A Sequential VGG16+CNN-Based Automated Approach With Adaptive Input for Efficient Detection of Knee Osteoarthritis Stages

Abstract

Osteoarthritis (OA) is a prevalent musculoskeletal disorder, predominantly affecting the knee joint and resulting in significant pain and functional limitations. The Kellgren-Lawrence (KL) grading system, traditionally utilized by radiologists, assesses OA severity through radiographic evaluation of knee bones on both sides. Recent advancements in computer-aided methods have focused on enhancing diagnostic precision and efficiency, leveraging automated classification models with Knee X-rays which represent a cornerstone imaging modality for OA severity assessment. In this study, a novel hybrid model combining Convolutional Neural Networks (CNN) and VGG16 architectures is proposed to achieve accurate OA detection. Various neural networks, including CNN, VGG16, VGG19, ResNet50, and CNN-ResNet, are implemented and compared alongside the proposed method. Additionally, data augmentation techniques are employed to address class imbalance, leading to enhanced accuracies across all models. Analysis demonstrates robust performance of all models on the training set. The proposed hybrid method (CNN-ResNet50) outperforms other state-of-the-art approaches, providing precise results across all five stages of OA according to the KL grading system by achieving an accuracy exceeding 93% on training, validation, and testing datasets.

Keywords

• Knee osteoarthritis
• X-Ray
• Kellgren and Lawrence grading
• VGG16/VGG19
• ResNet50
• convolutional neural network (CNN)