Scholarlink

<p>Osteoarthritis of the knee (OA) is a common degenerative condition that affects quality of life and mobility, especially in older adults. For disease management and treatment planning to be successful, early and precise diagnosis is essential. In order to classify the severity of osteoarthritis (OA) from knee X-ray pictures, this study suggests a new hybrid deep learning framework called Swin-O-NETS. It combines a Fast Extreme Learning Network (FELN) with a Modified Swin Transformer with Multi-Headed Channel Self-Attention for feature extraction. For evaluation, 2,047 radiographs from five Kellgren–Lawrence (KL) severity classes were taken from the Osteoarthritis Initiative (OAI) dataset on Kaggle. Our model outperformed traditional CNN, ResNet, DenseNet, and ensemble methods, achieving state-of-the-art performance with 99.4% accuracy, 99.0% precision, 98.9% recall, 98.3% specificity, and 98.8% F1-score respectively. These findings show that the suggested approach, which has better robustness and less computational complexity, can consistently help with early OA grading. Future research will concentrate on integrating multimodal imaging data, producing lightweight versions for use in real-time healthcare systems, and testing the model on larger multi-center clinical datasets.</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

Enhanced swin transformer with dual attention for knee osteoarthritis severity grading from X-ray images

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月30日

查看全文
  • K. Sudha,
  • A. Rajiv Kannan

摘要

Osteoarthritis of the knee (OA) is a common degenerative condition that affects quality of life and mobility, especially in older adults. For disease management and treatment planning to be successful, early and precise diagnosis is essential. In order to classify the severity of osteoarthritis (OA) from knee X-ray pictures, this study suggests a new hybrid deep learning framework called Swin-O-NETS. It combines a Fast Extreme Learning Network (FELN) with a Modified Swin Transformer with Multi-Headed Channel Self-Attention for feature extraction. For evaluation, 2,047 radiographs from five Kellgren–Lawrence (KL) severity classes were taken from the Osteoarthritis Initiative (OAI) dataset on Kaggle. Our model outperformed traditional CNN, ResNet, DenseNet, and ensemble methods, achieving state-of-the-art performance with 99.4% accuracy, 99.0% precision, 98.9% recall, 98.3% specificity, and 98.8% F1-score respectively. These findings show that the suggested approach, which has better robustness and less computational complexity, can consistently help with early OA grading. Future research will concentrate on integrating multimodal imaging data, producing lightweight versions for use in real-time healthcare systems, and testing the model on larger multi-center clinical datasets.