Scholarlink

<p>Chronic pulmonary diseases, including tuberculosis&#xa0;(TB), sarcoidosis, pulmonary alveolar proteinosis&#xa0;(PAP), and pulmonary fibrosis, impose a substantial burden on global health and are associated with significant morbidity and mortality. Accurate and timely diagnosis using chest radiographs remains a critical but inherently challenging task due to the overlapping radiological features characterizing these conditions. This study presents a comprehensive machine learning&#xa0;(ML) and deep learning&#xa0;(DL) framework for binary classification of chest X-rays into normal and abnormal (disease) categories. Using Haralick texture-based features derived from the Gray-Level Co-occurrence Matrix&#xa0;(GLCM) alongside a stacking ensemble of Support Vector Machine&#xa0;(SVM), K-Nearest Neighbors&#xa0;(KNN), and Logistic Regression, the ML model achieved 98.00% accuracy. Among the evaluated DL architectures, a fine-tuned AlexNet model yielded the highest test accuracy of 98.92%, with an area under the receiver operating characteristic curve&#xa0;(AUC) of 0.997. The dataset comprised 1934 chest radiographs drawn from the publicly available Kaggle repository (fernando2rad/x-ray-lung-diseases-images-9-classes), which was originally compiled from established clinical imaging repositories including the NIH ChestX-ray14, the Montgomery County TB dataset, and the Shenzhen Hospital dataset. All experiments employed strict patient-level train–test splitting and fivefold cross-validation to mitigate data leakage. Grad-CAM saliency maps were generated for the best-performing DL model to provide interpretable visual explanations of classification decisions. Bootstrap confidence intervals confirmed the robustness of the reported metrics. DL models demonstrated superior generalization, underscoring their strong potential for integration into automated pulmonary diagnostic workflows in clinical radiology practice.</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

Automated detection of chronic pulmonary diseases via stacked machine learning and CNN-based analysis of chest X-rays

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月20日

查看全文
  • Alhussein Abdullah,
  • Mahmoud Rabea,
  • Amr Omar,
  • Mai Mabrouk

摘要

Chronic pulmonary diseases, including tuberculosis (TB), sarcoidosis, pulmonary alveolar proteinosis (PAP), and pulmonary fibrosis, impose a substantial burden on global health and are associated with significant morbidity and mortality. Accurate and timely diagnosis using chest radiographs remains a critical but inherently challenging task due to the overlapping radiological features characterizing these conditions. This study presents a comprehensive machine learning (ML) and deep learning (DL) framework for binary classification of chest X-rays into normal and abnormal (disease) categories. Using Haralick texture-based features derived from the Gray-Level Co-occurrence Matrix (GLCM) alongside a stacking ensemble of Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Logistic Regression, the ML model achieved 98.00% accuracy. Among the evaluated DL architectures, a fine-tuned AlexNet model yielded the highest test accuracy of 98.92%, with an area under the receiver operating characteristic curve (AUC) of 0.997. The dataset comprised 1934 chest radiographs drawn from the publicly available Kaggle repository (fernando2rad/x-ray-lung-diseases-images-9-classes), which was originally compiled from established clinical imaging repositories including the NIH ChestX-ray14, the Montgomery County TB dataset, and the Shenzhen Hospital dataset. All experiments employed strict patient-level train–test splitting and fivefold cross-validation to mitigate data leakage. Grad-CAM saliency maps were generated for the best-performing DL model to provide interpretable visual explanations of classification decisions. Bootstrap confidence intervals confirmed the robustness of the reported metrics. DL models demonstrated superior generalization, underscoring their strong potential for integration into automated pulmonary diagnostic workflows in clinical radiology practice.