Scholarlink

<p>The SARS-CoV-2 pandemic has underscored the need for robust and interpretable computer-aided diagnostic systems in radiological imaging. In this work, we present a hybrid deep learning architecture that integrates Convolutional Neural Networks (CNNs) with Transformer modules to improve COVID-19 detection from chest radiographs. The proposed model combines DenseNet121 for hierarchical feature extraction with Vision Transformer (ViT) components to capture long-range dependencies in medical images. Experimental evaluation on the publicly available COVID-19 Radiography Database (hosted on Kaggle and Mendeley) demonstrates an accuracy of 89.39%, along with high sensitivity and specificity on the test set. To ensure clinical relevance and interpretability, a multi-modal explainability framework comprising Grad-CAM, SHAP, and Layer-wise Relevance Propagation (LRP) was employed. The generated saliency maps showed strong alignment with radiologist-annotated regions of interest (ROIs), validating the model’s decisions. Comparative analysis with state-of-the-art models such as ResNet-50, standalone DenseNet-121, and ViT revealed that our hybrid approach achieves statistically significant improvements across diagnostic performance metrics. Moreover, the model exhibits reduced computational complexity and real-time inference capability, making it suitable for deployment in resource-constrained clinical settings. This research contributes a high-performing and explainable framework for medical image classification, with promising applications in automated screening and diagnosis of respiratory diseases, including COVID-19.</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 AI hybrid CNN-transformer models for enhanced COVID-19 detection in chest X-rays

  • 期刊论文

  • 发表日期：

    2025年9月8日

查看全文
  • C. V. Aravinda,
  • B. C. Arjun,
  • Mohamed Samiulla Khan,
  • Santhosh John,
  • Shaik Asif Hussain

摘要

The SARS-CoV-2 pandemic has underscored the need for robust and interpretable computer-aided diagnostic systems in radiological imaging. In this work, we present a hybrid deep learning architecture that integrates Convolutional Neural Networks (CNNs) with Transformer modules to improve COVID-19 detection from chest radiographs. The proposed model combines DenseNet121 for hierarchical feature extraction with Vision Transformer (ViT) components to capture long-range dependencies in medical images. Experimental evaluation on the publicly available COVID-19 Radiography Database (hosted on Kaggle and Mendeley) demonstrates an accuracy of 89.39%, along with high sensitivity and specificity on the test set. To ensure clinical relevance and interpretability, a multi-modal explainability framework comprising Grad-CAM, SHAP, and Layer-wise Relevance Propagation (LRP) was employed. The generated saliency maps showed strong alignment with radiologist-annotated regions of interest (ROIs), validating the model’s decisions. Comparative analysis with state-of-the-art models such as ResNet-50, standalone DenseNet-121, and ViT revealed that our hybrid approach achieves statistically significant improvements across diagnostic performance metrics. Moreover, the model exhibits reduced computational complexity and real-time inference capability, making it suitable for deployment in resource-constrained clinical settings. This research contributes a high-performing and explainable framework for medical image classification, with promising applications in automated screening and diagnosis of respiratory diseases, including COVID-19.