Scholarlink

<p>Age-related macular degeneration (AMD) is a leading cause of irreversible vision loss among the elderly population worldwide, with a significant impact on quality of life. Early and accurate diagnosis of AMD stages is critical for timely intervention. However, the subtle and heterogeneous nature of AMD manifestations makes its diagnosis challenging, necessitating advanced imaging modalities and automated diagnostic tools. We propose a context-aware AMD stage diagnosis method using a self-supervised U-Net and Graph Neural Networks (GNNs) on OCT imaging. Our methodology integrates a U-Net architecture trained on a pretext inpainting task (PT model) for super-pixel segmentation, a pre-trained encoder module of the PT model for robust feature extraction, and a fusion of GNNs and dense layers for classification. A pioneering context-aware loss function encapsulates spatial and structural interdependencies among super-pixels, enhancing diagnostic precision through the exploitation of local and global contextual information. The proposed framework demonstrates strong performance in diagnosing AMD stages from OCT images. Specifically, our model achieved an average accuracy of 94.5% and an <i>F</i>1-score of 0.93, outperforming state-of-the-art methods such as traditional CNNs (accuracy of 88.2%, <i>F</i>1-score of 0.86) and ResNet-based approaches (accuracy of 91.1%, <i>F</i>1-score of 0.90) (<i>p</i> &lt; 0.05). The consistent decrease in loss values, high AUC scores (0.96), and accurate classification rates in the confusion matrix collectively underscore the model's reliability and effectiveness. By leveraging self-supervised pre-training to overcome limited annotated data and combining spatial relational modeling (via GCN) with dense layers for discriminative feature refinement, the framework effectively captures both local and global patterns critical for AMD staging. These results highlight the potential of the proposed approach to serve as a reliable tool for early and precise AMD diagnosis, enabling timely clinical decision-making.</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

Context-aware diagnosis of age-related macular degeneration (AMD) stages: a unified approach using self-supervised U-Net and graph neural networks on OCT imaging

  • 期刊论文

  • 发表日期：

    2025年6月7日

查看全文
  • Toktam Khatibi

摘要

Age-related macular degeneration (AMD) is a leading cause of irreversible vision loss among the elderly population worldwide, with a significant impact on quality of life. Early and accurate diagnosis of AMD stages is critical for timely intervention. However, the subtle and heterogeneous nature of AMD manifestations makes its diagnosis challenging, necessitating advanced imaging modalities and automated diagnostic tools. We propose a context-aware AMD stage diagnosis method using a self-supervised U-Net and Graph Neural Networks (GNNs) on OCT imaging. Our methodology integrates a U-Net architecture trained on a pretext inpainting task (PT model) for super-pixel segmentation, a pre-trained encoder module of the PT model for robust feature extraction, and a fusion of GNNs and dense layers for classification. A pioneering context-aware loss function encapsulates spatial and structural interdependencies among super-pixels, enhancing diagnostic precision through the exploitation of local and global contextual information. The proposed framework demonstrates strong performance in diagnosing AMD stages from OCT images. Specifically, our model achieved an average accuracy of 94.5% and an F1-score of 0.93, outperforming state-of-the-art methods such as traditional CNNs (accuracy of 88.2%, F1-score of 0.86) and ResNet-based approaches (accuracy of 91.1%, F1-score of 0.90) (p < 0.05). The consistent decrease in loss values, high AUC scores (0.96), and accurate classification rates in the confusion matrix collectively underscore the model's reliability and effectiveness. By leveraging self-supervised pre-training to overcome limited annotated data and combining spatial relational modeling (via GCN) with dense layers for discriminative feature refinement, the framework effectively captures both local and global patterns critical for AMD staging. These results highlight the potential of the proposed approach to serve as a reliable tool for early and precise AMD diagnosis, enabling timely clinical decision-making.