Knee osteoarthritis detection an AI driven severity estimation

In this work, we use the Kellgren and Lawrence (KL) grading system to investigate deep learning models or the automated detection and classification of knee osteoarthritis (OA) from radiographs. For prompt care, early identification of knee arthritis is essential, and radiography is an affordable, widely available diagnostic method. We evaluated the accuracy, precision, recall, and F1 score of a number of cutting-edge models, including VGG16, Xception, DenseNet121, InceptionV3, ResNet50, and EfficientNet07. With testing accuracy of 96.90% and 95.90%, respectively, and strong F1 scores of 0.9690 and 0.9790, Xception and InceptionV3 outperformed the other models under evaluation, demonstrating robust performance and trustworthy generalization. While ResNet50, EfficientNet07, and an ensemble model had lower accuracy and F1 scores, indicating possible limitations, DenseNet121 and VGG16 also shown strong performance in dealing with the complex knee arthritis grading system. According to our research, Xception and InceptionV3 are very good at automatically classifying knee OA, providing potential instruments for clinical treatment planning and early identification.