CNN-Based Classification of Knee Arthritis Severity from X-ray Images

This research work represents a convolutional neural network (CNN) technique for knee arthritis classification from X-ray images. The knee arthritis X-ray images are classified into five classes: Normal, Doubtful, Mild, Moderate, and Severe. Resizing, normalizing, and furnishing the steps of pre-processing aided in enriching the dataset and generalizing the strength of the model. The CNN model was carefully designed with multiple convolutional, activation, and max-pooling layers, followed by dense layers with dropout regularization to alleviate excessive fitting. The model’s training and validation were done at an 80:20 split of the dataset, attaining a final test accuracy of $58 \%$. The study highlights the potential of CNNs in the classification of images, especially for detecting knee arthritis despite relatively moderate accuracy. The areas highlighted for refinement include the performance evaluation, comprising accuracy, precision, and F1-score. Further study will focus on enhancing the accuracy of the model across deeper architectures, including additional clinical data and integrating advanced techniques such as transfer learning. This study focuses on facilitating the generalization of diagnostic tool automation that aids clinical decision-making and enriches the outcomes of patients in the context of knee arthritis, promoting healthcare innovation, well-being, and reduced healthcare disparities.