Scholarlink

<p>Early diagnosis of melanoma skin disease is critical for effective treatments as it is the deadliest form of skin illness and spreads quickly to other areas of the body unless identified and treated early. In clinical diagnoses of different disorders, non-invasive medical computer vision or medical imaging plays an increasingly important role. Such approaches give an automated image analysis tool for an accurate and quick assessment of the disease. The primary aim of this study is to detect melanoma skin cancer in its early stages by getting more accurate findings because there is an exponential increase in skin cancer patients worldwide, high medical expenditures, and an increase in death risk due to not commencing the diagnosis at an early stage, which is due to late detection. Our ViTDelMel approach, a vision transformer-based deep learning framework for early melanoma detection using dermoscopic images, used the HAM10K online dermoscopic images database. Our approach leverages the latest vision transformer backend along with a multi-head self-attention module for more accurate detection of melanoma. We also implemented three well-known DL models, i.e., AlexNet, SqueezeNet and DenseNet121 DL models and compared results with our approach. VitDelMel achieved the maximum classification accuracy of 97.62% whereas AlexNet, SqueezeNet, and DenseNet121 achieved 92.04%, 92.16%, and 92.77% respectively. We also compared our approach with the latest architecture in the literature and found better results. Our network is helpful for the early diagnosis of melanoma skin cancer and can classify melanoma and non-melanoma accurately and efficiently.</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

ViTDelMel: Vision Transformer-Based Melanoma Skin Disease Detection Through Dermoscopic Images

  • 期刊论文

  • 发表日期：

    2025年11月5日

查看全文
  • Zubair Saeed,
  • Ali Raza

摘要

Early diagnosis of melanoma skin disease is critical for effective treatments as it is the deadliest form of skin illness and spreads quickly to other areas of the body unless identified and treated early. In clinical diagnoses of different disorders, non-invasive medical computer vision or medical imaging plays an increasingly important role. Such approaches give an automated image analysis tool for an accurate and quick assessment of the disease. The primary aim of this study is to detect melanoma skin cancer in its early stages by getting more accurate findings because there is an exponential increase in skin cancer patients worldwide, high medical expenditures, and an increase in death risk due to not commencing the diagnosis at an early stage, which is due to late detection. Our ViTDelMel approach, a vision transformer-based deep learning framework for early melanoma detection using dermoscopic images, used the HAM10K online dermoscopic images database. Our approach leverages the latest vision transformer backend along with a multi-head self-attention module for more accurate detection of melanoma. We also implemented three well-known DL models, i.e., AlexNet, SqueezeNet and DenseNet121 DL models and compared results with our approach. VitDelMel achieved the maximum classification accuracy of 97.62% whereas AlexNet, SqueezeNet, and DenseNet121 achieved 92.04%, 92.16%, and 92.77% respectively. We also compared our approach with the latest architecture in the literature and found better results. Our network is helpful for the early diagnosis of melanoma skin cancer and can classify melanoma and non-melanoma accurately and efficiently.