Scholarlink

<p>Oral cancer is the most prevalent type of cancer, and early detection is crucial for reducing mortality rates. This creates opportunities for advancing computer image analysis tools that alleviate the workload of pathologists. In recent years, deep learning has been utilized to automate the detection of various pathologies through digital images. Consequently, this paper presents an edge-supervised deep learning framework for the automated classification of oral cancer histopathology images. The proposed approach aims to enhance nuclear edges and pixel intensities, convert the histopathology image to grayscale, and extract its edges. Afterward, the resulting output is processed through a sequence of newly proposed feature extraction blocks that integrate the image and edge features. This study examines two datasets: The first includes normal subjects and individuals with oral squamous cell carcinoma (OSCC), while the second comprises OSCC, leukoplakia without dysplasia, and leukoplakia with dysplasia. The experimental results indicate that the proposed framework outperforms convolutional neural networks (CNN) and vision transformer models, with the first dataset achieving a balanced classification accuracy (BCC) of 96.80%. The second dataset, on the other hand, revealed that 90.66% of BCCs were presence or absence for dysplasia, and 92.46% were positive for leukoplakia without dysplasia, leukoplakia with dysplasia, and OSCC. Other performance metrics were also calculated, including recall, sensitivity, F1-score, and specificity. The findings suggest that the framework is a promising diagnostic tool to assist pathologists in classifying oral cancer images more accurately.</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

Edge-supervised convolutional neural network for histopathological classification of oral cancer images

  • 期刊论文

  • 发表日期：

    2025年7月4日

查看全文
  • Ahmed M. Gab Allah,
  • Mohamed M. S. Gaballa,
  • Nada M. Elshennawy,
  • Amr Elkholy

摘要

Oral cancer is the most prevalent type of cancer, and early detection is crucial for reducing mortality rates. This creates opportunities for advancing computer image analysis tools that alleviate the workload of pathologists. In recent years, deep learning has been utilized to automate the detection of various pathologies through digital images. Consequently, this paper presents an edge-supervised deep learning framework for the automated classification of oral cancer histopathology images. The proposed approach aims to enhance nuclear edges and pixel intensities, convert the histopathology image to grayscale, and extract its edges. Afterward, the resulting output is processed through a sequence of newly proposed feature extraction blocks that integrate the image and edge features. This study examines two datasets: The first includes normal subjects and individuals with oral squamous cell carcinoma (OSCC), while the second comprises OSCC, leukoplakia without dysplasia, and leukoplakia with dysplasia. The experimental results indicate that the proposed framework outperforms convolutional neural networks (CNN) and vision transformer models, with the first dataset achieving a balanced classification accuracy (BCC) of 96.80%. The second dataset, on the other hand, revealed that 90.66% of BCCs were presence or absence for dysplasia, and 92.46% were positive for leukoplakia without dysplasia, leukoplakia with dysplasia, and OSCC. Other performance metrics were also calculated, including recall, sensitivity, F1-score, and specificity. The findings suggest that the framework is a promising diagnostic tool to assist pathologists in classifying oral cancer images more accurately.