Scholarlink

Diabetic retinopathy (DR) is a major cause of vision loss worldwide, emphasizing the need for timely and precise diagnosis. This study introduces a convolutional neural network (CNN) method utilizing EfficientNetB0 as a base network, for classifying DR stages: No DR, Mild DR, Moderate DR, Severe DR and Proliferative DR using fundus images. The model was trained and validated on the APTOS 2019 dataset of retinal images representing various stages of DR, with preprocessing steps applied to enhance critical features. The CNN achieved strong classification performance by utilizing an 80-20 train-test split. Results indicate that the model effectively distinguishes between different stages of DR, providing a valuable tool for automating early detection and aiding ophthalmologists in clinical diagnosis, ultimately reducing diagnostic time and improving patient outcomes. The model achieved 93.5% accuracy, with a specificity of 90.2% and a sensitivity of 91.7%, indicating balanced performance in identifying DR stages while minimizing misclassifications.

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

Classification of Diabetic Retinopathy Anomalies in Fundus Images Using Convolutional Neural Networks

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Amine El Hossi,
  • Abdelali Elmoufidi,
  • Mourad Nachaoui

摘要

Diabetic retinopathy (DR) is a major cause of vision loss worldwide, emphasizing the need for timely and precise diagnosis. This study introduces a convolutional neural network (CNN) method utilizing EfficientNetB0 as a base network, for classifying DR stages: No DR, Mild DR, Moderate DR, Severe DR and Proliferative DR using fundus images. The model was trained and validated on the APTOS 2019 dataset of retinal images representing various stages of DR, with preprocessing steps applied to enhance critical features. The CNN achieved strong classification performance by utilizing an 80-20 train-test split. Results indicate that the model effectively distinguishes between different stages of DR, providing a valuable tool for automating early detection and aiding ophthalmologists in clinical diagnosis, ultimately reducing diagnostic time and improving patient outcomes. The model achieved 93.5% accuracy, with a specificity of 90.2% and a sensitivity of 91.7%, indicating balanced performance in identifying DR stages while minimizing misclassifications.