Management and Engineering, University of Padua, Vicenza, Italy ... Methods: Custom software was written in C++ to read 3D markers trajectories, ground ...
Modelling SS-0071 REAL-TIME ESTIMATION OF KNEE JOINT CONTACT FORCES DURING WALKING USING OPENSIM AND A CALIBRATED EMG-DRIVEN NEUROMUSCULOSKELETAL MODEL Claudio Pizzolato 1,*Monica Reggiani 2Luca Modenese 1David G. Lloyd 1 1Centre
for Musculoskeletal Research, Griffith Health Institute, Griffith University, Gold Coast, Australia, 2Department of
Management and Engineering, University of Padua, Vicenza, Italy Introduction and Objectives: Real-time estimates of joint contact forces and musculotendon unit (MTU) forces can potentially enable rapid patient evaluation and biofeedback for gait retraining. MTU forces can be estimated in real-time using our calibrated electromyography (EMG)-driven neuromusculoskeletal (NMS) model (CEINMS). This uses EMG signals and three-dimensional (3D) joint angles to drive a NMS model that produces instantaneous kinematics and kinetics of individual MTUs [4, 5]. Then, the MTU forces estimated by CEINMS can be used with external joint moments to calculate the knee joint contact forces [3]. While rectified and filtered EMG signals can be directly used as input for CEINMS, joint angles and moments are estimated from motion capture data using inverse kinematics (IK) and inverse dynamics (ID) algorithms in OpenSim [1]. This is time consuming and currently precludes real-time estimation. However, using OpenSim to calculate 3D joint angles is essential to ensure consistency between motion analysis and musculoskeletal models used to estimate MTU kinematics. Furthermore, OpenSim permits personalisation of musculoskeletal models, created from imaging data to reflect an individual’s specific anatomy. Therefore, we aimed to i) develop real-time OpenSim IK and ID procedures, ii) integrate these with CEINMS, and iii) compare the real-time estimates of 3D joint angles and moments, and knee joint contact forces to those from offline processing. Methods: Custom software was written in C++ to read 3D markers trajectories, ground reaction forces (GRF), and EMG signals from a Vicon motion capture system in real-time. GRF, rectified EMG and 3D joint angles were filtered with a 6Hz low pass 4th order Butterworth filter, and processed using OpenSim IK and ID routines, which were modified to run on frame-by-frame. Multiple processing threads were used to increase the data throughput. MTU forces from CEINMS and external joint moments from ID were finally used to calculate medial and lateral knee joint contact forces [3]. CEINMS was calibrated in the background, on additional processor threads, by adjusting selected muscle parameters to reduce the error between joint moments estimated by CEINMS and ID. The real-time outputs were visualised in real-time using a Graphic User Interface (GUI). From a single subject walking on a treadmill an 8-camera Vicon system collected motion data of a full body marker-set (68 markers) at 100Hz, and GRF and surface EMG signals from 16 muscles of a single leg at 1000Hz. Segmental dimensions and muscle parameters of the OpenSim model were anthropometrically scaled to the subject. The scaled model was then used to compute the set of multidimensional cubic B-splines that characterised the subject MTU kinematics [6] in CEINMS. During the walking trials the real-time software was used to produce and save 3D joint angles and moments from hip, knee, and ankle, and knee joint contact forces. The marker trajectories were also saved for subsequent offline
1310
www.isbglasgow.com
processing. The real-time and offline pathways were then compared using a modified coefficient of multiple correlation (CMC), which assessed the similarity of waveforms [2]. Results: From 6 consecutive gait cycles we found similar waveforms between the offline and real-time processing for hip, knee and ankle 3D joint angles and moments, with CMCs always greater than 0.95 (Table 1). While the computation time was about 30ms, the use of multiple threads enabled the results to be produced at the same frequency as the marker trajectory data (100Hz). The real-time estimation of knee joint contact forces is now being evaluated. Conclusion: Our real-time system produced estimates of the 3D joint angles and moments nearly identical to those determined using offline processing. The small variations present were in the filtering and differentiation of 3D joint angles in the offline method that should not affect joint contact forces estimates. This is the first software to calculate, in real-time, 3D joint angles and moments using OpenSim coupled with an individually scaled and calibrated EMG-driven NMS model. Our framework enables real-time estimation MTU and joint contact forces that can be used for gait retraining biofeedback. Table:
Caption: Table 1: CMC for 3D joint angles and moments between offline and real-time processing from different joints and degrees of freedom (1 subject, 6 gait cycles). FE = flexion extension, AA = adduction abduction, IR = internal rotation. References: [1] Delp SL et al., IEEE Trans Biomed Eng, 54:1940-1950, 2007. [2] Ferrari A et al., Gait & posture, 31:540-542, 2010. [3] Gerus P et al., Journal of biomechanics, 46:2778-2786, 2013. [4] Lloyd DG et al., Journal of biomechanics, 36:765-776, 2003. [5] Sartori M et al., Plos One, 7:e52618, 2012. [6] Sartori M et al., Journal of biomechanics, 45:595-601, 2012. Disclosure of Interest: None Declared
@ISB_Glasgow
1311
