Related Experiment Video
Updated: Jul 1, 2026
Automated Joint Space Detection Improves Bone Segmentation Accuracy
Published on: November 28, 2025
Automated Kellgren-Lawrence grading of knee osteoarthritis using a multi-scale attention-based deep learning
Henghui Zhang1, Chui Kong2, Hanwen Chang1
1Shanghai Key Laboratory of Orthopaedic Implants, Department of Orthopaedic Surgery, Shanghai Jiao Tong University School of Medicine, School of Biomedical Engineering, Shanghai Ninth People's Hospital, Shanghai Jiao Tong University, Shanghai, China.
This study introduces a novel deep learning framework for accurate knee osteoarthritis (KOA) radiographic grading using the Kellgren-Lawrence (KL) system. The model demonstrates strong generalization across datasets, improving objective clinical assessment.
Area of Science:
- Radiology
- Artificial Intelligence
- Medical Imaging Analysis
Background:
- Accurate radiographic grading of knee osteoarthritis (KOA) using the Kellgren-Lawrence (KL) system is crucial for clinical decisions.
- Current deep learning models for KOA grading often lack multi-scale feature integration and effective attention mechanisms, leading to poor generalization.
Purpose of the Study:
- To develop and validate a novel multi-scale attention-guided deep learning framework for automated and reliable KL grading of KOA.
- To address limitations of existing models, including single-scale feature reliance and poor external dataset generalization.
Main Methods:
- A deep learning framework integrating a Feature Pyramid Network (FPN) for multi-scale feature fusion.
- Incorporation of a dual-attention mechanism (Squeeze-and-Excitation and Self-Attention) for enhanced feature recalibration and spatial dependency modeling.
- Utilized knowledge distillation to improve model generalization, trained on 5,000 KOA radiographs, and validated on an independent dataset of 2,000 radiographs.
Main Results:
- The proposed model achieved superior performance on internal validation (F1: 0.726, accuracy: 0.726) and maintained robust generalization on external validation (F1: 0.656, accuracy: 0.685).
- Misclassifications were limited to adjacent KL grades, indicating clinical reliability and minimal extreme errors.
- Gradient-weighted Class Activation Mapping (Grad-CAM) visualizations confirmed that the model's attention focused on clinically relevant regions, enhancing interpretability.
Conclusions:
- The multi-scale attention-guided framework provides reliable, automated KL grading of KOA from radiographs with strong cross-dataset generalization.
- This approach overcomes limitations of previous models and offers potential for standardized, objective, and interpretable clinical assessment of KOA severity.