=192.656-174.364X1,lung=176.828-147.197X1,kidney=167.674-162.547X1,and pancreas=187.640-162.5

关 键 词：分形理论 分维值 组织分类 

分 类 号：R361[医药卫生—病理学]