Patient-Specific Neuromusculoskeletal Models for Improving the Effectiveness of Human-Inspired Gait Rehabilitation Robots

Abstract

Rehabilitation robots are widely used to assist patients with neurological disorders in performing exercise tasks. Compared with physiotherapists, gait rehabilitation robots seldom tire, are able to accomplish gait retraining precisely, and provide quantitative feedback to show the movement patterns of the patients. However, the effectiveness of robot treatment methods is still under debate. Current gait rehabilitation robots are limited in the following two aspects. Firstly, they do not have the knowledge to account for patient variability. Secondly, they do not take into account the patient’s intention and engagement in the training. Therefore, this study investigated the use of biomechanical methodologies, including gait analysis technique, threedimensional musculoskeletal modelling and simulation technique and muscle force estimation methodologies to enhance the effectiveness of gait rehabilitation robots. This study aims to develop neuromusculoskeletal models, which include the patientspecific musculoskeletal properties and model the patients’ effort in muscle level. Two new models are developed in this research: the patient-specific muscle force estimation model (PMFE) and the patient-specific electromyography (EMG)-driven neuromuscular model (PENm). The PMFE and the PENm predict joint moment and muscle forces through kinematic information and EMG signals, respectively. The PMFE improves traditional inverse dynamic-static optimization model by realizing real-time calculation and ensuring good model prediction accuracy. Besides employing patient-specific musculoskeletal model for accurately modelling, the PMFE employs an analytical algorithm, the Lagrange multiplier method, in the static optimization procedure for real-time calculation. The musculoskeletal model is also simplified to one extensor and one flexor muscle around hip and knee joint for realtime calculation. The PMFE is evaluated by comparing the joint moments and individual muscle forces calculated via the PMFE and the computed muscle control method for healthy adolescents. Results show that the PMFE calculated joint moments and individual muscle forces accurately. As a case study of the PMFE, a patient-specific biological command based controller (PSBc) is developed based on the PMFE to control a human-inspired exoskeleton. The simulation and real-world experiment results show that the exoskeleton is controlled by the proposed PSBc with good accuracy. The second model, the PENm makes the following improvements for predicting individual muscle forces accurately in real-time. Firstly, the PENm incorporates EMG signals from two muscles around knee joint and using minimum musculotendon parameters in the model optimization process. Secondly, a dynamic computational model is developed based on Zajac’s computation flowchart to ensure the PENm predict muscle force in real-time. Thirdly, the PENm is based on a simplified patient-specific musculoskeletal model, which provides accurate patientspecific musculotendon parameters and muscle kinematics parameters. Fourthly, a combined force-length-velocity relationship is implemented to generate accurate muscle forces. The PENm is evaluated by comparing the joint moment and muscle forces via the PENm and the inverse dynamics and EMG activations for both healthy and cerebral palsy adolescents. Results show that the PENm can predict accurate joint moment in real-time. The PENm also provide more in-depth information on muscle functions. In summary, the design of gait rehabilitation robotic control strategies and clinical gait assessment can benefit from applications of the proposed biomechanical models. This research has collaborated with Department of Exercise Science and Shanghai Sunshine Hospital. The thesis has been published in two peer-reviewed SCI journals and presented at three international conferences.