Knee Osteoarthritis Severity Classification Using Enhanced EfficientNet-B4 with Preprocessing Techniques

Knee osteoarthritis (KOA) is a long-term persistent joint disease that impacts millions globally, causing discomfort, joint stiffness, and restricted mobility. Early and accurate classification of KOA severity helps in effective disease management and appropriate treatment planning. In this study, our approach used an enhanced EfficientNet-B4 model for automated KOA severity classification using X-ray images. Our model was trained on the Knee Osteoarthritis Dataset with Severity Grading, comprising more than 10,000 X-ray images categorized into five severity levels: healthy, doubtful, minimal, moderate, and severe. To improve generalization, we employed advanced image preprocessing techniques, including data augmentation, image normalization, and feature enhancement. The model achieved 96% accuracy, 97% precision, recall of 96% and a 96% F1-score, surpassing conventional CNN-based approaches. These findings demonstrate the potential of deep learning for clinical integration, offering a scalable and efficient AI-driven solution for KOA diagnosis and severity assessment.