Enhancing Orthopaedic Diagnosis through Convolutional Neural Networks for Knee X-ray Analysis

Abstract

Knee Osteoarthritis is a prevalentillness that requires accurate detection and severity assessment for effective treatment and prognosis. The Kellgren-Lawrence (KL) grading scale, commonly accustomed evaluate abnormalities in X-ray scans, is susceptible to individual variability in grading. Early identification can potentially delay the development of osteoarthritis in knees (KOA). Most of the existing methods prefer ensemble models to improve the accuracy of KOA grade classification, even though it requires more computational cost and space.To address this issue our workobjectives to develop AI driven modelwhich focus on earlier prediction of knee osteoarthritis and bridge the gap between the structural changes in knee joint and symptoms of the patient. This study utilizes an innovative method using YOLO models to train, detect and localize the specific structural changes in knee radiographs, which are indicative of severity grades in association with KL grading system.For automated classification OAI dataset is used and the model's functioning is verified using ten-fold cross-validation. In a five-class OA assessment test, the anticipated technique attains 98.43% mean accuracy, outperforming existing methods. Notably, the study finds that fewer multiclass labels yield better performance, with binary classification reaching an accuracy of 98.2. This research highlights the potential of AI analytics in enhancing treatment and prognosis by accurately predicting reduction in the space between joints (JSN). It advocates for the use of CNNs in KOA detection and severity assessment, using the KL grading system (Grade-0 to Grade-4). The findings advocate for transparent CNN-based analysis techniques, urging medical researchers and developers to adopt AI-assisted tools in clinical settings to enhance KOA diagnoses.