Knee-xRAI: Explainable AI Framework for Knee Osteoarthritis Grading
Knee-xRAI is a modular AI framework for automatic Kellgren-Lawrence grading of knee osteoarthritis. It independently quantifies joint space narrowing, osteophytes, and subchondral sclerosis using U-Net++ segmentation, SE-ResNet-50, and a hybrid texture-CNN classifier. The structured feature vector feeds XGBoost and ConvNeXt paths for explainable classification. The framework addresses inter-reader variability and opacity in deep learning approaches.
Key facts
- Knee-xRAI is a modular framework for KL grading of knee osteoarthritis.
- It quantifies three cardinal features: JSN, osteophytes, and subchondral sclerosis.
- Uses U-Net++ for JSN measurement, SE-ResNet-50 for osteophyte grading, and hybrid texture-CNN for sclerosis.
- Generates a 50-dimensional structured feature vector.
- XGBoost path supports SHAP-based feature attribution.
- ConvNeXt hybrid path combines structured features.
- Aims to reduce inter-reader variability and improve explainability.
- Published on arXiv with ID 2604.23435.
Entities
Institutions
- arXiv