聚合多通道特征的青光眼自动检测

发布时间：2018-07-22 16:18

【摘要】：眼底影像的自动分析是计算机辅助青光眼筛查和诊断的关键基础.为提高青光眼辅助诊断的准确性,基于彩色眼底图,提出一种聚合多通道特征的青光眼自动检测算法.首先基于多尺度分析技术,通过聚合多通道图像特征,从颜色分布、多尺度Gabor滤波和梯度方向分布等方面表示视盘形态和结构在彩色眼底图上的细微变化;然后设计基于随机森林的分类器,在青光眼数据集上训练分类器模型,并利用集成学习技术鉴别青光眼,从而实现一种基于图像特征的青光眼自动检测算法;最后在2个具有挑战性的青光眼公开数据集(RIM-ONE r2和Drishti_GS)上对青光眼检测算法进行测试和验证,分别得到了0.869 0和0.800 4的曲线下面积值.实验结果表明,该算法在保证青光眼检测敏感性的同时能够显著提高其特异性,对青光眼辅助筛查和诊断具有很好的参考价值.
[Abstract]:Automatic analysis of fundus images is the key to computer-aided glaucoma screening and diagnosis. In order to improve the accuracy of assisted glaucoma diagnosis, an automatic glaucoma detection algorithm based on color fundus image was proposed. Firstly, based on multi-scale analysis technology, the color distribution, multi-scale Gabor filtering and gradient direction distribution are used to show the subtle changes of the shape and structure of the disc in the color fundus image by aggregating the multi-channel image features. Then a classifier based on random forest is designed to train the classifier model on the glaucoma data set and an automatic glaucoma detection algorithm based on image features is realized by using integrated learning technology to identify glaucoma. Finally, two challenging open data sets of glaucoma (RIM-ONE R2 and Drishti GS) were used to test and verify the glaucoma detection algorithm, and the area values under the curve of 0.869 0 and 0.800 4 were obtained, respectively. The experimental results show that the algorithm not only ensures the sensitivity of glaucoma detection, but also improves its specificity, which has a good reference value for glaucoma assisted screening and diagnosis.
【作者单位】：中南大学信息科学与工程学院;中南大学眼科医学影像处理中心;中南大学湘雅二医院眼科;中南大学文学与新闻传播学院;
【基金】：国家自然科学基金(61573380,61672542)
【分类号】：TP391.41

【相似文献】