Scholarlink

<p>One of the major causes of visual impairment and blindness in the world among patients affected with diabetes is described as Diabetic Retinopathy (DR). It now needs to be combined with early diagnosis; however, manual screening is time-consuming, incomplete, and constrained by the availability of retina specialists, particularly in low-resource settings. The results of this review provide an overview analysis of the deep learning approaches of automated DR segmentation and classification on retinal fundus images, both traditional machine learning frameworks and the emerging models, modern architectures such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), transformer-based models, attention layer, and ensemble models. This paper emphasizes the significance of preprocessing techniques, including CLAHE, green channel extraction, and image normalization, in enhancing lesion visibility and improving the quality of inputs for training the model. Nevertheless, the complexities lie in managing the imbalance problem in the dataset, differences in lesion types, the lack of generalizability, limitations of interpretation, and the high computational costs. Recent advances, including Federated Learning, Multimodal Data Fusion, lightweight models (such as EfficientNet-Lite), and Explainable AI, offer potential for improvement. The focus of future studies should be on a privacy-preserving, interpretable, and real-time DR detection system, which is studied and tested in a clinical trial.</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

A Comprehensive Review of Deep Learning Approaches for Automated Detection, Segmentation, and Grading of Diabetic Retinopathy

  • 期刊论文

  • 发表日期：

    2025年12月2日

查看全文
  • Varsha Bhoyar,
  • Mitul Patel

摘要

One of the major causes of visual impairment and blindness in the world among patients affected with diabetes is described as Diabetic Retinopathy (DR). It now needs to be combined with early diagnosis; however, manual screening is time-consuming, incomplete, and constrained by the availability of retina specialists, particularly in low-resource settings. The results of this review provide an overview analysis of the deep learning approaches of automated DR segmentation and classification on retinal fundus images, both traditional machine learning frameworks and the emerging models, modern architectures such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), transformer-based models, attention layer, and ensemble models. This paper emphasizes the significance of preprocessing techniques, including CLAHE, green channel extraction, and image normalization, in enhancing lesion visibility and improving the quality of inputs for training the model. Nevertheless, the complexities lie in managing the imbalance problem in the dataset, differences in lesion types, the lack of generalizability, limitations of interpretation, and the high computational costs. Recent advances, including Federated Learning, Multimodal Data Fusion, lightweight models (such as EfficientNet-Lite), and Explainable AI, offer potential for improvement. The focus of future studies should be on a privacy-preserving, interpretable, and real-time DR detection system, which is studied and tested in a clinical trial.