A neuromusculoskeletal tracking method for estimating muscle forces in human gait from experimental movement data

Abstract

The research results contained in this dissertation relate to a novel approach to estimating individual muscle forces in human movement by exploiting typical experimental observations acquired in movement laboratories. A neuromusculoskeletal model is made to move as observed and exert the same forces on the environment as recorded in the laboratory. Electrical activity of muscles can also be used to guide the solution process such that in the end, the muscle activity of the model is in better agreement with these recordings while still producing the desired movement. The innovation of this process is the efficient combination of inverse and forward analysis techniques. These classical techniques combined with nonlinear control theory form the basis of a neuromusculoskeletal tracking methodology for systematically replicating human performance in a computer model. The purpose is to capitalize on the non-invasive nature of this methodology to extract internal information about muscle forces and subsequent bone and soft-tissue loads during human movement. This information is sought by orthopedic surgeons and movement scientists alike in order to determine the function of individual muscles and to understand what interventions/treatments may be the most effective at restoring function and comfort to their patients. This treatise has accomplished three primary objectives: 1) it provides the detailed development of a non-invasive method for estimating muscle forces that includes complete system dynamics and is computationally tractable; 2) performs a benchmark analysis to validate the increased accuracy and computational advantages of the tracking approach, and 3) applies neuromusculoskeletal tracking to one of the most challenging problems in biomechanics, which is human gait simulation and analysis. In reaching these objectives four principle findings were made. 1) Tracking has provided results that are superior to previous dynamic optimization methods and at 3 to 4 orders of magnitude savings in computational costs, with the relative savings increasing with model complexity. 2) When random and systematic error/noise is present in kinematic data (due to skin movement, sampling, environmental interference, and data processing techniques), then ground reaction forces are better predictors of the true movement of the system. Under these circumstances, closely tracking experimentally estimated model kinematics is insu cient to demonstrate movement accuracy and ground reaction forces must be closely duplicated to indicate accuracy. 3) Because of its relative speed, neuromusculoskeletal tracking has proven to be a powerful validation tool since poor results or even tracking failure occurs if the model is not adequately representative of the subject data. Therefore, models must be evolved until the desired accuracy is obtained. 4) Controller weightings can further improve simulation accuracy by tracking certain reference data (such as ground reaction forces) more closely than others (i.e. motion of the toes). However, obtaining the set of weightings that balance tracking accuracy across multiple references is not a trivial task especially when there are a large number of reference signals to consider. Although improvements in tracking accuracy can be obtained by the optimization of weightings, they may not justify the high computational cost.