Abstract
Knee osteoarthritis (KOA) is a major cause of disability, particularly among older adults, because of the degeneration of articular cartilage in the knee joint. This disorder is characterized by stiffness, reduced mobility, and pain, which makes medical diagnosis challenging, especially given the current limitations in achieving timely and accurate detection and progression analysis. Moreover, the manual interpretation of X-ray images for KOA grading is subjective and diverse among clinicians. Hence, this research proposes the gated axial attention mechanism based on DenseNet121 (GAA-DenseNet121) for the KOA classification. The DenseNet121 effectively extracts multi-level spatial features across layers, whereas GAA improves the feature maps by concentrating on the most clinically relevant regions, such as the medial and lateral joint compartments. The experimental discoveries illustrate that the proposed GAA-DenseNet121 approach obtained a better accuracy of 99.25% and 99.21% on Mendeley and OAI datasets individually, as compared to the existing approaches such as CenterNet.