Scholarlink

In recent years, lung diseases have significantly increased in the community. This condition causes millions of deaths annually. To address this issue, medicine requires effective, reliable, and rapid diagnostic methods for lung diseases. This paper proposes an efficient technique for classifying lung diseases from CXR images. In the study, pre-trained and fine-tuned deep learning models such as DenseNet, MobileNet, Inception, ResNet, EfficientNet, VGG16, and VGG19 were utilized to accurately predict four categories of lung diseases: bacterial pneumonia, viral pneumonia, tuberculosis, and COVID-19 based on X-ray images. The training process was divided into two steps, which form the proposed sequential fine-tuning method. First, the models have their final layers removed and replaced with one dense layer of 1024 neurons, with the original model layers frozen. The network trains only the newly added final layers. Subsequently, the model is unfrozen, and all weights are retrained. According to experimental findings, the proposed method is simple and performs better than other approaches. Specifically, the InceptionV3 model achieved a classification accuracy of 91.85%. The proposed method can be effectively used for lung disease classification.

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

Sequential Fine-Tuning Deep Learning for Lung Disease Classification Based on X-Ray Images

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Phuoc-Hai Huynh

摘要

In recent years, lung diseases have significantly increased in the community. This condition causes millions of deaths annually. To address this issue, medicine requires effective, reliable, and rapid diagnostic methods for lung diseases. This paper proposes an efficient technique for classifying lung diseases from CXR images. In the study, pre-trained and fine-tuned deep learning models such as DenseNet, MobileNet, Inception, ResNet, EfficientNet, VGG16, and VGG19 were utilized to accurately predict four categories of lung diseases: bacterial pneumonia, viral pneumonia, tuberculosis, and COVID-19 based on X-ray images. The training process was divided into two steps, which form the proposed sequential fine-tuning method. First, the models have their final layers removed and replaced with one dense layer of 1024 neurons, with the original model layers frozen. The network trains only the newly added final layers. Subsequently, the model is unfrozen, and all weights are retrained. According to experimental findings, the proposed method is simple and performs better than other approaches. Specifically, the InceptionV3 model achieved a classification accuracy of 91.85%. The proposed method can be effectively used for lung disease classification.