Predicting Center of Gravity Displacement During Walking Using a Single Inertial Sensor and Deep Learning Technique

Abstract

Center of gravity (CG) trajectory during walking has been used as an important indicator to understand the mechanism of pathological walking. Kinematic method using optical motion analysis system and force plate method using ground reaction force are conventionally used to calculate CG trajectory during gait. However, there is a limitation in that expensive experimental equipment is needed to measure it. The purpose of this study was to develop a long short-term memory (LSTM) network model that could estimate the CG trajectory during gait based on a low-cost portable inertial sensor signal. Twenty healthy adult males (age: 24.2±2.1 years; height: 171.4±5.2 cm; weight: 67.5±7.4 kg) who had no history of musculoskeletal disorders were participated in the study. Based on the optical motion capture system, CG trajectories have been calculated using highly accurate kinematic method. Input variables of the LSTM model consisted of a total of six signals of three-axis acceleration and angular velocity. Data were divided into training/validation/testing data at a ratio of 80/10/10. Performance was evaluated utilizing 10-fold cross-validation. As a result of model estimation, the Pearson correlation coefficient of about 0.9 or more and root mean square error values ranging from 3.9 to 31 mm were obtained in anterior/posterior, proximal/distal, and medial/lateral directions. These results confirmed that the CG trajectory during walking could be estimated by using low cost, small wearable inertial sensor. This novel CG prediction method has potential to evaluate pathological gait by incorporating patient walking data.