Scholarlink

<p>Tuberculosis (TB), caused by Mycobacterium tuberculosis, is a highly contagious disease and one of the leading causes of death globally. Medical imaging research has aimed to equip doctors with AI-powered techniques for early detection, monitoring, and diagnosis of TB. Although recent efforts to automatically identify TB from chest X-ray (CXR) images have shown promising results, many existing methods have been tested on limited and homogeneous datasets. We believe this high performance may not extend to diverse, real- world data sources. Our current work seeks to bridge the gap by enhancing prediction performance on larger datasets. We propose a practical solution for detecting tuberculosis from CXR images using advanced Machine Learning and Computer Vision algorithms. Our framework incorporates three state-of-the-art deep neural networks: a modified Efficient Net, a modified Vision Transformer, and a modified Hybrid Efficient Net with Vision Transformer. Additionally, we strengthen the learning process with diverse augmentation techniques. We assessed our proposed method using a large dataset created by combining various public datasets. This dataset was divided into training (80%), validation (10%), and testing (10%) sets. To further analyze our approach, we compared it with two state-of-the-art systems. The results are promising, with our ViT_Base_EfficientNet_B1_224 achieving a maximum accuracy of 97.72% and an AUC of 98.97%. The experimental results show that our tool surpasses the baseline systems in multiple quality metrics.</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

Improving tuberculosis detection in chest X-rays using Vision Transformers and transfer learning with XGBoost

  • 期刊论文

  • 发表日期：

    2026年7月21日

查看全文
  • Ebtasam Ahmad Siddiqui,
  • Vijayshri Chaurasia,
  • Madhu Shandilya,
  • Praveen Kumar Tyagi,
  • Ajay Sharma,
  • Swati Gautam,
  • Rahul Prasad,
  • Uday Pratap Singh

摘要

Tuberculosis (TB), caused by Mycobacterium tuberculosis, is a highly contagious disease and one of the leading causes of death globally. Medical imaging research has aimed to equip doctors with AI-powered techniques for early detection, monitoring, and diagnosis of TB. Although recent efforts to automatically identify TB from chest X-ray (CXR) images have shown promising results, many existing methods have been tested on limited and homogeneous datasets. We believe this high performance may not extend to diverse, real- world data sources. Our current work seeks to bridge the gap by enhancing prediction performance on larger datasets. We propose a practical solution for detecting tuberculosis from CXR images using advanced Machine Learning and Computer Vision algorithms. Our framework incorporates three state-of-the-art deep neural networks: a modified Efficient Net, a modified Vision Transformer, and a modified Hybrid Efficient Net with Vision Transformer. Additionally, we strengthen the learning process with diverse augmentation techniques. We assessed our proposed method using a large dataset created by combining various public datasets. This dataset was divided into training (80%), validation (10%), and testing (10%) sets. To further analyze our approach, we compared it with two state-of-the-art systems. The results are promising, with our ViT_Base_EfficientNet_B1_224 achieving a maximum accuracy of 97.72% and an AUC of 98.97%. The experimental results show that our tool surpasses the baseline systems in multiple quality metrics.