Scholarlink

<p>Maintaining optimal lung health is essential for overall well-being, yet acute and chronic respiratory diseases (CRDs) continue to pose significant global health challenges, negatively impacting mortality, morbidity, and productivity. This study introduces an advanced diagnostic framework for detecting and classifying pulmonary conditions specifically Normal, Lung Opacity, and Viral Pneumonia from chest X-ray datasets, as well as Normal and Pneumonia cases from CT scans using deep transfer learning techniques. The framework leverages three state-of-the-art convolutional neural networks (CNNs): ResNet50, ResNet101, and EfficientNetB0 for initial image classification. In addition, two hybrid architectures are investigated: (1) feature-level integration, which merges features extracted from ResNet50 and EfficientNetB0, and (2) decision-level fusion, which combines the outputs of ResNet50 and ResNet101. To enhance clinical interpretability, Explainable Artificial Intelligence (XAI) techniques, including Local Interpretable Model-agnostic Explanations (LIME), SHapley Additive exPlanations (SHAP), and Integrated Gradients, are employed to highlight diagnostically relevant regions and provide transparency in model decision-making. Experimental results demonstrate that the decision-level fusion model (ResNet50 + ResNet101) significantly outperforms all baseline approaches. Specifically, it achieves 95.68% accuracy, 96.33% precision, 97.66% recall, and 97.33% F1 score on chest X-rays, and 99.61% accuracy, 99.50% precision, 99.50% recall, and a perfect 100% F1 score on CT scans. These findings underscore the effectiveness of hybrid architectures in medical image analysis and highlight the advantages of integrating transfer learning with XAI techniques, offering both enhanced diagnostic accuracy and improved model transparency for clinical decision support in pulmonary medicine.</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

Explainable fusion-based transfer learning for lung diseases classification

  • 期刊论文

  • 发表日期：

    2026年1月13日

查看全文
  • Aicha Akrout,
  • Amira Echtioui,
  • Mohamed Ghorbel

摘要

Maintaining optimal lung health is essential for overall well-being, yet acute and chronic respiratory diseases (CRDs) continue to pose significant global health challenges, negatively impacting mortality, morbidity, and productivity. This study introduces an advanced diagnostic framework for detecting and classifying pulmonary conditions specifically Normal, Lung Opacity, and Viral Pneumonia from chest X-ray datasets, as well as Normal and Pneumonia cases from CT scans using deep transfer learning techniques. The framework leverages three state-of-the-art convolutional neural networks (CNNs): ResNet50, ResNet101, and EfficientNetB0 for initial image classification. In addition, two hybrid architectures are investigated: (1) feature-level integration, which merges features extracted from ResNet50 and EfficientNetB0, and (2) decision-level fusion, which combines the outputs of ResNet50 and ResNet101. To enhance clinical interpretability, Explainable Artificial Intelligence (XAI) techniques, including Local Interpretable Model-agnostic Explanations (LIME), SHapley Additive exPlanations (SHAP), and Integrated Gradients, are employed to highlight diagnostically relevant regions and provide transparency in model decision-making. Experimental results demonstrate that the decision-level fusion model (ResNet50 + ResNet101) significantly outperforms all baseline approaches. Specifically, it achieves 95.68% accuracy, 96.33% precision, 97.66% recall, and 97.33% F1 score on chest X-rays, and 99.61% accuracy, 99.50% precision, 99.50% recall, and a perfect 100% F1 score on CT scans. These findings underscore the effectiveness of hybrid architectures in medical image analysis and highlight the advantages of integrating transfer learning with XAI techniques, offering both enhanced diagnostic accuracy and improved model transparency for clinical decision support in pulmonary medicine.