Scholarlink

Diabetic Retinopathy (DR) is an eye disease associated with chronic diabetes. It remains the primary cause of visual impairment and blindness among the global working-age population. Early detection of DR is crucial for ensuring timely diagnosis and effective treatment. This paper proposes a new homogeneous ensemble-based approach constructed using a set of hybrid architectures as base learners and two combination rules (weighted and hard voting) for referable DR detection, using fundus images from the Messidor-2, Kaggle DR, and APTOS datasets. The hybrid architectures are created using deep feature extraction techniques, dimensionality reduction techniques to reduce the size of the extracted features, and a decision tree algorithm (DT) for classification. The results showed the potential of the proposed new approach which achieved high accuracy values over the three datasets: 90.65%, 93.01%, and 83.32% using the APTOS, Kaggle DR, and Messidor-2 datasets respectively. Therefore, we recommend using the proposed approach since it is impactful for referable DR classification, and it represents a promising tool to assist ophthalmologists in diagnosing DR.

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

Feature Extraction and Dimensionality Reduction to Evaluate Decision Tree Ensembles for Diabetic Retinopathy Detection

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Chaymaa Lahmar,
  • Ali Idri

摘要

Diabetic Retinopathy (DR) is an eye disease associated with chronic diabetes. It remains the primary cause of visual impairment and blindness among the global working-age population. Early detection of DR is crucial for ensuring timely diagnosis and effective treatment. This paper proposes a new homogeneous ensemble-based approach constructed using a set of hybrid architectures as base learners and two combination rules (weighted and hard voting) for referable DR detection, using fundus images from the Messidor-2, Kaggle DR, and APTOS datasets. The hybrid architectures are created using deep feature extraction techniques, dimensionality reduction techniques to reduce the size of the extracted features, and a decision tree algorithm (DT) for classification. The results showed the potential of the proposed new approach which achieved high accuracy values over the three datasets: 90.65%, 93.01%, and 83.32% using the APTOS, Kaggle DR, and Messidor-2 datasets respectively. Therefore, we recommend using the proposed approach since it is impactful for referable DR classification, and it represents a promising tool to assist ophthalmologists in diagnosing DR.