基于特征增强的高血压视网膜病变分类方法研究

Hypertensive retinopathy classification method based on feature enhancement mechanism research

高血压视网膜病变是由高血压所引起的眼底疾病,传统分类方法主要是基于区域特征进行分析,识别依据较为单一,准确度不高。为提高分类准确度,本文提出了一种基于特征增强机制的高血压视网膜病变分类方法,设计了基于不同图像色彩空间的眼底图像特征增强方法,增强眼底照片中的病灶特征,提高模型的输入特征值,将处理后的图片输入改进的DenseNet模型中进行分类,从而提高高血压视网膜病变(HR)分类的准确度。采用公开数据集OIA-ODIR对本文提出的基于特征增强的高血压视网膜病变分类方法进行测试,其敏感性、特异性、准确率分别达到97.09%、98.79%、98.67%,与现有的HR分类方法进行分析对比,本文提出的分类方法效果更佳。

Hypertensive retinopathy is a fundus disease caused by high blood pressure,the traditional classification method is mainly based on the regional features to analyze,the identification basis is relatively single,and the accuracy is not high.In order to improve the classification accuracy,this paper proposes a classification method of hypertensive retinopathy based on the feature enhancement mechanism and designs a fundus image feature enhancement method based on different image color spaces,i.e.,to enhance the lesion features in the fundus photographs,to increase the input feature value of the model,and to input the processed images into the improved DenseNet model for classification,so as to achieve the accuracy of the classification of hypertensive retinopathy(HR).lesion(HR)classification accuracy.The sensitivity,specificity,and accuracy of the proposed feature-enhanced classification method for hypertensive retinopathy(HR)were tested using the publicly available dataset OIA-ODIR and reached 97.09%,98.79%,and 98.67%.Compared with the existing HR classification method,the classification method proposed in this article is better.