Scholarlink

Among other respiratory diseases, COVID-19 and pneumonia need to be identified quickly and precisely in order to guarantee proper patient care. Six cutting-edge deep learning CNN architectures—DenseNet121, ResNet50, VGG16, MobileNetV2, InceptionV3, and Xception—are thoroughly compared in this study in order to classify chest X-ray images into four groups: Normal, COVID-19, Bacterial Pneumonia, and Viral Pneumonia. A comprehensive set of metrics, including precision, recall, F1-score, and ROC curves, are used to assess the model’s performance using a carefully selected dataset of 9,208 X-ray pictures (1,281 COVID-19, 3,270 Normal, 3,001 Bacterial Pneumonia, and 1,656 Viral Pneumonia). The highest overall accuracy 91.69% is achieved by Xception, as demonstrated by the results, followed by InceptionV3 at 90.99%, whereas DenseNet121 demonstrates superior performance in COVID-19 detection (F1-score: 98.62%). The work offers insights for clinical application by highlighting the trade-offs between diagnostic performance and model complexity. All models were trained using the same hyperparameters (50 epochs, Adam optimizer, learning rate 0.0001) and tested on a stratified test set. This work could contribute to the growing body of research on AI-assisted radiological diagnostics, which is particularly relevant to pandemic response scenarios.

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

Performance Comparison of CNN Architectures for Multi-class Chest X-Ray Classification of COVID-19 and Pneumonia

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Md. Raihanul Haque,
  • Md. Rakibul Islam,
  • Shahriar Siddique Arjon,
  • Md. Siam Uddin Molla Antor,
  • Tamanna Yasmin,
  • Md. Rashik Uddin

摘要

Among other respiratory diseases, COVID-19 and pneumonia need to be identified quickly and precisely in order to guarantee proper patient care. Six cutting-edge deep learning CNN architectures—DenseNet121, ResNet50, VGG16, MobileNetV2, InceptionV3, and Xception—are thoroughly compared in this study in order to classify chest X-ray images into four groups: Normal, COVID-19, Bacterial Pneumonia, and Viral Pneumonia. A comprehensive set of metrics, including precision, recall, F1-score, and ROC curves, are used to assess the model’s performance using a carefully selected dataset of 9,208 X-ray pictures (1,281 COVID-19, 3,270 Normal, 3,001 Bacterial Pneumonia, and 1,656 Viral Pneumonia). The highest overall accuracy 91.69% is achieved by Xception, as demonstrated by the results, followed by InceptionV3 at 90.99%, whereas DenseNet121 demonstrates superior performance in COVID-19 detection (F1-score: 98.62%). The work offers insights for clinical application by highlighting the trade-offs between diagnostic performance and model complexity. All models were trained using the same hyperparameters (50 epochs, Adam optimizer, learning rate 0.0001) and tested on a stratified test set. This work could contribute to the growing body of research on AI-assisted radiological diagnostics, which is particularly relevant to pandemic response scenarios.