GoogleNet-Based Classification System for Knee Osteoarthritis Detection in Radiographic Images

Knee osteoarthritis (OA) is a prevalent degenerative joint disease significantly impacting quality of life, particularly among older populations. This paper presents a sophisticated deep learning-based approach for identifying and categorizing knee OA. Radiographic images were processed with Faster R-CNN for precise localization of pathological features and GoogleNet for classification into four grades of severity. A dataset of 1650 manually annotated grayscale X-rays was used, achieving an overall classification accuracy of 91.9%, with grade-specific accuracies as follows: Grade 1 (90.57%), Grade 2 (90.63%), Grade 3 (93.71%), and Grade 4 (96.11%). The model exhibited high sensitivity across all grades, reaching 93.7% for Grade 3, and specificity values remained strong, peaking at 99.3% for Grade 4. Error rates ranged from 3.89% to 9.43%, while F1 scores demonstrated reliable performance, averaging above 91%. This automated method improves efficiency and lowers diagnostic subjectivity, particularly in advanced cases where the model showed its highest performance. The system's capacity to reduce diagnostic workloads and facilitate individualized treatment plans is highlighted by its telemedicine adaptability and clinical environment scalability. To improve musculoskeletal illness management and increase reliability in real-world applications, future improvements will concentrate on increasing classification accuracy for intermediate grades.