Assessing Knee Osteoarthritis Severity and Biomechanical Changes After Total Knee Arthroplasty Using Self-organizing Maps

Abstract

This study aimed to develop an unsupervised Self-Organizing Map (SOM) based framework to map variability in longitudinal Osteoarthritis (OA) gait biomechanics, and characterize progression pathways within the SOM. Three-dimensional knee kinematics and kinetics observations of asymptomatic (n = 236), moderate knee OA (n = 341), severe knee OA (pre Total Knee Arthroplasty (TKA); n = 145) and post-TKA (n = 201) gait were collected. Principal component analysis (PCA) was applied to frontal and sagittal knee angle and moment waveforms, resulting in an uncorrelated PC score dataset describing 95% of gait waveform variability. PC scores, spatiotemporal gait, and demographic features were applied to the SOM, followed by hierarchical clustering. Clusters were validated by examining inter-cluster differences by chi-squared, k-way ANOVA and Kruskal Wallis tests. OA clinical severity transitioned from mostly asymptomatic to mostly severe across the SOM’s x-axis. Age and BMI increased, and gait speed decreased across the SOM. This coincided with worsening knee biomechanics, captured by reduced flexion angle magnitudes, reduced stance-phase flexion moment range, and reduced knee adduction moment mid-stance unloading. Three clusters within the SOM were characterized as 1) High Function Gait; 2) Low Function Gait; and 3) Moderate Function Gait. Knee biomechanics during OA gait can be characterized using SOMs to provide a multidimensional interpretation of gait biomechanics severity pathways. Longitudinal changes in individual SOM location can provide insight into OA progression pathways, with utility to support interventions targeting current or predicted individual functional needs.