Scholarlink

Diabetic retinopathy is an eye disease that is the leading cause of vision loss in one-third of people with diabetes, it progresses silently without any noticeable symptoms from the patient’s side, so the early detection can preserve the patient’s vision loss and improve the quality of life. In this paper, we present a combination of deep learning pre-trained Inception-ResNet model with state-of-the-art image cleaning, enhancement and pre-processing techniques for early classification of diabetic retinopathy by using about 3660 retinal images obtained from the APTOS 2019 blindness detection public dataset. The evaluation of the approach is made based on different measures such as accuracy, precision, recall F1-score, confusion matric, receiver operating characteristic (ROC) curve and kappa score, showed a promising result for accurately classifying the stage of diabetic retinopathy. Our model achieved an accuracy of 98.0%, an average F1 score of 97.6% and a kappa score of 0.970. The results suggest that the approach can contribute to the early detection of diabetic retinopathy and improve the quality of life of people with diabetes.

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

Transfer Learning-Based Classification of Diabetic Retinopathy Using a Pre-trained InceptionResNet Model

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Suhad Malayshi,
  • Ahmad Hasasneh

摘要

Diabetic retinopathy is an eye disease that is the leading cause of vision loss in one-third of people with diabetes, it progresses silently without any noticeable symptoms from the patient’s side, so the early detection can preserve the patient’s vision loss and improve the quality of life. In this paper, we present a combination of deep learning pre-trained Inception-ResNet model with state-of-the-art image cleaning, enhancement and pre-processing techniques for early classification of diabetic retinopathy by using about 3660 retinal images obtained from the APTOS 2019 blindness detection public dataset. The evaluation of the approach is made based on different measures such as accuracy, precision, recall F1-score, confusion matric, receiver operating characteristic (ROC) curve and kappa score, showed a promising result for accurately classifying the stage of diabetic retinopathy. Our model achieved an accuracy of 98.0%, an average F1 score of 97.6% and a kappa score of 0.970. The results suggest that the approach can contribute to the early detection of diabetic retinopathy and improve the quality of life of people with diabetes.