Kinetic gait analysis using inertial motion capture. New tools for knee osteoarthritis
Angelos Karatsidis is a PhD student in the Biomedical Signals and Systems (BSS) group. His supervisors are prof.dr.ir. P.H. Veltink from the faculty Electrical Engineering Mathematics and Computer Science (EEMCS) and prof.dr.ir. J. Harlaar from Delft University of Technology / VUMC Amsterdam
Human movement analysis is an important field enabling us to understand the nature of several musculoskeletal disorders. For instance, altered mechanical loading of the affected knee joint has been demonstrated as a significant risk factor of knee osteoarthritis, leading to high rates of disability in the elderly population. In this context, assessing loading is crucial to develop new, patient-specific, conservative, non-pharmacological treatments for the effective management of joint pathologies. However, to date, analysis of human movement in terms of both kinematics and kinetics is conventionally performed within laboratory environments, demanding optical motion capture and force plate systems.
Despite the high performance of laboratory systems, their spatial restrictions, increased costs, operation complexity, and time are important limitations, obstructing the translation of state-of-the-art biomechanics research into routine clinical practice. Therefore, there is an emerging need for new tools for biomechanical analysis capturing both kinematics and kinetics in a flexible and convenient, but still accurate and reliable manner. To that end, inertial measurement units have recently evolved into a robust technique to capture human movement kinematics in any environment, but highly ambulatory assessment of kinetics remains a challenge.
In this dissertation, performed within the KNEEMO EU-funded Initial Training Network, we propose the use of high-end inertial motion capture systems to estimate human body kinetics, as well as to provide real-time biofeedback to the subject.
In Chapter 1, we introduce the history, definitions, and existing methods to obtain human movement in terms of kinematics and kinetics. Next, we present the clinical relevance of this research for knee osteoarthritis and the specific research objectives and questions.
In Chapter 2, we propose a method to estimate the external ground reaction forces and moments during gait tasks using only inertial measurement units. The reconstructed kinematics available through a commercial inertial motion capture system together with anthropometric measurements are used to as input in a whole-body inverse dynamics approach, deriving the total external force and moment applied on the body. A known problem in such approaches is the indeterminacy occurring between the feet in the bipedal phase. An empirical gait event-driven function called smooth transition assumption has been utilized to solve this problem. We compared the ground reaction force and moment estimates derived from either inertial or optical motion capture input against a gold standard force plate reference. The agreement between the estimated and reference GRF&M was categorized over a self-selected walking speed as excellent for the vertical, anterior and sagittal GRF&M components and as strong for the lateral, frontal, and transverse GRF&M.
In Chapter 3, we developed and validated the afore-mentioned method further in estimating knee joint moments during gait modifications. The examined gait modifications have been proposed to alter the knee joint loading, reflected by the joint moments, and therefore are assumed to be ways to unload the knee joint without external orthotic devices. Our analysis focused on the estimation of knee adduction and flexion moments assessed with the use of inertial sensors only, and compared these estimates to the ones obtained from a conventional inverse dynamics approach using optical motion capture and force plates. The accuracy analysis showed that moments obtained from the proposed method presented moderate to strong correlations for the knee adduction and flexion moment, over the stance phase. In addition, knee adduction moment changes from a baseline due to gait modification were quantified with mean differences less than 0.2% of body weight times body height between the proposed and reference approach in toe-in, toe-out, and lateral trunk lean walking patterns. These findings are encouraging for the further clinical adoption of inertial motion capture systems in assessing kinetics.
Chapter 4 takes a step towards integrating existing highly detailed musculoskeletal models with inertial motion capture. Recent techniques in musculoskeletal modeling allow the estimation of loads from motion while distributing the loads across a number of muscles using fatigue minimization principles. The significant advantage of such a technique is that it can be applied in several activities without dependence on a database. Available musculoskeletal models typically operate using reduced number of degrees of freedom and different segment definitions with respect to what commercial inertial motion capture systems use. Therefore, a method to reduce the number of degrees of freedom is proposed based on solving over-determinate inverse kinematics of a few virtual markers driven by the inertial model. For evaluation we compared the lower limb kinematics and kinetics of three different systems: 1) inertial motion capture with predicted ground reaction forces 2) optical motion capture with predicted ground reaction forces and 3) optical motion capture with measured ground reaction forces. The sagittal plane joint angles of ankle, knee, and hip presented excellent Pearson correlations and root-mean-squared-differences of about 5 degrees, respectively showcasing the robustness of inertial motion capture to assess joint angles. In addition, the ground and joint reaction forces and moments from the first and second kinematically-driven models provided in most cases similar performance to the third reference model demonstrating that inertial motion capture can be considered as an input option for musculoskeletal models and kinetics predictions outside laboratory environments.
In Chapter 5, we investigate the second objective of the thesis, focusing on a wearable biofeedback system for gait retraining applications. The foot progression angle has been selected as the most conveniently assessable and effective parameter to alter knee joint moments. A visual feedback system has been implemented with the use of augmented reality. The recent advances in this field allow accurate placement of virtual objects in the environment of use. Using the capabilities of the device we adjust the color and spatial properties of the objects in such a way that they convey information about the performed and targeted movement in real-time. The wearable system tracked FPA with an accuracy of 2.4 degrees RMS and intra-class correlation equal to 0.94 across all target angles and subjects, when compared to an optical motion capture reference. In addition, the effectiveness of the biofeedback, reflected by the number of steps with FPA value ± 2 degrees from the target, was found to be around 50% in both wearable and laboratory approaches. These findings demonstrate that retraining of the FPA using wearable inertial sensing and visual feedback is feasible with effectiveness matching closely an established laboratory method. The proposed wearable setup may reduce the complexity of gait retraining applications and facilitate their transfer to routine clinical practice.
Finally, in Chapter 6, we described the major insights of this research, including the limitations and sources of error, and provided directions for future development, validation and use of the proposed techniques. The research described in this thesis is only a first step to assess full kinetics using inertial sensors and several improvements should be introduced to reach clinical or daily life use. Reducing the number and size of sensors, as well as finding alternative power sources would facilitate the unobtrusive use. Regarding performance, methods to improve the sensor-to-segment calibration and scaling of the biomechanical models should be found. As for the biofeedback application, multimodal feedback should be explored to maximize the effect of gait retraining applications.
