基于深度卷积神经网络的眼底豹纹分割量化及应用  

Fundus tessellation segmentation and quantization based on the deep convolution neural network

作　　者：郭振[1] 陈凌智 王立龙 吕传峰 谢国彤 高艳[1] 李君[1] Guo Zhen;Chen Lingzhi;Wang Lilong;Lyu Chuanfeng;Xie Guotong;Gao Yan;Li Jun(Qingdao Eye Hospital of Shandong Firs t Medical University,Shandong Eye Institute,State Key Laboratory Cultivation Base,Shandong Provincial Key Laboratory of Ophthalmology,Qingdao 266071,China;Ping An Healthcare Technology,Beijing 100010,China)

机构地区：[1]山东第一医科大学附属青岛眼科医院,山东省眼科研究所,山东省眼科学重点实验室-省部共建国家重点实验室培育基地,青岛266071 [2]平安医疗科技有限公司,北京100010

出　　处：《中华眼底病杂志》2022年第2期114-119,共6页Chinese Journal of Ocular Fundus Diseases

基　　金：山东第一医科大学学术提升计划(2019ZL001)。

摘　　要：目的基于深度卷积神经网络(DCNN)方法自动测量彩色眼底像上全局和局部豹纹分布密度。方法应用研究。将2021年5〜7月于山东第一医科大学附属青岛眼科医院北部院区行近视手术的患者514例1028只眼的1005张彩色眼底像建立人工智能(AI)数据库。采用RGB颜色通道重标定方法(CCR算法)、基于Lab颜色空间的CLAHE算法、多重迭代照度估计的Retinex算法、具有色彩保护的多尺度Retinex算法对图像进行预处理。对比观察上述4种图像增强方法以及使用Dice损失、边缘重叠率损失和中心线损失对豹纹分割模型效果的影响。建立眼底豹纹分割模型识别全图范围内豹纹结构区域;构建眼底组织结构检测模型用于视盘及黄斑中心凹定位。计算视野范围内后极部豹纹密度(FTD)、黄斑区豹纹密度(MTD)、视盘区豹纹密度(PTD)。结果应用CCR算法图像预处理和训练损失组合后,豹纹分割模型的Dice系数、准确率、灵敏度、特异性、约登指数分别达到0.7234、94.25%、74.03%、96.00%和70.03%。模型自动测量的FTD、MTD,PTD值与人工标注测量值平均绝对误差分别为0.0143、0.0207、0.0267,均方根误差则分别为0.0178、0.0323、0.0365。结论基于DCNN分割和检测方法能自动测量近视患者眼底全局和局部区域的豹纹分布密度,可以更准确地辅助临床监测和评估眼底豹纹改变对近视发展的影响。Objective To propose automatic measurement of global and local tessellation density on color fundus images based on a deep convolutional neural network(DCNN)method.Methods An applied study.An artificial intelligence(AI)database was constructed,which contained 1005 color fundus images captured from 1024 eyes of 514 myopic patients in the Northern Hospital of Qingdao Eye Hospital from May to July,2021.The images were preprocessed by using RGB color channel re-calibration method(CCR algorithm),CLAHE algorithm based on Lab color space,Retinex algorithm for multiple iterative illumination estimation,and multi-scale Retinex algorithm.The effects on the segmentation of tessellation by adopting the abovemetioned image enhancement methods and utilizing the Dice,Edge Overlap Rate and clDice loss were compared and observed.The tessellation segmentation model for extracting the tessellated region in the full fundus image as well as the tissue detection model for locating the optic disc and macular fovea were built up.Then,the fundus tessellation density(FTD),macular tessellation density(MTD)and peripapillary tessellation density(PTD)were calculated automatically.Results When applying CCR algorithm for image preprocessing and the training losses combination strategy,the Dice coefficient,accuracy,sensitivity,specificity and Jordan index for fundus tessellation segmentation were 0.7234,94.25%,74.03%,96.00%and 70.03%,respectively.Compared with the manual annotations,the mean absolute errors and root mean square errors of FTD,MTD,PTD automatically measured by the model were 0.0143,0.0207,0.0267 and 0.0178,0.0323,0.0365,respectively.Conclusion The DCNN-based segmentation and detection method can automatically measure the tessellation density in the global and local regions of the fundus of myopia patients,which can more accurately assist clinical monitoring and evaluation of the impact of fundus tessellation changes on the development of myopia.

关 键 词：近视 神经网络(计算机) 彩色眼底像 豹纹分布密度 

分 类 号：R778.11[医药卫生—眼科] TP183[医药卫生—临床医学] TP391.41[自动化与计算机技术—控制理论与控制工程]