Scholarlink

<p>Contrary to common belief, skin diseases can be life-threatening and have a serious impact on patients’ lives. Autoimmune blistering skin diseases (AIBD) are among them. The diagnosis based on clinical examination is subjective and irreproducible as it depends upon the expertise and experience of the treating dermatologists. A correct etiological diagnosis of AIBD requires extensive investigations that are costly and time-consuming, leading to delays in confirmation of the diagnosis. Early diagnosis and treatment of AIBD are very important for a better prognosis. Artificial intelligence techniques, particularly convolutional neural networks (CNNs), are being used across several medical domains, including dermatology, and have the potential to assist dermatologists in making early-stage decisions. This paper explores various AI-based&#xa0;approaches. The proposed models utilized traditional approaches, deep learning techniques, and hybrid methods to classify AIBD by analyzing clinical images. The models were compared using performance parameters, and their shortcomings and potential improvement strategies were analyzed based on the experimental results. In the traditional approach, the random forest attained the highest accuracy of 57%, while a hybrid model, Xception integrated with support vector machine (SVM), achieved an overall accuracy of 81%. However, in deep learning techniques, certain models performed very well and showed accuracy that ranged from 84 to 98%, particularly after additional fine-tuning processes.</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

Autoimmune blistering skin disease classification: classical machine learning, deep learning, and hybrid approaches

  • 期刊论文

  • 发表日期：

    2025年6月20日

查看全文
  • Manbir Singh,
  • Maninder Singh,
  • Dipankar De,
  • Sanjeev Handa,
  • Rahul Mahajan,
  • Vinod Hanumanthu,
  • Debajyoti Chatterjee

摘要

Contrary to common belief, skin diseases can be life-threatening and have a serious impact on patients’ lives. Autoimmune blistering skin diseases (AIBD) are among them. The diagnosis based on clinical examination is subjective and irreproducible as it depends upon the expertise and experience of the treating dermatologists. A correct etiological diagnosis of AIBD requires extensive investigations that are costly and time-consuming, leading to delays in confirmation of the diagnosis. Early diagnosis and treatment of AIBD are very important for a better prognosis. Artificial intelligence techniques, particularly convolutional neural networks (CNNs), are being used across several medical domains, including dermatology, and have the potential to assist dermatologists in making early-stage decisions. This paper explores various AI-based approaches. The proposed models utilized traditional approaches, deep learning techniques, and hybrid methods to classify AIBD by analyzing clinical images. The models were compared using performance parameters, and their shortcomings and potential improvement strategies were analyzed based on the experimental results. In the traditional approach, the random forest attained the highest accuracy of 57%, while a hybrid model, Xception integrated with support vector machine (SVM), achieved an overall accuracy of 81%. However, in deep learning techniques, certain models performed very well and showed accuracy that ranged from 84 to 98%, particularly after additional fine-tuning processes.