基于眼底图像的糖尿病视网膜病变分类系统设计
投稿时间:2020-02-03  修订日期:2020-06-15  点此下载全文
引用本文:
摘要点击次数: 173
全文下载次数: 0
作者单位邮编
许梦莹 河南科技大学 医学技术与工程学院 471000
李振伟* 河南科技大学 医学技术与工程学院 471000
杨晓利 河南科技大学 医学技术与工程学院 
贾蒙丽 河南科技大学 医学技术与工程学院 
基金项目:河南省科技发展计划项目(202102310534)
中文摘要:摘要: 针对眼底图像,设计一个糖尿病视网膜病变(Diabeticretinopathy , DR)分类系统,通过对视网膜血管图像进行定量分析来实现对DR病程的分类。采用Messidor数据集的眼底照片图像,这个数据集共包含100个研究项目,其中32张未患DR的眼底照片,24张患NPDR。根据数据集中DR患者和非DR人群的眼底图像以及眼科专家的分类结果,利用数字图像处理技术分析特征值的统计意义,判断该图像所反映的DR病程。预处理为提取特征值前的图像增强、主像素成分分析、匹配滤波以及Gabor滤波,对预处理后的图像进行直径、角度和分形维数等特征值提取。最终结果展示了直径、角度和分形维数的准确率达到了93%、96%、81.8%,提供有效的辅助诊断手段。糖尿病视网膜病变的特征值分析包括直径、角度和分形维数准确率较高。对于缺乏医疗条件的地区很有价值。
中文关键词:眼底图像  糖尿病视网膜病变  滤波  特征提取
 
Design of a classification system fordiabetic retinopathy based on fundus images
Abstract:Abstract: Design a diabetic retinopathy (DR) classification system by quantitative analysis of retinal blood vessel images to classify the course of DR. Using the Messidor dataset. This dataset contains 100images, of which 32 fundus photos without DR and 24 with NPDR. According to the fundus images of DR and non-DR patients in the database and the classification results of ophthalmologists, digital image processing techniques to analyze the statistical significance of the feature values and judge the DR course. The pre-processing includes image enhancement, principal pixel component analysis, matched filtering, and Gabor filtering before extracting eigenvalues, and extracts eigenvalues such as diameter, angle, and fractal dimension from the pre-processed image. The final results show that the accuracy of the diameter, angle and fractal dimension reached 93%, 96%, 81.8%, providing effective aided diagnostic methods. The DR eigenvalue analysis, including diameter, angle and fractal dimension, has a high accuracy rate. It is very valuable in areas without medical conditions.
keywords:Fundus Image  DR  Filtering  Feature Extraction
查看全文   查看/发表评论   下载pdf阅读器