Enhanced Knee Osteoarthritis Classification using a Modified EfficientNetB5 Architecture with Self-Attention Mechanism

Knee osteoarthritis, a debilitating and threatening musculoskeletal condition, underscores the critical importance of early detection for timely treatment. Given the seriousness of this health issue, this study explores the use of deep learning as an effective diagnostic tool. The study introduces an innovative method for diagnosing and classifying knee osteoarthritis, achieved through the modification of an EfficientNetB5 architecture and the integration of the Self Attention mechanism, addressing current limitations. The attention mechanism selectively emphasizes crucial features, enabling the model to prioritize and weigh pertinent information in the classification process. To ensure a strong approach, a custom callback is used in training coupled with strategies to solve issues of class imbalance. This approach greatly enhances the diagnosis of musculoskeletal health, meeting the critical need for early intervention with an exceptional 90.86% accuracy with a comparatively smaller dataset. The intentional alteration of the model, along with an emphasis on the Self Attention mechanism, signifies a break from custom and a sincere attempt to improve medical image analysis. The strategy’s proven effectiveness not only supports it but also points out areas for improvement, perhaps leading to more accurate diagnostic tools. This technique represents a significant advancement in the treatment of osteoarthritis in the knee, a condition that may have fatal implications.