...
  • 主题
高级搜索  >
历史搜索 清空历史
首页> 外文期刊>Journal of ambient intelligence and humanized computing >Robust retinal blood vessel segmentation using convolutional neural network and support vector machine
【24h】

Robust retinal blood vessel segmentation using convolutional neural network and support vector machine

机译：使用卷积神经网络和支持向量机的强大视网膜血管分割

获取原文
获取原文并翻译 | 示例
           
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

In recent decades, automatic retinal blood vessel segmentation and classification (RBVSC) helps to determine many diseases such as glaucoma, hypertension, macular-degeneration, diabetes-mellitus, etc. The early recognition of these disorders is essential for preventing patients from blindness. In this work, a new supervised system was developed to enhance the performance of RBVSC. At first, the input retinal images were collected from two datasets such as: Digital Retinal Image for Vessel Extraction (DRIVE) and STARE (STructured Analysis of the Retina). Then, the retinal vessels were segmented utilizing mean orientation based super-pixel segmentation. Besides, Convolutional Neural Network (CNN) was applied to extract the feature vectors from segmented regions. Finally, a binary classifier [Support Vector Machine (SVM)] performs classification on the extracted features for classifying the "vessel" and "non-vessel" regions. The combination of CNN and SVM automatically learns the feature values from raw images and classifies the patterns easily. From the experimental study, the proposed system improved RBVSC up to 2-4% compared to other existing systems and classification methodologies: Deep Neural Network (DNN), Random Forest (RF) and Naive Bayes (NB) by means of specificity, accuracy, sensitivity and kappa index.
机译：近几十年来，自动视网膜血管分割和分类（RBVSC）有助于确定许多疾病，如青光眼，高血压，黄斑，糖尿病等。对这些疾病的早期识别对于预防盲目的患者至关重要。在这项工作中，开发了一个新的监督系统，以提高RBVSC的表现。首先，从两个数据集收集输入视网膜图像，例如：用于血管提取（驱动）和凝视（视网膜的结构化分析）的数字视网膜图像。然后，使用基于平均基于的超像素分割来分段视网膜血管。此外，应用卷积神经网络（CNN）以从分段区域提取特征向量。最后，二进制分类器[支持向量机（SVM）]对提取的特征进行分类，用于对“船只”和“非船只”区域进行分类。 CNN和SVM的组合自动从原始图像中学习特征值，并轻松对模式进行分类。从实验研究中，与其他现有系统和分类方法相比，提出的系统改善了RBVSC高达2-4％：深神经网络（DNN），随机森林（RF）和幼稚贝叶斯（NB）通过特异性，准确性，敏感性和κ指数。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号