Scholarlink

Techniques for detecting and diagnosing abnormalities in the chest and lungs have made significant progress, but chest X-ray (CXR) diagnosis remains popular due to its availability, efficiency, and high reliability. However, diagnostic errors in CXR still persist due to heavy workloads, the complexity of images, and disparities in expertise among hospitals. This creates an urgent need for automated deep learning-based support tools to minimize errors. Although deep learning has demonstrated potential in medical image classification and prediction, many studies have not fully integrated current trends and modern methods in predicting CXR abnormalities. Previous research works only addressed the issue of processing 14 abnormality labels while neglecting the “No findings” label, which is inconsistent with real-world scenarios. Therefore, in this study, we integrated classification and object detection methods, combined with data imbalance handling techniques before analysis, to improve the accuracy of detecting abnormalities. As a result, the research team developed an application that incorporates these techniques to assist doctors in diagnosis and treatment.

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

Detecting Abnormalities in Chest X-Ray Images by Combining Classification Methods, Object Detection, and Data Balancing

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Nhu Q. T. Nguyen,
  • Dat Q. Nguyen,
  • Dat T. Ta,
  • Huy V. Tran

摘要

Techniques for detecting and diagnosing abnormalities in the chest and lungs have made significant progress, but chest X-ray (CXR) diagnosis remains popular due to its availability, efficiency, and high reliability. However, diagnostic errors in CXR still persist due to heavy workloads, the complexity of images, and disparities in expertise among hospitals. This creates an urgent need for automated deep learning-based support tools to minimize errors. Although deep learning has demonstrated potential in medical image classification and prediction, many studies have not fully integrated current trends and modern methods in predicting CXR abnormalities. Previous research works only addressed the issue of processing 14 abnormality labels while neglecting the “No findings” label, which is inconsistent with real-world scenarios. Therefore, in this study, we integrated classification and object detection methods, combined with data imbalance handling techniques before analysis, to improve the accuracy of detecting abnormalities. As a result, the research team developed an application that incorporates these techniques to assist doctors in diagnosis and treatment.