Methods

A framework in Figure 1 outlines the entire method of this study from the data collection to the estimation model development. Due to an importance of initial loading characteristics of TKA population [6], gait variables related to the initial loading behavior were carefully considered when both biomechanical and inertial gait variables were selected. Feature selection was performed to obtain statistically meaningful inertial gait variable subsets, and hierarchical linear regressions were created to determine directional contributions of inertial gait variables to the key biomechanical gait variables of the TKA population.

A. Data Description

Data were acquired in the Neuromuscular Biomechanics Laboratory at the University of Delaware. An instrumented split-belt treadmill (Bertec Corp, Columbus, Ohio) and an 8- camera motion capture system (Motion Analysis Corp, Santa Rosa, CA) were used to collect kinetic and kinematic gait variables. Concurrently, two ankle-worn accelerometers (Noraxon USA, Scottsdale, AZ) were attached above the lateral malleoli using elastic bands to collect three-dimensional acceleration data. The sampling frequency of acceleration data was 200 Hz. For the left leg-worn sensor, the X-axis pointed up to shank, the Y-axis pointed forward, and the Z-axis pointed away to the left. The X-axis of the right sensor pointed up to the shank, the Y-axis pointed backward, and the Z-axis pointed the outward direction. The biomechanical data from the force place, and the cameras and inertial data from the wearables, were synched via hardware trigger. Biomechanical gait variables and inertial gait variables were computed by using Visual 3D (C-Motion, Bethesda, MD) and custom software developed in the MATLAB 9.0 (Mathworks, Natick, MA) environment, respectively.