Analysis of Knee Osteoarthritis Severity: A Deep Learning Paradigm for Automated Assessment from Plain Radiographs

Knee osteoarthritis (OA) is a chronic disease, also known as progressive joint disease which occurs due to continuous loss of cartilage. It generally affects the older person with age above 45 years. It is also known as “wear and tear” joint disease because it mostly affects the joints of humans like knee, hand, hips and spine. Osteoarthritis symptoms frequently grow over the time and result in worsening of articular cartilage. Pain, stiffness, tenderness, swelling, and bone spur are most common signs of osteoarthritis. Doctors normally use radiographic X-ray images for identification of OA. In medical field this is a highly time-consuming process; hence, to automate this process and rapidly diagnose, deep learning algorithms such as convolutional neural networks (CNN), are employed for the identification of knee osteoarthritis (OA) severity in X-ray scans, using the Kellgren-Lawrence (KL) grading scheme. The dataset utilized for this research was sourced from Kaggle, consisting of knee X-ray images. The dataset was partitioned into a training set including 5778 images, a testing set including 1661 images, and a validation dataset containing 826 images. The proposed study encompasses the quantification of grade-specific severity in knee osteoarthritis. The measured rates for each severity grade are G0 = 91%, G1 = 1%, G2 = 49.77%, G3 = 67.85%, and G4 = 88.46%.