Scholarlink

Deep learning algorithms have transformed medical imaging techniques in recent years, resulting in notable progress in categorizing multiple eye disorders. Early detection and assessments for multi-eye disease can increase the possibility of visual results. However, it takes a lot of effort and time to manually examine for eye disorders. This study investigates diverse deep learning techniques such as Convolutional Neural Network (CNN), AlexNet, VGG 16, Residual Network (ResNet), Inception v3, DarkNet, UNet, LinkNet, and EfficientNet B7 to analyze and characterize eye disease on collected datasets. The process begins with the effectiveness of various datasets and pre-processing methods to standardize and enhance image quality. After pre-processing, to evaluate the robustness and classification accuracy of several pre-trained architectures, extensive testing is examined for precise multiple eye disorders diagnosis. The rigorous evaluation of these models revealed that advanced architectures like EfficientNet B7 and DarkNet are the most effective, and always display better precision, accuracy, sensitivity, F1-scores, and specificity for eye disease classifications. The outcomes unambiguously show that, in the DR, Myopia, AMD, Glaucoma, and cataract detection tasks, the EfficientNet B7 performs noticeably better than any other models, demonstrating the greatest accuracy of (0.95–0.97) and outstanding performance in the remaining metrics. On the other hand, DarkNet 53 and UNet show outstanding accuracy (0.93–0.96) across all three categories of ocular illnesses, respectively. However, ResNet and UNet, while providing good results, lack interpretability, which is crucial for medical professionals to optimism and use these deep learning models effectively detection and classification of five datasets. In addition to LinkNet 34 and DarkNet 53, these models also exhibit exceptional performance across many criteria, indicating how effective they are at classifying eye disorders. Future research should address issues to enable the therapeutic use of these models, despite their potential to enhance diagnosis accuracy and 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

Analysis and Characterization of Multi-eye Diseases Using Deep Learning Algorithms

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Antara Malakar,
  • Ankur Ganguly,
  • Swarnendu Kumar Chakraborty

摘要

Deep learning algorithms have transformed medical imaging techniques in recent years, resulting in notable progress in categorizing multiple eye disorders. Early detection and assessments for multi-eye disease can increase the possibility of visual results. However, it takes a lot of effort and time to manually examine for eye disorders. This study investigates diverse deep learning techniques such as Convolutional Neural Network (CNN), AlexNet, VGG 16, Residual Network (ResNet), Inception v3, DarkNet, UNet, LinkNet, and EfficientNet B7 to analyze and characterize eye disease on collected datasets. The process begins with the effectiveness of various datasets and pre-processing methods to standardize and enhance image quality. After pre-processing, to evaluate the robustness and classification accuracy of several pre-trained architectures, extensive testing is examined for precise multiple eye disorders diagnosis. The rigorous evaluation of these models revealed that advanced architectures like EfficientNet B7 and DarkNet are the most effective, and always display better precision, accuracy, sensitivity, F1-scores, and specificity for eye disease classifications. The outcomes unambiguously show that, in the DR, Myopia, AMD, Glaucoma, and cataract detection tasks, the EfficientNet B7 performs noticeably better than any other models, demonstrating the greatest accuracy of (0.95–0.97) and outstanding performance in the remaining metrics. On the other hand, DarkNet 53 and UNet show outstanding accuracy (0.93–0.96) across all three categories of ocular illnesses, respectively. However, ResNet and UNet, while providing good results, lack interpretability, which is crucial for medical professionals to optimism and use these deep learning models effectively detection and classification of five datasets. In addition to LinkNet 34 and DarkNet 53, these models also exhibit exceptional performance across many criteria, indicating how effective they are at classifying eye disorders. Future research should address issues to enable the therapeutic use of these models, despite their potential to enhance diagnosis accuracy and patient outcomes.