Scholarlink

<p>Pulmonary nodules are one of the main causes of lung cancer. Nodule detection is therefore key to the prevention of lung cancer-related deaths. In this work, a new paradigm is therefore presented that revolves around supervised and unsupervised machine learning approaches, superpixels, and graph neural networks to detect and segment nodules automatically. The proposed method, namely, Enhanced Graph Propagated Detector of nodules, harbors an enhanced uniform superpixel generation model and couples it with an improved version of graph propagation networks to diagnose nodules accurately in computed tomography images without the need for a high amount of annotated data. The architecture presented in this work is light weight and shows high capability in nodule detection and segmentation compared to several state-of-the-art models in the field. Experimental results show a significant reduction in superpixel generation time, a true positive rate of approximately 95%, and a low number of false positives on the Lung Image Database Consortium image collection dataset.</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

EGPD-Nodule: An enhanced lung nodule detection and segmentation framework utilizing superpixels and graph neural networks

  • 期刊论文

  • 发表日期：

    2026年6月2日

查看全文
  • Sudipta Modak,
  • Luis Rueda,
  • Esam Abdel-Raheem

摘要

Pulmonary nodules are one of the main causes of lung cancer. Nodule detection is therefore key to the prevention of lung cancer-related deaths. In this work, a new paradigm is therefore presented that revolves around supervised and unsupervised machine learning approaches, superpixels, and graph neural networks to detect and segment nodules automatically. The proposed method, namely, Enhanced Graph Propagated Detector of nodules, harbors an enhanced uniform superpixel generation model and couples it with an improved version of graph propagation networks to diagnose nodules accurately in computed tomography images without the need for a high amount of annotated data. The architecture presented in this work is light weight and shows high capability in nodule detection and segmentation compared to several state-of-the-art models in the field. Experimental results show a significant reduction in superpixel generation time, a true positive rate of approximately 95%, and a low number of false positives on the Lung Image Database Consortium image collection dataset.