Scholarlink

<p>Accurate delineation of skin lesions in images is important for skin cancer detection. Existing methods often struggle with inherent complexities, such as irregular boundaries, textures, and artefacts in skin lesions. The study proposes a hybrid model comprising the edge-accurate LEDNet and Swin-UMamba for multiscale segmentation. The irregular boundaries and complex textures of skin lesions can be captured more effectively through this integration than with previous stand-alone methods. The structure of LEDNet includes components that enable it to segment lesions of various types effectively. Swin-Mamba is an encoder that uses Mamba-based architecture with the additional component of the VSS block. The proposed model is evaluated on the Ph<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(^2\)</EquationSource> </InlineEquation>, ISIC-2017 and ISIC-2018 skin cancer datasets and demonstrates robust performance across all datasets. The method achieved a Dice Similarity Coefficient (DSC) of 0.9734, a sensitivity of 0.9697, a specificity of 0.9858 and an accuracy of 0.9847 with ISIC 2017, DSC of 0.9753, a sensitivity of 0.9494, a specificity of 0.9902 and an accuracy of 0.9713 with ISIC 2018; and a DSC of 0.9801, a sensitivity of 0.9892, a specificity of 0.9966 and an accuracy of 0.9932 with Ph. These results show that the proposed hybrid framework has the potential to bring important benefits in the segmentation of skin lesions and is promising in clinical dermatology.</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

A hybrid approach for accurate skin lesion segmentation using LEDNet and Swin-UMamba

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月6日

查看全文
  • Muhammad Ahtsam Naeem,
  • Shangming Yang,
  • Muhammad Asim Saleem,
  • Ashir Javeed,
  • Tanveer Ahmad

摘要

Accurate delineation of skin lesions in images is important for skin cancer detection. Existing methods often struggle with inherent complexities, such as irregular boundaries, textures, and artefacts in skin lesions. The study proposes a hybrid model comprising the edge-accurate LEDNet and Swin-UMamba for multiscale segmentation. The irregular boundaries and complex textures of skin lesions can be captured more effectively through this integration than with previous stand-alone methods. The structure of LEDNet includes components that enable it to segment lesions of various types effectively. Swin-Mamba is an encoder that uses Mamba-based architecture with the additional component of the VSS block. The proposed model is evaluated on the Ph \(^2\) , ISIC-2017 and ISIC-2018 skin cancer datasets and demonstrates robust performance across all datasets. The method achieved a Dice Similarity Coefficient (DSC) of 0.9734, a sensitivity of 0.9697, a specificity of 0.9858 and an accuracy of 0.9847 with ISIC 2017, DSC of 0.9753, a sensitivity of 0.9494, a specificity of 0.9902 and an accuracy of 0.9713 with ISIC 2018; and a DSC of 0.9801, a sensitivity of 0.9892, a specificity of 0.9966 and an accuracy of 0.9932 with Ph. These results show that the proposed hybrid framework has the potential to bring important benefits in the segmentation of skin lesions and is promising in clinical dermatology.