Scholarlink

<p>Diabetic retinopathy most common preventable cause of vision impairment primarily affects people in their working years worldwide. The need for more accessible and effective methods of identifying managing diagnosing and treating retinal disorders has been brought to light by recent research. When designing a computer-aided diagnosis tool for diabetic retinopathy, it is crucial to evaluate both challenges of achieving a reliable early diagnosis at an affordable cost and effectiveness of screening programs. Biomedical engineers and computer scientists now have innovative ways to meet the demands of clinical practice thanks to developments in machine learning and deep learning techniques. The primary objective of this study is to develop a methodology for classifying the severity of diabetic retinopathy that integrates pre-processing, segmentation, and severity classification. For pre-processing, it is recommended to use median filtering, followed by segmentation with Gradient Descent—Sea Turtle Foraging Algorithm-based SwinUNet (GD-STFA-based SwinUNet). The diabetic retinopathy severity is then classified through the proposed ensemble learning using Xception,MoNet andPolyNet. Diabetic retinopathy cases are thus classified into five severity levels by proposed ensemble learning: No Diabetic Retinopathy, Mild Diabetic Retinopathy, Moderate Diabetic Retinopathy, Severe Diabetic Retinopathyand Proliferative Diabetic Retinopathy. Experimental outcomes are compared with standard models and the proposed ensemble learning (MoNetPolyNetXception) technique outperforms them with 98.9% accuracy, 99% sensitivity, 98.5% specificity, and 98.4% precision rate.</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

An Ensemble Learning Approach Using Deep Learning Models For Diabetic Retinopathy Severity Classification

  • 期刊论文

  • 发表日期：

    2025年7月11日

查看全文
  • A. S. Sabeena,
  • M. K. Jeyakumar

摘要

Diabetic retinopathy most common preventable cause of vision impairment primarily affects people in their working years worldwide. The need for more accessible and effective methods of identifying managing diagnosing and treating retinal disorders has been brought to light by recent research. When designing a computer-aided diagnosis tool for diabetic retinopathy, it is crucial to evaluate both challenges of achieving a reliable early diagnosis at an affordable cost and effectiveness of screening programs. Biomedical engineers and computer scientists now have innovative ways to meet the demands of clinical practice thanks to developments in machine learning and deep learning techniques. The primary objective of this study is to develop a methodology for classifying the severity of diabetic retinopathy that integrates pre-processing, segmentation, and severity classification. For pre-processing, it is recommended to use median filtering, followed by segmentation with Gradient Descent—Sea Turtle Foraging Algorithm-based SwinUNet (GD-STFA-based SwinUNet). The diabetic retinopathy severity is then classified through the proposed ensemble learning using Xception,MoNet andPolyNet. Diabetic retinopathy cases are thus classified into five severity levels by proposed ensemble learning: No Diabetic Retinopathy, Mild Diabetic Retinopathy, Moderate Diabetic Retinopathy, Severe Diabetic Retinopathyand Proliferative Diabetic Retinopathy. Experimental outcomes are compared with standard models and the proposed ensemble learning (MoNetPolyNetXception) technique outperforms them with 98.9% accuracy, 99% sensitivity, 98.5% specificity, and 98.4% precision rate.