Scholarlink

Abstract <p>Lung cancer remains one of the leading causes of death worldwide, with lung nodules serving as early indicators of the disease. This study introduces an Efficient Lung Nodule Segmentation Mechanism (ELNSM) to enhance lung cancer detection. This approach consists of three stages: image acquisition, pre-processing, and segmentation. Input images from publicly available datasets are pre-processing using Modified Guided Filtering (MGF) and Upgraded Contrast-Limited Adaptive Histogram Equalization (UCLAHE) to reduce noise and improve contrast. The lightweight Swin Transformer-based Parameterized Hypercomplex Residual U-Net (LightSwin-PHRUNet) is then used to precisely segment lung nodules, with hyper-parameter tuning using the Lyrebird Optimization Algorithm (LBOA). The proposed model is evaluated on the CT-scan, LIDC-IDRI, and LUNG-PET-CT-DX datasets, achieving Jaccard Scores of 0.98, 0.97, and 0.95; Dice Scores of 0.98, 0.98, and 0.97; and Mean IoU values of 0.98, 0.96, and 0.95, respectively. These results demonstrate the robustness and consistency of the proposed segmentation model across varied and challenging medical imaging datasets.</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

Lightweight Swin Transformer Based Parameterized Hypercomplex Convolutional Residual U-Net for Lung Nodule Segmentation

  • 期刊论文

  • 发表日期：

    2026年4月29日

查看全文
  • Saurabh Singh Raghuvanshi,
  • K. V. Arya,
  • Vinal Patel

摘要

Abstract

Lung cancer remains one of the leading causes of death worldwide, with lung nodules serving as early indicators of the disease. This study introduces an Efficient Lung Nodule Segmentation Mechanism (ELNSM) to enhance lung cancer detection. This approach consists of three stages: image acquisition, pre-processing, and segmentation. Input images from publicly available datasets are pre-processing using Modified Guided Filtering (MGF) and Upgraded Contrast-Limited Adaptive Histogram Equalization (UCLAHE) to reduce noise and improve contrast. The lightweight Swin Transformer-based Parameterized Hypercomplex Residual U-Net (LightSwin-PHRUNet) is then used to precisely segment lung nodules, with hyper-parameter tuning using the Lyrebird Optimization Algorithm (LBOA). The proposed model is evaluated on the CT-scan, LIDC-IDRI, and LUNG-PET-CT-DX datasets, achieving Jaccard Scores of 0.98, 0.97, and 0.95; Dice Scores of 0.98, 0.98, and 0.97; and Mean IoU values of 0.98, 0.96, and 0.95, respectively. These results demonstrate the robustness and consistency of the proposed segmentation model across varied and challenging medical imaging datasets.