-
1
A novel multi-scale attention-guided deep learning framework was developed for automated grading of knee osteoarthritis using the Kellgren–Lawrence system.
-
2
The framework integrates a Feature Pyramid Network, dual-attention mechanisms, and knowledge distillation to enhance model performance and generalization.
-
3
Internal validation showed superior performance metrics, including F1 score of 0.726, precision of 0.740, and accuracy of 0.726, outperforming baseline models.
-
4
External validation demonstrated robust generalization with an F1 score of 0.656 and accuracy of 0.685, indicating clinical reliability of the model.
-
5
The model's attention aligns with clinically relevant regions, enhancing interpretability and supporting standardized assessment of KOA severity.
