Computer Engineering and Applications ›› 2022, Vol. 58 ›› Issue (8): 230-236.DOI: 10.3778/j.issn.1002-8331.2010-0427

• Graphics and Image Processing • Previous Articles     Next Articles

Classification Method of Hypertensive Retinopathy Based on Regional Feature Fusion

WANG Wei, PU Yiwen   

  1. 1.Foundation Department, Liaoning Technical University, Huludao, Liaoning 125105, China
    2.School of Electronic and Information Engineering, Liaoning Technical University, Huludao, Liaoning 125105, China
  • Online:2022-04-15 Published:2022-04-15



  1. 1.辽宁工程技术大学 基础教学部,辽宁 葫芦岛 125105
    2.辽宁工程技术大学 电子与信息工程学院,辽宁 葫芦岛 125105

Abstract: Due to the lack of distinctive clinical features of hypertensive retinopathy(HR), it is difficult to effectively classify using conventional methods. Targeting on this problem, it proposes a regional feature fusion HR classification method with global and local characteristic, that is, based on the global model, merging its local characteristic arteriovenous nicking(AVN) classification model to enhance the HR classification effect. On AVN classification, it proposes a novel method to detect arteriovenous intersections. The proposed algorithm calculates the locations of these intersections through logical computing, AVN images are then extracted in regions of interest from HR affected rear of an eye photographs. Tested the proposed merging model with private databases, the accuracy, sensitivity and specificity are 93.5%, 69.83% and 98.33%, respectively. The experimental results show that this new model works well and gets better effect compared with the existing methods in the single-stage classification model.

Key words: regional feature fusion, hypertensive retinopathy classification, arteriovenous nicking classification, cross point detection

摘要: 由于高血压性视网膜病变(hypertensive retinopathy,HR)病灶特征不明显,传统分类算法难以对其进行有效分类。针对这一问题,提出一种具有整体特征和局部特征的区域特征融合HR分类方法,即在整体HR分类模型的基础上,融合局部特征动静脉交叉压迫(arteriovenous nicking,AVN)分类模型来增强HR分类效果。在AVN分类中,提出一种新型的交叉点检测算法,该算法对分类后的动静脉进行逻辑与运算以求出交叉点位置,利用感兴趣区域提取方法从HR眼底图像中提取AVN图像块。提出的融合模型在私有数据集上进行了评估,准确率、敏感性和特异性分别为93.50%、69.83%和98.33%。在单阶段分类模型中分别与已有的方法进行对比,实验结果证明所提出模型效果较好。

关键词: 区域特征融合, 高血压性视网膜病变分类, 动静脉交叉压迫分类, 交叉点检测