Hybrid Ensemble Model for Knee Osteoarthritis Grading: Integrating CNNs with GLCM Features and XAI

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Orthopedics

Background:

• Knee osteoarthritis (KOA) involves cartilage degradation and joint narrowing.
• This leads to increased friction and structural joint damage.
• Accurate severity grading is crucial for effective treatment.

Purpose of the Study:

• To develop an automated framework for classifying KOA severity using knee X-rays.
• To integrate deep learning with traditional texture analysis for improved accuracy.
• To provide a reliable and interpretable method for radiographic KOA grading.

Main Methods:

• A composite hybrid framework using joint-centered cropping and data augmentation.
• Feature extraction via fine-tuned ResNet-101 and EfficientNetB7 models.
• Integration of deep features with Gray Level Co-occurrence Matrix (GLCM) texture descriptors.
• Soft-voting ensemble for final predictions and class weighting to address imbalance.

Main Results:

• The ensemble model achieved 73% test accuracy in a four-class setting (KL-0, KL-2, KL-3, KL-4).
• Achieved macro-F1 score of approximately 0.70 and weighted-F1 score of approximately 0.73.
• Grad-CAM analysis confirmed the model's focus on the knee joint region.

Conclusions:

• Combining ensemble deep learning with handcrafted texture features offers a robust approach.
• The framework provides a reliable and interpretable method for grading radiographic KOA.
• This automated approach can aid in clinical decision-making for knee osteoarthritis management.