Scholarlink

<p>Accurate classification of ocular diseases from color fundus images is an important task in medical image analysis. To improve the performance of multi-class ocular disease classification, we propose a hybrid deep learning dual-branch architecture that combines deep features extracted from two convolutional neural networks, namely ResNet50 and XceptionNet, each enhanced with Convolutional Block Attention Modules (CBAM) to refine spatial and channel representations. CBAM adaptively emphasizes informative features while suppressing less relevant ones, thereby improving the discriminative capacity of the model. The extracted features are subsequently flattened and concatenated to leverage the complementary strengths of residual learning and depthwise separable convolutions, followed by fully connected layers for classification. To enhance classification performance, a composite dataset was constructed by integrating selected classes from two publicly available fundus image datasets. The final dataset consists of 10 classes, including nine ocular diseases and healthy cases. Experimental results demonstrate that the proposed model achieves an overall accuracy of 96.05% and a weighted average F1-score of 96.05%. Furthermore, Grad-CAM was employed to provide visual interpretability of the model’s predictions.</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

Towards enhanced ophthalmic care: integrating ameliorated ResNet50 and XceptionNet deep features utilizing fundus images

  • 期刊论文

  • 发表日期：

    2026年6月23日

查看全文
  • Aya Mostafa,
  • Marwa Elpeltagy,
  • Mohamed A. Abdelhamed,
  • Aly M. Elsemary

摘要

Accurate classification of ocular diseases from color fundus images is an important task in medical image analysis. To improve the performance of multi-class ocular disease classification, we propose a hybrid deep learning dual-branch architecture that combines deep features extracted from two convolutional neural networks, namely ResNet50 and XceptionNet, each enhanced with Convolutional Block Attention Modules (CBAM) to refine spatial and channel representations. CBAM adaptively emphasizes informative features while suppressing less relevant ones, thereby improving the discriminative capacity of the model. The extracted features are subsequently flattened and concatenated to leverage the complementary strengths of residual learning and depthwise separable convolutions, followed by fully connected layers for classification. To enhance classification performance, a composite dataset was constructed by integrating selected classes from two publicly available fundus image datasets. The final dataset consists of 10 classes, including nine ocular diseases and healthy cases. Experimental results demonstrate that the proposed model achieves an overall accuracy of 96.05% and a weighted average F1-score of 96.05%. Furthermore, Grad-CAM was employed to provide visual interpretability of the model’s predictions.