Scholarlink

Lung diseases are critical health conditions that significantly impact quality of life and can be life-threatening. Causes include environmental pollution, smoking, genetics, and viral infections. Chest X-ray is a common diagnostic tool, but image analysis requires time and expertise. Deep learning approaches have proven to be highly effective in interpreting medical images. This study develops a model combining EfficientNet with Feature Pyramid Network for lung disease classification and detection, trained and tested on the COVID-19 Radiography dataset, which contains 21,165 X-ray images across four disease classes: COVID19, Lung Opacity, Normal Lung, and Pneumonia. A 70:15:15 ratio was employed to partition the dataset into training, validation, and testing portions. Results obtained through experimentation reveal that the combined model achieved 92.93% accuracy on the training set, outperforming individual models like standalone EfficientNet (91.82%), InceptionV3 (91.73%), ResNet50 (90.81%), Xception (90.08%), and VGG19 (90.69%). On the test set, the combined model achieved 96.00% accuracy, while individual models like EfficientNet, InceptionV3, ResNet50, and VGG19 reached 88.50%, 93.50%, 91.25%, and 92.25%, respectively.

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

Leveraging Deep Learning for Lung Disease Diagnosis and Classification Through X-ray Imaging

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Nguyen Minh Khiem,
  • Pham Ngoc Quyen,
  • Tran Duy Quang,
  • Ngo-Ho Anh-Khoi

摘要

Lung diseases are critical health conditions that significantly impact quality of life and can be life-threatening. Causes include environmental pollution, smoking, genetics, and viral infections. Chest X-ray is a common diagnostic tool, but image analysis requires time and expertise. Deep learning approaches have proven to be highly effective in interpreting medical images. This study develops a model combining EfficientNet with Feature Pyramid Network for lung disease classification and detection, trained and tested on the COVID-19 Radiography dataset, which contains 21,165 X-ray images across four disease classes: COVID19, Lung Opacity, Normal Lung, and Pneumonia. A 70:15:15 ratio was employed to partition the dataset into training, validation, and testing portions. Results obtained through experimentation reveal that the combined model achieved 92.93% accuracy on the training set, outperforming individual models like standalone EfficientNet (91.82%), InceptionV3 (91.73%), ResNet50 (90.81%), Xception (90.08%), and VGG19 (90.69%). On the test set, the combined model achieved 96.00% accuracy, while individual models like EfficientNet, InceptionV3, ResNet50, and VGG19 reached 88.50%, 93.50%, 91.25%, and 92.25%, respectively.