Advancements in Automated Classification of Knee Osteoarthritis Severity using EfficientNetV2-S

Automated classification of knee osteoarthritis severity from MRI images is essential for timely diagnosis and treatment planning. Traditional methods relying on manual grading are subjective and time-consuming, necessitating automated solutions for accuracy and efficiency. Previous studies have employed various deep learning architectures and traditional machine learning approaches. These include AlexNet, KL-3 grading, EfficientNet-B5, Adamax optimization, and ResNet101 DNN, achieving accuracies ranging from 69% to 93.24%. Current methodologies suffer from limited accuracy, scalability issues, and dependency on specific imaging modalities or grading systems. They often require extensive manual feature extraction and struggle with generalization across diverse datasets. To address these limitations, our proposed model utilizes EfficientNetV2-S with custom learning rate adjustment. This model leverages transfer learning from ImageNet, enhancing feature extraction capabilities and improving generalization across varied MRI datasets of knee osteoarthritis. EfficientNetV2-S is selected for its superior performance in balancing computational efficiency and accuracy. The model is fine-tuned with a custom learning rate schedule to optimize convergence and robustness, enabling precise classification of knee osteoarthritis severity. Experimental results demonstrate the effectiveness of our approach, achieving a training set accuracy of $98.47 \%$. This outperforms existing methodologies, highlighting the potential of our model in clinical applications for accurate and automated knee osteoarthritis severity classification from MRI images.