Scholarlink

<p>Tuberculosis (TB) remains the foremost infectious disease-related cause of mortality in India, emphasizing the need for rapid and precise diagnostic approaches. Chest X-ray (CXR) imaging, a cornerstone in TB diagnosis, traditionally depends on radiological expertise, which can lead to inconsistencies and inefficiencies. The advent of artificial intelligence and machine learning has introduced advanced methodologies to mitigate these challenges, particularly through the deployment of deep learning frameworks for automated CXR analysis. This research presents an innovative method that enhances TB detection by incorporating spatial attention mechanisms into a convolutional neural network (CNN) based feature fusion strategy. Using pre-trained CNN architectures such as Inception V3, DenseNet121, and VGG16, augmented with spatial attention layers, the proposed approach refines feature extraction by dynamically emphasizing pertinent image regions. These extracted features are subsequently integrated and classified via XGBoost, a robust gradient-boosting algorithm further refined through Bayesian optimization. Bayesian optimization contributes to computational efficiency by accelerating hyperparameter tuning such as N-estimators, learning rate, max depth, gamma, min-child-weight, subsample, and colsample-bytree thereby reducing training time while maintaining superior classification performance. Comprehensive experimentation on publicly accessible CXR datasets reveals that the fusion model significantly surpasses conventional techniques, attaining an accuracy of 95.50%, precision of 97.59%, recall of 93.30%, specificity of 97.70%, F1-score of 95.40%, and an AUC of 0.98, outperforming existing TB detection model such as NFNet. Statistical validation using the Wilcoxon signed-rank test (<i>p</i> &lt; 0.1) suggests a trend favoring our approach. Given its high accuracy and efficiency, the proposed approach holds strong potential for real-world applications such as mobile health clinics, mobile health solutions including deployment in remote clinics, and AI-assisted TB screening in resource-limited settings, aiding early diagnosis and timely intervention. Current findings demonstrate the efficacy of the proposed fusion approach in enhancing TB diagnostics, offering a highly effective and reliable tool for clinical implementation.</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

Integrating CNN-based feature fusion and spatial attention for tuberculosis detection with Bayesian optimized XGBoost

  • 期刊论文

  • 发表日期：

    2025年4月28日

查看全文
  • Samia Khan,
  • Farheen Siddiqui,
  • Mohd. Abdul Ahad

摘要

Tuberculosis (TB) remains the foremost infectious disease-related cause of mortality in India, emphasizing the need for rapid and precise diagnostic approaches. Chest X-ray (CXR) imaging, a cornerstone in TB diagnosis, traditionally depends on radiological expertise, which can lead to inconsistencies and inefficiencies. The advent of artificial intelligence and machine learning has introduced advanced methodologies to mitigate these challenges, particularly through the deployment of deep learning frameworks for automated CXR analysis. This research presents an innovative method that enhances TB detection by incorporating spatial attention mechanisms into a convolutional neural network (CNN) based feature fusion strategy. Using pre-trained CNN architectures such as Inception V3, DenseNet121, and VGG16, augmented with spatial attention layers, the proposed approach refines feature extraction by dynamically emphasizing pertinent image regions. These extracted features are subsequently integrated and classified via XGBoost, a robust gradient-boosting algorithm further refined through Bayesian optimization. Bayesian optimization contributes to computational efficiency by accelerating hyperparameter tuning such as N-estimators, learning rate, max depth, gamma, min-child-weight, subsample, and colsample-bytree thereby reducing training time while maintaining superior classification performance. Comprehensive experimentation on publicly accessible CXR datasets reveals that the fusion model significantly surpasses conventional techniques, attaining an accuracy of 95.50%, precision of 97.59%, recall of 93.30%, specificity of 97.70%, F1-score of 95.40%, and an AUC of 0.98, outperforming existing TB detection model such as NFNet. Statistical validation using the Wilcoxon signed-rank test (p < 0.1) suggests a trend favoring our approach. Given its high accuracy and efficiency, the proposed approach holds strong potential for real-world applications such as mobile health clinics, mobile health solutions including deployment in remote clinics, and AI-assisted TB screening in resource-limited settings, aiding early diagnosis and timely intervention. Current findings demonstrate the efficacy of the proposed fusion approach in enhancing TB diagnostics, offering a highly effective and reliable tool for clinical implementation.