Scholarlink

<p>Pneumonia and various bone problems are potentially serious infections that require prompt medical attention, especially in vulnerable populations. Early diagnosis and appropriate treatment are critical for recovery and reducing the risk of complications. While CNN-based models for pneumonia classification on X-rays can offer significant benefits, such as fast and accurate diagnosis, they come with notable disadvantages. These include issues with data quality, class imbalance, overfitting, lack of interpretability, and the risk of misdiagnosis. Also, transfer learning, where models pre-trained on large public datasets e.g., ImageNet are fine-tuned on medical images, has limitations. Medical images are very different from natural images, and the feature representations learned from non-medical datasets might not be directly applicable. In this regard, we designed a deep learning model based on the attention mechanism to accurately extract features and classify X-ray images. By combining self-attention and spatial attention, we create a powerful hybrid mechanism that is more efficient and effective in extracting relevant features. The proposed model achieved high accuracy and low loss for the Chest X-ray dataset (99.82% accuracy, 0.24% loss), Bone Fracture Multi-Region X-ray Dataset (100.00% accuracy, 3.40% loss) and Shoulder X-ray dataset (97.53% accuracy, 0.42% loss) classifications.</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

An accurate attention based method for multi-tasking X-ray classification

  • 期刊论文

  • 发表日期：

    2025年9月1日

查看全文
  • Zahra Raeisi,
  • Shayan Rokhva,
  • Amirsadegh Roshanzamir,
  • Reza ahmadi lashaki

摘要

Pneumonia and various bone problems are potentially serious infections that require prompt medical attention, especially in vulnerable populations. Early diagnosis and appropriate treatment are critical for recovery and reducing the risk of complications. While CNN-based models for pneumonia classification on X-rays can offer significant benefits, such as fast and accurate diagnosis, they come with notable disadvantages. These include issues with data quality, class imbalance, overfitting, lack of interpretability, and the risk of misdiagnosis. Also, transfer learning, where models pre-trained on large public datasets e.g., ImageNet are fine-tuned on medical images, has limitations. Medical images are very different from natural images, and the feature representations learned from non-medical datasets might not be directly applicable. In this regard, we designed a deep learning model based on the attention mechanism to accurately extract features and classify X-ray images. By combining self-attention and spatial attention, we create a powerful hybrid mechanism that is more efficient and effective in extracting relevant features. The proposed model achieved high accuracy and low loss for the Chest X-ray dataset (99.82% accuracy, 0.24% loss), Bone Fracture Multi-Region X-ray Dataset (100.00% accuracy, 3.40% loss) and Shoulder X-ray dataset (97.53% accuracy, 0.42% loss) classifications.