Machine learning for clustering and classification of early knee osteoarthritis using single-leg standing kinematics

Detection of early osteoarthritis (EOA) of the knee is crucial for effective management and improved outcomes. This study investigated the application of machine learning techniques to single-leg standing (SLS) kinematics to classify and predict EOA. (1) To identify distinct groups based on SLS kinematic patterns using unsupervised learning algorithms, (2) to develop supervised learning models to predict EOA status, and (3) to identify the most influential kinematic variables associated with EOA.

Total 43 manufacturing workers (86 legs) aged 40-70 years were evaluated. The participants were categorized using an Early Osteoarthritis Questionnaire. Single-leg standing kinematics was captured using 2D video analysis to assess the horizontal displacement of six key anatomical points (trunk, pelvis, femur, knee, lower leg, and ankle) in the frontal plane. K-means clustering was used for unsupervised learning, whereas six supervised machine learning algorithms were trained and validated for EOA classification.

In our machine learning models, we used 258 data points derived from three repeated measurements per participant. K-means clustering revealed three distinct groups based on SLS kinematics and demographic characteristics. The random forest algorithm achieved the highest classification accuracy (area under the receiver operating characteristic curve = 1.000, accuracy = 1.000) in distinguishing between individuals with and without EOA. Pelvic and ankle horizontal displacements were identified as the most influential predictors of EOA classification.

Machine learning analysis of SLS kinematics shows significant potential for the early detection of knee osteoarthritis. Identification of key kinematic predictors, particularly pelvic and ankle movements, provides new insights into targeted interventions and screening protocols for rehabilitation.