基于深度学习技术分析健康人眼底血管图像特征  

Analysis of image features of fundus blood vessel in healthy human eye based on deep learning techniques

作　　者：惠梦雨 石晶琳[2] 俞晓菡 李舰 张彦 汤峥丽[5] 余上海 高悦 刘平 张华 Hui Mengyu;Shi Jinglin;Yu Xiaohan;Li Jian;Zhang Yan;Tang Zhengli;Yu Shanghai;Gao Yue;Liu Ping;Zhang Hua(Institute of Hepatology,Shuguang Hospital Affiliated to Shanghai University of Traditional Chinese Medicine,Shanghai 201203,China;Department of Ophthalmology,Shuguang Hospital Affiliated to Shanghai University of Traditional Chinese Medicine,Shanghai 201203,China;JF Healthcare Corp,Nanchang 330095,Jiangxi Province,China;Shanghai Donghai Vocational and Technical College,Shanghai 200241,China;Department of Medical Check-up Center,Shuguang Hospital Affiliated to Shanghai University of Traditional Chinese Medicine,Shanghai 201203,China;Institute of Interdisciplinary Integrative Medicine Research,Shanghai University of Traditional Chinese Medicine,Shanghai 201203,China)

机构地区：[1]上海中医药大学附属曙光医院肝病研究所,上海市201203 [2]上海中医药大学附属曙光医院眼科,上海市201203 [3]江西中科九峰智慧医疗科技有限公司,江西省南昌市330095 [4]中国上海市东海职业技术学院,200241 [5]上海中医药大学附属曙光医院体检中心,上海市201203 [6]上海中医药大学交叉科学研究院,上海市201203

出　　处：《国际眼科杂志》2024年第10期1542-1550,共9页International Eye Science

基　　金：上海市卫生和计划生育委员会中医药科技创新项目(No.ZYKC201702005)。

摘　　要：目的:基于深度学习技术探讨健康人眼底血管特征,以期发现眼底动、静脉的正常值范围,以及性别、年龄、体质量指数(BMI)、血压等生理因素与眼底血管特征间的关系。方法:运用专业的眼底照相机拍摄健康人眼底图像,并采集受试者的血压及化验检查指标。通过改进的U-Net模型分割眼底动、静脉血管,采用计算机视觉技术提取血管颜色、形态及Haralick纹理特征。结果:共拍摄体检者眼底图像4487例,筛选出健康且眼底图像清晰者326例,其中男200例,女126例。左、右眼以及眼底动、静脉的形态、颜色及纹理特征上存在差异,血管宽度(width)动脉均值为1.146、静脉为1.430,动静脉比约为4∶5。不同年龄段(21-30、31-40、41-50岁)健康人眼底动、静脉特征:与21-30岁及31-40岁的比,41-50岁的动、静脉idm、f12特征值以及静脉asm值增大,动、静脉con、ent、den特征值以及静脉sen值降低。与21-30岁比,31-40岁的动脉f12值增大、静脉con值降低(均P<0.05)。不同性别健康人眼底血管特征:与男性比,女性眼底动、静脉的sav、sva值、动脉的curved以及静脉的b mean、bsd、var、sen、ent值均增大,静脉的area值减小(均P<0.05)。不同程度BMI的健康人眼底动、静脉特征无差异(均P>0.05)。不同程度血压健康人的眼底动脉area、width,静脉con、idm、dva、den值具有差异(均P<0.05)。结论:健康人左、右眼及眼底动、静脉间特征均有差异,与性别、年龄及血压等生理因素具有相关性,具有潜在的微循环标志物的价值。AIM:To explore the fundus vascular characteristics of healthy individuals based on deep learning techniques,with a view to discovering the range of normal values of the fundus arteries and veins,as well as the relationship between physiological factors,such as gender,age,body mass index(BMI),blood pressure,and fundus vasculature characteristics.METHODS:Fundus images of healthy people were taken from a professional fundus camera,and the subject’s blood pressure and laboratory test was collected.Additionally,the fundus arteries and veins were segmented by the improved U-Net model,and the color,morphology and Haralick texture features of the vessels were extracted from computer vision technology.RESULTS:A total of 4487 cases fundus images were taken and 326 cases with healthy and clear fundus images were screened,including 200 males and 126 females.There were differences in the morphology,color,and textural characteristics of the left and right eyes,as well as of the fundus arterioles and veins,with a mean vessel width(width)of 1.146 in the arteries and 1.430 in the veins,and an arteriovenous ratio about 4∶5.Fundus artery and vein characteristics in healthy individuals of different ages(21-30,31-40,41-50):compared with the healthy population aged 21-30 and 31-40 years,arterial and venous inverse difference moment(idm),f12 and venous angular second moment(asm)values increased,and arterial and venous contrast(con),entropy(ent),difference entropy(den),and venous sum entropy(sen)values decreased in 41-50 years.Compared with the 21-30 years age group,arterial f12 values increased and venous con values decreased in 31-40 years(all P<0.05).Fundus vascular characteristics of healthy individuals of different sexes:compared with male,fundus arterial and venous sum average(sav),sum variance(sva)values,arterial curved values,and venous b mean,bsd,variance(var),sen,ent values increased in female,while venous area value of female decreased(all P<0.05).There were no statistically significant differences in fundus arteriosus an

关 键 词：健康人 眼底图像 深度学习 眼底血管(动脉、静脉) 

分 类 号：R770.4[医药卫生—眼科]