A Data-Driven Approach for Estimating Postural Control Using an Inertial Measurement Unit

In this paper, we propose a probabilistic multi-Gaussian parameter estimation technique which addresses the complex relationship between acceleration and ground force signals used to derive a human’s static center of pressure. The intent of this work is to develop an accurate accelerometer-based method for determining postural control and neuromuscular status which is more portable and cost-effective than force plate-based techniques. Acceleration data was collected using an inertial measurement unit while ground reaction forces were simultaneously measured using a force plate. Various metrics were calculated from both sensors and probabilistic data models were built to characterize the relationships between the two sensors. These models were used to predict force-based postural control metrics corresponding to observed acceleration metrics. Data collected from one participant was used as a training set to which the test data of two individuals were then applied. We conclude that converted acceleration-based metrics on average can accurately predict all the corresponding force-based metrics we studied here. Furthermore, the proposed multi-Gaussian parameter estimation approach outperforms a more basic linear transformation technique for 75% of the metrics studied, as evidenced by an increase in correlation coefficients between true and estimated force plate metrics.