Scholarlink

Skin lesions are a sort of irregularity on the skin. Skin lesions can be mainly categorized into benign or malignant. Malignant skin cancers pose a serious risk to the life of patients. Computer-assisted diagnosis of skin diseases by using artificial intelligence-based techniques can be a solution to detect skin diseases in their early development stages. This method complements the existing human expert-level diagnosis of such diseases and can benefit patients where there is a scarcity of expert dermatologists or other facilities. In this paper, a model for skin lesion classification of dermoscopic images for the diagnosis of skin diseases is proposed by using the fusion-based architecture consisting of Vision Transformer and ConvNext. We have used the HAM10000, ISIC2016, and ISIC2017 datasets for experimentation which contain a large collection of dermatoscopic pictures from many sources depicting typical skin lesions with pigment. The proposed fusion-based model architecture achieves significant performance on the datasets utilized for experimentation, which clearly depicts the efficacy of the proposed model for skin lesion classification.

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

Skin Lesion Classification Using CNN and Transformer Networks for Computer-Assisted Diagnosis

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Prasad Kanhegaonkar,
  • Sruthi Ponugoti,
  • Surya Prakash

摘要

Skin lesions are a sort of irregularity on the skin. Skin lesions can be mainly categorized into benign or malignant. Malignant skin cancers pose a serious risk to the life of patients. Computer-assisted diagnosis of skin diseases by using artificial intelligence-based techniques can be a solution to detect skin diseases in their early development stages. This method complements the existing human expert-level diagnosis of such diseases and can benefit patients where there is a scarcity of expert dermatologists or other facilities. In this paper, a model for skin lesion classification of dermoscopic images for the diagnosis of skin diseases is proposed by using the fusion-based architecture consisting of Vision Transformer and ConvNext. We have used the HAM10000, ISIC2016, and ISIC2017 datasets for experimentation which contain a large collection of dermatoscopic pictures from many sources depicting typical skin lesions with pigment. The proposed fusion-based model architecture achieves significant performance on the datasets utilized for experimentation, which clearly depicts the efficacy of the proposed model for skin lesion classification.