Scholarlink

<p>Johne’s disease, caused by <i>Mycobacterium avium</i> subspecies <i>paratuberculosis</i>, poses a significant threat to livestock health and the agricultural economy. Early and accurate detection through histopathological examination is the gold standard for diagnosis, but requires expert knowledge and remains labor-intensive and subject to human variability. While deep learning has been widely applied to human histopathology, no prior work has systematically explored its use for Johne’s disease. Addressing this gap, we investigate the potential of convolutional neural networks (CNNs) and vision transformers (ViTs) to automate the classification of histopathological slide images as positive or negative for Johne’s disease. A total of 14 public CNN transfer learning architectures, including VGG-19, ResNet, and InceptionV3, a custom lightweight CNN, and three ViTs were trained and evaluated using five-fold cross-validation to ensure robustness and prevent data leakage. Performance was assessed across six complementary metrics (accuracy, precision, recall, specificity, F-score, and Matthews correlation coefficient) to capture different aspects of classification quality. Results demonstrate that several models achieved high accuracy and consistency (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(&gt;95\%\)</EquationSource> </InlineEquation>), with the custom model showing competitive performance while maintaining computational efficiency. Gradient-weighted class activation mapping visualizations were employed to provide interpretability into model decision-making. Despite promising findings, limitations such as dataset diversity and image resolution were identified and will guide future improvements. By introducing and benchmarking CNNs and ViTs for the first time in Johne’s disease histopathology, this study establishes a performance baseline and highlights pathways for reliable AI-assisted diagnostic support in veterinary pathology.</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

Benchmarking binary classification of Johne’s disease using a multi-resolution histopathological image dataset

  • 期刊论文

  • 发表日期：

    2025年11月20日

查看全文
  • Mohammad Fraiwan,
  • Wael Hananeh

摘要

Johne’s disease, caused by Mycobacterium avium subspecies paratuberculosis, poses a significant threat to livestock health and the agricultural economy. Early and accurate detection through histopathological examination is the gold standard for diagnosis, but requires expert knowledge and remains labor-intensive and subject to human variability. While deep learning has been widely applied to human histopathology, no prior work has systematically explored its use for Johne’s disease. Addressing this gap, we investigate the potential of convolutional neural networks (CNNs) and vision transformers (ViTs) to automate the classification of histopathological slide images as positive or negative for Johne’s disease. A total of 14 public CNN transfer learning architectures, including VGG-19, ResNet, and InceptionV3, a custom lightweight CNN, and three ViTs were trained and evaluated using five-fold cross-validation to ensure robustness and prevent data leakage. Performance was assessed across six complementary metrics (accuracy, precision, recall, specificity, F-score, and Matthews correlation coefficient) to capture different aspects of classification quality. Results demonstrate that several models achieved high accuracy and consistency ( \(>95\%\) ), with the custom model showing competitive performance while maintaining computational efficiency. Gradient-weighted class activation mapping visualizations were employed to provide interpretability into model decision-making. Despite promising findings, limitations such as dataset diversity and image resolution were identified and will guide future improvements. By introducing and benchmarking CNNs and ViTs for the first time in Johne’s disease histopathology, this study establishes a performance baseline and highlights pathways for reliable AI-assisted diagnostic support in veterinary pathology.