Scholarlink

Chest X-rays are critical for diagnosing respiratory diseases like pneumonia and tuberculosis, yet manual interpretation can be prone to errors and delays, affecting patient care. This study explores the application of transfer learning using pre-trained models ResNet-50 and EfficientNet-B3, initially trained on large datasets, to improve disease classification from the NIH Chest X-ray Dataset. Our findings demonstrate that transfer learning enhances diagnostic accuracy and reduces training time, with EfficientNet-B3 having 62% accuracy outperforming ResNet-50 with 55% accuracy and also in scalability. However, the ResNet-50 model also performs better in some specific class classification, short training time, and fast inference speed than the Efficient-B3 model. This research highlights the potential of transfer learning to provide faster, more reliable diagnostic support in clinical settings while addressing challenges such as dataset diversity and model interpretability for future investigations.

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

Adapting Medical Imaging: Transfer Learning for Chest X-ray Diagnosis

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Shalini Gambhir,
  • Sumit,
  • Aman Raj,
  • Khushi Sharma

摘要

Chest X-rays are critical for diagnosing respiratory diseases like pneumonia and tuberculosis, yet manual interpretation can be prone to errors and delays, affecting patient care. This study explores the application of transfer learning using pre-trained models ResNet-50 and EfficientNet-B3, initially trained on large datasets, to improve disease classification from the NIH Chest X-ray Dataset. Our findings demonstrate that transfer learning enhances diagnostic accuracy and reduces training time, with EfficientNet-B3 having 62% accuracy outperforming ResNet-50 with 55% accuracy and also in scalability. However, the ResNet-50 model also performs better in some specific class classification, short training time, and fast inference speed than the Efficient-B3 model. This research highlights the potential of transfer learning to provide faster, more reliable diagnostic support in clinical settings while addressing challenges such as dataset diversity and model interpretability for future investigations.