A Sequential VGG16+CNN based Automated Approach with Adaptive Input for Efficient Detection of Knee Osteoarthritis Stages

Osteoarthritis (OA) stands as the most prevalent musculoskeletal disorder, particularly affecting the knee joint and causing substantial pain and functional impairment. Radiologists traditionally employ the Kellgren–Lawrence (KL) grading system, analyzing radiographic evidence from both sides of the knee bones, to evaluate OA severity. Knee X-Rays are considered a golden imaging modality to analyse the severity of Osteoarthritis. Many computer-aided methods aimed at enhancing diagnostic accuracy and efficiency, leveraging advancements in automated classification models utilizing Knee X-Rays. These innovations hold promise for improving the diagnosis and management of OA. In this paper, A new model (hybrid of CNN and VGG16) is proposed to utilize strength of both architectures to obtain accurate results for OA detection. Different neural networks (CNN, VGG16, VGG19, ResNet50, CNN-RestNet) are implemented and compared with the proposed method. Additionally, data augmentation contributes to enhanced accuracies across all models by resolving class imbalance problem. It is analysed that all models performed well on training set however, for testing set, the proposed hybrid method (CNN-ResNEt50) outperformed other state-of-the-art methods and produce accurate results for all five stages of OA using KL grading method. The proposed method obtained above 93% accuracy for training, validation and testing data.