Scholarlink

Background <p>Diabetic retinopathy (DR) is a condition caused by long-term diabetes and clots the retina's blood vessels. Early recognition and categorization of DR are vital for timely diagnosis and preventing vision loss.</p> Objective <p>This paper presents a novel DR-Cinque for detecting and classifying DR in cinque classes from retinal OCT images.</p> Methods <p> Initially, the retinal images are collected from three various publicly available databases and the noisy artifacts are removed by adaptive mean filter. The Otsu segmentation algorithm is applied for segmenting soft and hard exudates from the noise-free retinal images. The proposed DR-Cinque leverages Regularized Network (RegNet) to automatically retrieve appropriate features from the segmented regions. Finally, Artificial Neural Network (ANN) is employed to classify the images into various phases of DR severity ranging from no DR to proliferative DR.</p> Results <p>The proposed DR-Cinque achieves high accuracy and robustness for outperforming traditional methods in both detection and classification tasks. The efficiency of the proposed DR-Cinque was assessed using the network parameters viz., accuracy, F1 score, sensitivity, accuracy, and specificity.</p> Conclusions <p>The proposed DR-Cinque framework achieves 99.7% accuracy, which is higher than traditional deep learning networks. The proposed DR-Cinque enhances overall accuracy by 2.78%, 0.50%, 3.85%, and 2.46% compared to DRNet13, DR-UNet, Random Forest Classifier, and DenseNet 121 respectively. Experimental fallouts on standard DR datasets demonstrate the efficacy of the proposed DR-Cinque for making it a capable tool for automated screening in medical settings.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

DR-Cinque: diabetic retinopathy cinque classification via deep dual segmentation-based neural network

  • 期刊论文

  • 发表日期：

    2026年5月8日

查看全文
  • Karpagavadivu Karuppusamy,
  • Baranidharan Thangavelu,
  • Kavitha Mettupalayam Subramaniam,
  • Sumathi Thangavelu

摘要

Background

Diabetic retinopathy (DR) is a condition caused by long-term diabetes and clots the retina's blood vessels. Early recognition and categorization of DR are vital for timely diagnosis and preventing vision loss.

Objective

This paper presents a novel DR-Cinque for detecting and classifying DR in cinque classes from retinal OCT images.

Methods

Initially, the retinal images are collected from three various publicly available databases and the noisy artifacts are removed by adaptive mean filter. The Otsu segmentation algorithm is applied for segmenting soft and hard exudates from the noise-free retinal images. The proposed DR-Cinque leverages Regularized Network (RegNet) to automatically retrieve appropriate features from the segmented regions. Finally, Artificial Neural Network (ANN) is employed to classify the images into various phases of DR severity ranging from no DR to proliferative DR.

Results

The proposed DR-Cinque achieves high accuracy and robustness for outperforming traditional methods in both detection and classification tasks. The efficiency of the proposed DR-Cinque was assessed using the network parameters viz., accuracy, F1 score, sensitivity, accuracy, and specificity.

Conclusions

The proposed DR-Cinque framework achieves 99.7% accuracy, which is higher than traditional deep learning networks. The proposed DR-Cinque enhances overall accuracy by 2.78%, 0.50%, 3.85%, and 2.46% compared to DRNet13, DR-UNet, Random Forest Classifier, and DenseNet 121 respectively. Experimental fallouts on standard DR datasets demonstrate the efficacy of the proposed DR-Cinque for making it a capable tool for automated screening in medical settings.