Scholarlink

<p>Diabetic Retinopathy (DR) is a severe complication of diabetes that can lead to vision impairment and irreversible blindness if not diagnosed and treated at an early stage. In its initial phases, DR is often asymptomatic, making early detection particularly challenging. Manual examination of retinal images is labor-intensive and error-prone due to the subtle and varied nature of retinal lesions. As a result, accurately classifying the different stages of DR remains a major challenge. To address this issue, we propose a deep learning-based approach utilizing eight pre-trained convolutional neural network (CNN) models: VGG16, VGG19, Xception, InceptionV3, ResNet50, InceptionResNetV2, MobileNetV2, and EfficientNetB0. We evaluated these models on two publicly available datasets: Dataset A (APTOS 2019) and Dataset B (Messidor-EyePACS). Among the tested models, ResNet50 achieved the highest accuracy of 86% on the APTOS 2019 dataset, surpassing existing state-of-the-art methods in DR classification. To improve model transparency and support clinical trust, we integrated SHAP (Shapley Additive Explanations) for visual interpretability. This technique highlighted the most relevant image regions used by the models in making their predictions, offering valuable insights into the decision-making process. Our findings demonstrate that combining high-performing deep learning models with explainability tools can significantly enhance the accuracy, transparency, and clinical applicability of automated DR screening systems.</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

Interpretable transfer learning framework for diabetic retinopathy stage classification

  • 期刊论文

  • 发表日期：

    2025年9月22日

查看全文
  • Dalel Bouabidi,
  • Amira Echtioui

摘要

Diabetic Retinopathy (DR) is a severe complication of diabetes that can lead to vision impairment and irreversible blindness if not diagnosed and treated at an early stage. In its initial phases, DR is often asymptomatic, making early detection particularly challenging. Manual examination of retinal images is labor-intensive and error-prone due to the subtle and varied nature of retinal lesions. As a result, accurately classifying the different stages of DR remains a major challenge. To address this issue, we propose a deep learning-based approach utilizing eight pre-trained convolutional neural network (CNN) models: VGG16, VGG19, Xception, InceptionV3, ResNet50, InceptionResNetV2, MobileNetV2, and EfficientNetB0. We evaluated these models on two publicly available datasets: Dataset A (APTOS 2019) and Dataset B (Messidor-EyePACS). Among the tested models, ResNet50 achieved the highest accuracy of 86% on the APTOS 2019 dataset, surpassing existing state-of-the-art methods in DR classification. To improve model transparency and support clinical trust, we integrated SHAP (Shapley Additive Explanations) for visual interpretability. This technique highlighted the most relevant image regions used by the models in making their predictions, offering valuable insights into the decision-making process. Our findings demonstrate that combining high-performing deep learning models with explainability tools can significantly enhance the accuracy, transparency, and clinical applicability of automated DR screening systems.