Scholarlink

Diabetic retinopathy (DR) is the leading global cause of adult blindness, and avoiding vision loss requires early identification. To help improve the detection of diabetic retinopathy, this work combines Explainable AI (XAI) techniques with collaborative Vision Transformer (ViT) and Convolutional Neural Network (CNN) architectures. Significantly using LIME (Local Interpretable Model-agnostic Explanations), to ensure the interpretability of a model. Five of these advanced CNN models which used to extract important characteristics of retinal images include VGG16, EfficientNetB0, MobileNetV2, NASNetMobile and Xception. The models are shown to perform well when tested on a large dataset of diabetic retinopathy, with good classification accuracy and ability to identify important characteristics that help diagnosis. The study integrates LIME to provide insights into each model’s decision making process, which adds to its transparency and reliability on prediction results. This collaborative framework aims to improve the way medical professionals decide when to identify diabetic retinopathy, by helping medical professionals make better decisions.

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

Advancing Diabetic Retinopathy Detection Through Collaborative Vision Transformer and CNN Architectures Integrated with Explainable AI

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • K. Afnaan,
  • Tripty Singh,
  • Kriti Priya shah,
  • Achal Baniya,
  • Mansi Sharma

摘要

Diabetic retinopathy (DR) is the leading global cause of adult blindness, and avoiding vision loss requires early identification. To help improve the detection of diabetic retinopathy, this work combines Explainable AI (XAI) techniques with collaborative Vision Transformer (ViT) and Convolutional Neural Network (CNN) architectures. Significantly using LIME (Local Interpretable Model-agnostic Explanations), to ensure the interpretability of a model. Five of these advanced CNN models which used to extract important characteristics of retinal images include VGG16, EfficientNetB0, MobileNetV2, NASNetMobile and Xception. The models are shown to perform well when tested on a large dataset of diabetic retinopathy, with good classification accuracy and ability to identify important characteristics that help diagnosis. The study integrates LIME to provide insights into each model’s decision making process, which adds to its transparency and reliability on prediction results. This collaborative framework aims to improve the way medical professionals decide when to identify diabetic retinopathy, by helping medical professionals make better decisions.