Cluster Analysis of Biomechanical Gait Data and Pain Score as a Potential Classification of Severity in Knee Osteoarthritis

Osteoarthritis (OA) is the most common type of arthritis affecting approximately 240 million people globally, with increasing prevalence with age. The knee is the most prevalent joint affected by OA and it causes physical disability and decreased motor function which consequently affects the activity of daily living including mobility. Pain is the main symptom that is characterized in OA, which is measured using self-rated scales or questionnaires to determine several aspects of the pain including the intensity, frequency, and pattern. Quantifying pain is a standard clinical practice to diagnose and monitor symptomatic OA, however, its application for severity assessment is not well explored. To date, the severity assessment of knee OA is only by radiographic severity assessment that does not necessarily reflect the symptomatic OA. In this study, gait analysis was performed on symptomatic knee OA patients. Distinctive gait kinematic features were extracted using principal component analysis (PCA). Pain score and the gait features including spatiotemporal and kinematics were used for clustering analysis. Two clustering algorithms, K-means and K-medoids were conducted to cluster samples with similar features to assess knee OA characterization. The clustering solutions were evaluated based on three measures which are the Davies Bouldin index, Calinzki Harabasz index, and Silhouette index. This study discovered that majority of the datasets which is 5 out of 9 datasets had the best performance (fulfill at least 2 out of 3 performance index criteria) when the number of clusters, k is 4 and using the k-means algorithm. These clustering models can be used in the future as the labeling class of symptomatic knee OA that is based on pain and gait characteristics of knee OA. Future studies are suggested to test other pain assessment scores, include other gait features such as kinetic and muscle activity features, and employ various types of feature selection methods to improve the clustering performance.