Phase Space Reconstruction and Euclidean Distance for Improving the Classification of Gait Patterns between Patients with Knee Osteoarthritis and Healthy Controls

This research introduces innovative methodologies to differentiate between the gait patterns of individuals with knee osteoarthritis (OA) and those without symptoms, a distinction achieved through the analysis of kinematic data dynamics. The employed methodology involves the reconstruction of phase space representations from the measured translations and rotations of the knee, thereby revealing the signatures of the underlying gait system through deterministic learning theory. The extraction of features is then carried out using phase space distances that amplify the differences between asymptomatic and OA knee patterns. In particular, a three-dimensional (3D) phase space model is developed in tandem with Euclidean distance metrics that are sensitive to the dynamic changes induced by the disease. The feature set derived from this process is subsequently used as input for classification models that label new gait trial data. Experimental trials conducted on data from 19 patients with OA and 28 asymptomatic subjects yielded an accuracy of 95.7% and 97.9% in two-fold and leave-one-subject-out cross-validation respectively, thereby demonstrating the effectiveness of these methods for characterization and diagnosis. Comparative evaluations further underscore the superior performance of these methods over existing ones. In essence, this study lays the groundwork for a dynamic framework that enables the data-driven quantification and detection of the impact of osteoarthritis on human walking patterns.