Transfer Learning Based EfficientNet for Knee Osteoarthritis Classification

Knee Osteoarthritis (KOA) is a prevalent condition that deteriorates with time and may lead to disability. Diagnosis depends on subjective symptom evaluation and radiograph analysis. The Kellgren-Lawrence (KL) grading system is widely used to assess the severity of knee osteoarthritis, with grades ranging from 0 (no osteoarthritis) to 4 (severe osteoarthritis). The detection and classification of KOA play a crucial role in medical diagnosis and treatment planning. It enables healthcare professionals to detect the condition at an early stage to take necessary precautions through medication to prevent its progression, leading to a better quality of life for those affected by it. In this work, we propose an approach for knee osteoarthritis classification using fine-tuned deep learning models. We employed the concept of transfer learning by utilizing three pre -trained EfficientNet models: EfficientNetB5, EfficientNet-B6, and EfficientNet-B7. By customizing the layers of the model, transfer learning enables us to use prior knowledge and models to enhance the performance of new tasks with limited data. The proposed system aims to automate the KL grading process by analyzing knee X-rays and classifying them into one of the five grades using fine-tuned EfficientNet models. Each model's performance is evaluated. The experimental results show that the EfficientNet-B7 model achieved the highest accuracy of 78.53%, while EfficientNet-B5 and EfficientNet-B6 attained accuracies of 75.14% and 76.47%, respectively.