Scholarlink

One of the main causes of vision loss among diabetics is Diabetic Retinopathy (DR). Early diagnosis and accurate treatment of this condition are crucial for providing timely treatment and preventing severe complications. Several deep learning models, like Inception, ResNet, and denseNet, have demonstrated their ability to accurately detect this condition, but they still need to be refined for use in the Indian diabetic population. The goal of this study is to analyze the various techniques that are used to improve the performance of the deep learning models in detecting DR in Indian patients. We used a large set of retinal images from the Indian diabetic population to perform a comprehensive evaluation. To optimize the model's performance, we first experimented with varying parameters, such as the learning rate, batch size, and data augmentation. We then gradually adjusted the models to improve their sensitivity, specificity, area under the AUC-ROC, and accuracy. The results of the study revealed that the various strategies used to improve the performance of the deep learning models in detecting DR significantly improved the accuracy of the models. The ResNet model was able to perform remarkably well with 98.4% accuracy, while the other models, such as the Inception and the denseNet, were able to achieve 96.4 and 92.3% accuracy, respectively. In addition, we discuss the impact of various fine-tuning parameters on the performance of the models. The learning rate of the ResNet model was lowered to 0.0001, and it was also enhanced through data augmentation methods such as flipping, random rotations, and translations. Transfer learning performed well across different architectures. The study demonstrates the ResNet model's capability to accurately detect DR in Indian diabetics. The findings of this research provide valuable information on the various fine-tuning techniques that can be utilized to improve the efficiency of deep learning models in this field. Their early detection and accurate diagnosis could help in the timely delivery of corrective therapy and improve patient outcomes.

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

Optimizing Fine-Tuning Strategies for Diabetic Retinopathy Detection: A Comparative Evaluation of ResNet, Inception, and DenseNet

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Komal B. Umare,
  • T. Kishore Kumar

摘要

One of the main causes of vision loss among diabetics is Diabetic Retinopathy (DR). Early diagnosis and accurate treatment of this condition are crucial for providing timely treatment and preventing severe complications. Several deep learning models, like Inception, ResNet, and denseNet, have demonstrated their ability to accurately detect this condition, but they still need to be refined for use in the Indian diabetic population. The goal of this study is to analyze the various techniques that are used to improve the performance of the deep learning models in detecting DR in Indian patients. We used a large set of retinal images from the Indian diabetic population to perform a comprehensive evaluation. To optimize the model's performance, we first experimented with varying parameters, such as the learning rate, batch size, and data augmentation. We then gradually adjusted the models to improve their sensitivity, specificity, area under the AUC-ROC, and accuracy. The results of the study revealed that the various strategies used to improve the performance of the deep learning models in detecting DR significantly improved the accuracy of the models. The ResNet model was able to perform remarkably well with 98.4% accuracy, while the other models, such as the Inception and the denseNet, were able to achieve 96.4 and 92.3% accuracy, respectively. In addition, we discuss the impact of various fine-tuning parameters on the performance of the models. The learning rate of the ResNet model was lowered to 0.0001, and it was also enhanced through data augmentation methods such as flipping, random rotations, and translations. Transfer learning performed well across different architectures. The study demonstrates the ResNet model's capability to accurately detect DR in Indian diabetics. The findings of this research provide valuable information on the various fine-tuning techniques that can be utilized to improve the efficiency of deep learning models in this field. Their early detection and accurate diagnosis could help in the timely delivery of corrective therapy and improve patient outcomes.