Scholarlink

<p>This study presents an integrated approach for retinal vessel segmentation and classification of arteries and veins using the ResU-Net architecture combined with a Multi-Scale Feature Fusion Module. The ResU-Net framework effectively captures both local and global features through multi-scale convolutional filters, allowing for the precise segmentation of vascular structures. By employing various convolutional kernels, the model identifies vessel details at multiple scales, ensuring accurate delineation of arteries and veins. The Multi-Scale Feature Fusion Module enhances this capability by optimizing local features and integrating outputs from different convolutional layers, facilitating a comprehensive representation of vessel characteristics. The Snippeting Features process further refines the output by prioritizing relevant details for improved classification accuracy. This robust framework demonstrates the efficacy of combining feature extraction, fusion, and classification techniques in medical imaging, leading to enhanced segmentation and identification of retinal vascular structures.The proposed model demonstrates exceptional performance across multiple datasets, achieving classification accuracies of 96.37–99.69% on the LES Retina dataset, with segmentation accuracy peaking at 0.9874. For the TREND dataset, accuracy values range from 0.9686 to 0.9986, while recall scores consistently reach near-perfect values, indicating the model’s reliability in both segmentation and classification tasks. Across three datasets, the model exhibits strong performance metrics, with accuracy scores mostly exceeding 0.97, highlighting its effectiveness in retinal vessel analysis.</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

Deep learning-based multi-scale framework for retinal artery and vein classification

  • 期刊论文

  • 发表日期：

    2025年7月16日

查看全文
  • Usharani Bhimavarapu

摘要

This study presents an integrated approach for retinal vessel segmentation and classification of arteries and veins using the ResU-Net architecture combined with a Multi-Scale Feature Fusion Module. The ResU-Net framework effectively captures both local and global features through multi-scale convolutional filters, allowing for the precise segmentation of vascular structures. By employing various convolutional kernels, the model identifies vessel details at multiple scales, ensuring accurate delineation of arteries and veins. The Multi-Scale Feature Fusion Module enhances this capability by optimizing local features and integrating outputs from different convolutional layers, facilitating a comprehensive representation of vessel characteristics. The Snippeting Features process further refines the output by prioritizing relevant details for improved classification accuracy. This robust framework demonstrates the efficacy of combining feature extraction, fusion, and classification techniques in medical imaging, leading to enhanced segmentation and identification of retinal vascular structures.The proposed model demonstrates exceptional performance across multiple datasets, achieving classification accuracies of 96.37–99.69% on the LES Retina dataset, with segmentation accuracy peaking at 0.9874. For the TREND dataset, accuracy values range from 0.9686 to 0.9986, while recall scores consistently reach near-perfect values, indicating the model’s reliability in both segmentation and classification tasks. Across three datasets, the model exhibits strong performance metrics, with accuracy scores mostly exceeding 0.97, highlighting its effectiveness in retinal vessel analysis.