Automated Knee Osteoarthritis Severity Classification Using ResNet: A Deep Learning Approach for X-ray Analysis

In this work, the ResNet-based CNN was applied to the classification of knee OA using X-ray images. Osteoarthritis is a crippling degenerative joint disease that affects millions worldwide; hence, the importance of early diagnosis for effective management. A total number of 1,656 knee X-rays were utilized in order to train the ResNet model, which represents grades from 0-a healthy one-to 4-severe osteoarthritis. The classification was performed according to clinical criteria that define the development of osteoarthritis by characteristic radiological features, including narrowing of joint space and formation of osteophytes. Overall accuracy for the model performance on the test set at sight is 94.57%. The classification report emphasizes mostly excellent performance over most grades. In Grade 0, which means healthy knees, the precision is 0.99, recall is 0.93, and F1-score is 0.96. This was done with Grades 1, 2, and 3, which returned F1-scores of 0.96, 0.95, and 0.94, respectively. In Grade 4, that is, severe OA, the performance was far worse, with a precision of 0.66 and an F1-score of 0.77, although this can be expected given to the smallest sample size in this category. These results further develop the conclusion that ResNet performs effective automation of knee osteoarthritis classification based on X-ray images with strong accuracy and robustness for most of the severity grades. This approach is a potentially useful tool to help diagnose and monitor osteoarthritis, especially in resource settings where access to expert radiological assessment is limited.