EMG-Informed neuromuscular model assesses the effects of varied bodyweight support on muscles during overground walking

Abstract

Bodyweight supported walking is a common gait rehabilitation method that can be used as an experimental approach to better understand walking biomechanics. Neuromuscular modeling can provide analytical means to add insight into how muscles coordinate to produce walking and other movements. We used an electromyography (EMG)-informed neuromuscular model in OpenSim to investigate changes in muscle parameters (muscle force, activation and fiber length) at varying bodyweight support levels: 0%, 24%, 45% and 69% bodyweight. Coupled constant force springs provided a vertical support force while we collected biomechanical data (EMG, motion capture and ground reaction forces) from healthy, neurologically intact participants walking overground at 1.2 m/s. The lateral & medial gastrocnemius both demonstrated a significant decrease in muscle force (p = 0.002 & p < 0.001) and activation (p = 0.007 & p <0.001) through push-off at higher levels of bodyweight support. The soleus, in contrast, had no significant change in muscle activation through push-off (p = 0.652) regardless of bodyweight support level even though soleus muscle force decreased with increasing bodyweight support similar to the gastrocnemius. Compared to the other plantarflexors, there was a difference in soleus muscle fiber behavior, with shorter muscle fiber lengths and faster shortening velocities at greater bodyweight support levels. The bicep femoris actually had increased muscle force during mid-stance with greater levels of bodyweight support (p <0.001). These results provide insight into how muscle force is decoupled from effective bodyweight during walking with bodyweight support due to changes in muscle fiber dynamics.

Introduction

Neuromuscular modeling allows biomechanists to investigate changes in muscle dynamics that produce complex musculoskeletal movements [1]. Applications of neuromuscular modeling include designing better prosthesis control, evaluating disordered movement in patient populations, and investigating the dynamic behavior of musculotendon units noninvasively [2], [3]. OpenSim is a commonly used open-source neuromuscular modeling and simulation software package with a library of previously developed and tested neuromuscular and musculoskeletal models [4]. Using experimental data from motion capture, ground reaction forces, and electromyography (EMG), it is possible to model changes in muscle fiber dynamics and muscle forces during human movement. The OpenSim toolbox Calibrated EMG-Informed NeuroMusculoSkeletal Modelling Toolbox (CEINMS) adds a layer of neural integration to biomechanical models and facilitates subject specific modeling [4]–[7].

Using a harness to provide bodyweight support during walking is an effective method to improve mobility in patients with neurological impairments [8]–[11] and can provide insight into fundamental biomechanical principles of human locomotion [12]–[15]. Although a treadmill is often used for bodyweight supported gait training, overground bodyweight supported walking allows the patient to control the speed, and allows the experimenter to easily introduce obstacles, turns, or stairs. All of these overground possibilities have some benefit for gait rehabilitation. There is growing evidence on how relatively constant bodyweight support during gait alters kinematics, kinetics, and muscle activity, but we have little insight into how muscle forces and muscle fiber dynamics are affected by bodyweight support during overground walking [15]–[20].

The goal of this study was to investigate changes in muscle force and muscle fiber dynamics across a range of bodyweight support levels during overground walking. To the best of our knowledge, no prior research has analyzed the effects of overground bodyweight support through computational neuromuscular modeling. Information about joint kinematics and ground reaction forces is helpful to understand how bodyweight support affects overground gait mechanics, but estimates of specific muscle forces and muscle fiber dynamics would provide valuable information about neuromuscular control under the bodyweight support conditions. Previous studies have examined changes in electromyography (EMG) amplitudes and net muscle moments about the joints for overground walking with bodyweight support [19], but it is not clear how muscle forces and displacements are affected by bodyweight support. Providing a better understanding of differential effects on muscle dynamics with bodyweight support could aid in optimizing gait therapies after musculoskeletal and neurological injuries (Zajac, F. E., Neptune, R. R., & Kautz, S. A. (2003). Biomechanics and muscle coordination of human walking: part II: lessons from dynamical simulations and clinical implications. Gait & posture, 17(1), 1–17; Pandy, M. G., & Andriacchi, T. P. (2010). Muscle and joint function in human locomotion. Annual review of biomedical engineering, 12, 401–433.)).

Methods

Experimental Design [19]

To measure biomechanics in a simulated reduced gravity environment, we fitted participants with a custom-built bodyweight support system and recorded biomechanical data as they walked overground. Our passive, dynamic bodyweight support system used coupled constant force springs as the actuation unit to deliver a near constant upwards force to the user during overground walking. The mechanical design for the bodyweight support system is described in MacLean et al. [19]. All 8 participants (6 females; 25±3 years of age, body mass 69±7 kg, mean±s.d.) recruited for this study provided informed consent as approved by the University of Florida’s Institutional Review Board. Participants walked on an 8 m walkway at 1.2 m/s, which was monitored by two pairs of infrared timing gates placed 2 m from the start and end of the walkway. Three in-line force plates (AMTI, Watertown, MA, USA) were embedded in the middle of the walkway and measured ground reaction forces. Participants walked with four bodyweight support levels (BW): 0% BW, 24% BW, 45% BW, and 69% BW and were given 3 minutes walking practice in each condition to habituate before we recorded data. We recorded surface EMG (Cometa, Bareggio, MI, Italy) of the tibialis anterior, soleus, medial gastrocnemius, lateral gastrocnemius, rectus femoris, vastus lateralis, vastus medialis, and biceps femoris. Both force plates and EMG data were recorded at 1000 Hz. We also used optical motion capture (OptiTrack, Orvallis, OR, USA) on the lower body to measure kinematics at 100 Hz. For each condition, we recorded 4 trials. More in-depth details of the experimental design can be found in MacLean et al. [19].

MOtoNMS [21]

We used the MATLAB (MathWorks, Natick, MA, USA) “Motion Data Elaboration Toolbox for NeuroMusculoSkeletal Applications” (MOtoNMS) to pre-process and prepare the experimental data for use in OpenSim. We filtered marker data and ground reaction forces with a zero lag, 10 Hz lowpass, 4th order Butterworth filter. EMG data were first filtered with a zero lag, 4^th order Butterworth bandpass filter with cut-off frequencies of 30 and 300 Hz. We then rectified the data and applied a zero lag, 6 Hz low pass, 4th order Butterworth filter. For each muscle and participant, EMG data were normalized to the maximum EMG activity recorded across all conditions. For musculoskeletal modeling, we calculated the ankle joint center as the mid-point of the medial and lateral ankle malleoli markers, the knee joint center as the mid-point of the lateral and medial femoral epicondyle markers, and the hip joint center using the Harrington hip joint method [22]. To increase the accuracy of our EMG-informed neuromuscular model we choose to use data from the stance phase only.

OpenSim Processing

We carried out our musculoskeletal modeling using OpenSim. We scaled an existing OpenSim model (gait2392 [23]) for each participant using their mass and motion capture data from a 3-second T-pose. We used inverse kinematics with the experimental marker data to find 3-dimensional joint angles for the ankles, knees, and hips at each frame. We then calculated the external moment about each joint using inverse dynamics on the calculated joint angles, measured ground reaction forces, and center of pressure. The built-in muscle analysis tool used joint moment and angle during the subject’s stance phase to solve for muscle fiber length, tendon length, fiber velocity and muscle moment arm.

CEINMS [6]

To estimate muscle-level mechanics from our experimental data, we used the CEINMS process. For the CEINMS calibration, we optimized the specific muscle parameters (optimal fiber length, pennation angle, tendon slack length, recursive coefficients and shape factor) for each musculotendon unit using the subject’s experimental data walking with zero supported bodyweight. We made standard assumptions about the contraction dynamics of musculotendon units such as: active force length curve, passive force length curve, force-velocity relationship, and tendon-force strain relationship. These assumptions were used within the CEINMS model to shape our modified-Hill type muscle model.

We refined our neuromuscular model analysis in CEINMS by customizing the execution program that dictated how to analyze muscle parameters given the input forces, moments, muscle activations, and joint kinematics. The execution program personalizes the intermediary processing steps within the CEINMS pipeline. These intermediary steps include the transition mechanism from neural to muscle activation and the physical response of muscles to changes in joint kinematics. We used the EMG-assisted CEINMS neuromuscular model, which employed a simulated annealing optimization [24] to improve the musculotendon unit responses using both the recorded EMG signals and joint kinematics. The equilibrium elastic tendon model simulated musculotendon unit forces with an exponential relationship from neural excitation to muscle activation, shown in the equations (1–2)[25]–[27].

$$u_j\left(t\right)= \alpha e_j\left(t-d\right)-{\beta }_1{u}_j\left(t-1\right)-{\beta }_2{u}_j\left(t-2\right)$$ (1)

u_j(t) = neural activation, α = muscle gain coefficient, e_j = j-th muscle excitation, β₁, β₂ = recursive coefficients, d = electromechanical delay

$$a_j\left(t\right)=\frac{e^{A_j{u}_j\left(t-1\right)}-1}{e^{A_j}-1}$$ (2)

a_j(t) = j-th muscle activation, A_j = non-linear shape factor

Analysis

CEINMS produced kinematic, kinetic, and musculotendon unit data which were normalized, grouped and averaged by bodyweight condition. We examined joint angles and net external moments for the ankle and knee to focus on the joints with the highest level of accuracy based on the experimental data collected. The CEINMS program normalized muscle fiber length to the muscle’s optimal fiber length from the subject calibration file. Likewise, MOtoNMS normalized muscle activation to the participant’s maximal muscle activation for each muscle. Then, we normalized muscle force to the participant’s body weight before averaging across trials for each bodyweight condition. After grouping and averaging the data, we split the data into three distinct phases: loading response, mid-stance and push-off. These three phases encapsulate important shifts in the subject’s musculotendon unit behavior during walking. They are defined by the first local maxima and the local minima in the subject’s vertical ground reaction force during stance phase, as shown in Fig 2. We segmented the data into the 3 phases and averaged across trials for all bodyweight conditions and participants.

Fig 2:

Stance phase was separated into three sections: loading response, mid-stance and push-off. These sections are separated by the first local maxima (F1) and minima (F2) in the subject’s vertical ground reaction forces.

The statistical analysis of the data investigated the effects of bodyweight support on each muscle and gait phase using repeated measures MANOVA and reported Wilk’s lambda for significance. Follow up univariate tests queried significance among response variables (muscle fiber length range, activation and force), Greenhouse-Geisser correction was applied. Finally, if the univariate test showed significance, we performed pairwise comparisons with a Bonferroni correction.

Results

Bodyweight support level substantially impacted ankle and knee, angles and external joint moments (Fig 3). Increased bodyweight support resulted in less ankle plantarflexion from 20–100% of stance phase, with the greatest decrease occurring at push-off. Ankle dorsiflexion moment decreased with increased support in the push-off phase. In the 69% bodyweight support condition, the ankle angle profile exhibited a much more plantarflexed joint angle compared to the 0% bodyweight support condition. The knee flexion angle decreased during the loading response and increased during the push-off with increased bodyweight support. During the initial loading response, knee joint extension moment magnitude greatly decreased with increased bodyweight support. In push-off, the higher levels of bodyweight support led to a shorter duration of knee flexion moment.

Fig 3: (.

Left) Ankle and knee external joint moment shown at 0% BW, 24% BW, 45% BW and 69% BW, standard deviation shown as shaded area. (Right) Ankle and knee angle shown at 0% BW, 24% BW, 45% BW and 69% BW, standard deviation shown as shaded area. Grey lines indicate F1 and F2 which split the gait cycle into loading response, mid-stance, and push-off.

The repeated measures MANOVA demonstrated a significant relationship between bodyweight support and the muscle parameters (force, activation and fiber length) for all muscles but the tibialis anterior (Fig 4). We found 3 of the 8 muscles were dependent on bodyweight support level in all 3 gait phases, and 4 muscles were impacted by bodyweight support level in 2 gait phases. The hip muscles recorded (bicep femoris & rectus femoris) did not have a significant relationship between bodyweight support and the muscle parameters for the push-off phase (Λ = 0.371 & Λ = 0.198). Similarly, the ankle plantarflexors (lateral & medial gastrocnemius) did not demonstrate a significant relationship during the mid-stance phase (Λ = 0.08 & Λ = 0.601). The tibialis anterior did not have a significant relationship in any of the three stance phases (loading response Λ = 0.42, mid-stance Λ = 0.259, push-off Λ = 0.081).

Fig 4:

Multivariate statistics repeated measures MANOVA (reporting Wilk’s Lambda), how bodyweight condition affects each muscle and phase for all response variables. LR is loading response, MS is mid-stance, and PS is push-off.

Of the 8 muscles studied, the predominate relationship in the 3 stance phases was a decrease in muscle force and activation with increased bodyweight support, although there were some instances where this relationship was not manifested and others in which the relationship was inversed, as seen in Fig 5.

Fig 5:

Univariate statistics, reporting the Greenhouse-Geisser sig. values. (Top) Muscle force, univariate statistics for how bodyweight condition affects each muscle and phase. (Middle) Fiber length range, univariate statistics for how bodyweight condition affects each muscle and phase. (Bottom) Muscle activation, univariate statistics for how bodyweight condition affects each muscle and phase.

The relationship between muscle activity and bodyweight support was generally inverse, with increased bodyweight support resulting in decreased muscle activity, but there were exceptions (Fig 5). The vastii (lateralis & medialis) showed a significant inverse trend during the loading response (p=0.001 & p=0.003). Similarly, the rectus femoris had a significant inverse trend in muscle activation during mid-stance (p=0.003). In contrast, the bicep femoris muscle activation had a direct trend during mid-stance, meaning that the muscle activation increased with bodyweight support (p<0.001). The lateral & medial gastrocnemius activations decreased with increased bodyweight support during the push-off phase (p=0.007 & p<0.001) (Fig 6). There was no significant change in muscle activity in response to bodyweight support level for the third plantarflexor muscle, the soleus, during push-off (p=0.652) (Fig 6). Lastly, the tibialis anterior showed no substantial change in muscle activity throughout stance phase.

Fig 6:

(Left) Soleus normalized force, normalized activation and normalized fiber length. (Middle) Gastrocnemius lateralis normalized force, normalized activation and normalized fiber length. (Right) Gastrocnemius medialis normalized force, normalized activation and normalized fiber length. Means shown with standard deviation as shaded error.

Similar to muscle activation, the relation between muscle force and bodyweight support was generally inverse, as bodyweight support increased muscle force decreased (Fig 5) (Fig 7). The lateral gastrocnemius, medial gastrocnemius and soleus all showed a significant decrease in muscle force with increased bodyweight support throughout push-off (p = 0.002; p <0.001; p <0.001) (Fig 6). Furthermore, the soleus also showed a decrease in muscle force during mid-stance (p <0.001) (Fig 6). On the other hand, there was no significant change in tibialis anterior muscle force during stance phase throughout bodyweight support levels. The vastii (lateralis & medialis) demonstrated a significant inverse trend between muscle force and bodyweight support during the loading response phase (p<0.001 & p<0.001). Similarly, the rectus femoris showed a significant inverse trend in muscle force for both loading response and mid-stance (p = 0.002 & p = 0.012). We found a slight increase in vastii muscle force during push-off between 24% and 45% BW. The bicep femoris exhibited a direct muscle force to bodyweight relationship in mid-stance, with muscle force increasing with bodyweight support (p <0.001).

Fig 7:

Upper leg muscles, vastus lateralis, vastus medialis, bicep femoris and rectus femoris shown as normalized muscle force mean±s.d. (error bars) for loading response, mid-stance and push-off. Lower leg muscles, lateral gastrocnemius, medial gastrocnemius, soleus and tibialis anterior shown as normalized muscle force mean±s.d. (error bars) for loading response, mid-stance and push-off.

Lastly, the muscle fiber length range decreased with increasing bodyweight support for all muscles. The relationship was significant for all muscles during loading response except the soleus and tibialis anterior (Fig. 5). During mid-stance, only the rectus femoris exhibited a significant relationship (p = 0.006) and for push-off only the lateral (p < 0.001) and medial (p < 0.001) vastus developed a similar trend.

Discussion

As expected, muscle force tended to decrease with increased bodyweight support, although there were some exceptions to this relationship. Muscle force was expected to decrease in stance phase with bodyweight support because the load on the legs (bodyweight) is reduced [16][28]. The level of muscle force is determined by a complex combination of factors, including neural drive, muscle fiber length, muscle fiber shortening velocity, and past activation history (Lieber RL, Ward SR. Skeletal muscle design to meet functional demands. Philos Trans R Soc Lond B Biol Sci. 2011, 366(1570):1466–76; Gordon A. M., Huxley A. F.& Julian F. J. 1966. The variation in isometric tension with sarcomere length in vertebrate muscle fibres. J. Physiol. Lond. 184, 170–192.; Hill A. V. 1953. The mechanics of active muscle. Proc. R. Soc. Lond. B 141, 104–117.; Hill A. V. 1938. The heat of shortening and the dynamic constants of muscle. Proc. R. Soc. Lond. B. 126, 136–195.; Zajac, F. E. 1989. Muscle and tendon: properties, models, scaling, and application to biomechanics and motor control. Critical reviews in biomedical engineering, 17(4), 359–411). Given that bodyweight support produces reductions in lower limb loading (i.e. ground reaction force), changes in extensor tendon displacement are likely for some, if not all, muscle-tendon units. Both decreases in neural drive and deviations from the optimal muscle fiber length likely contributed to the decreases in muscle force that we calculated.

All three plantarflexors modelled in this study exhibited decreased muscle force and fiber length with increased bodyweight support, but only two of the three plantarflexors had decreased muscle activity. The soleus muscle activity did not change with bodyweight support level but soleus force decreased with increased bodyweight support. A potential reason for the dissimilarities in response to bodyweight between the soleus and gastrocnemius may be contrasting skeletal muscle fiber types. Soleus muscle contains an abundance of type 1 muscle fibers, slow contractile fibers which facilitate endurance tasks, whereas the gastrocnemius has an equal share of slow (type 1) and fast (type 2) muscle fibers [30]. It is not uncommon to see greater dependence on soleus muscle relative to gastrocnemius muscle for ankle joint torque generation at lower mechanical limb loading levels in cats (Herzog, W., & Leonard, T. R. (1991). Validation of optimization models that estimate the forces exerted by synergistic muscles. Journal of biomechanics, 24, 31–39; Herzog, W. 2000. Muscle properties and coordination during voluntary movement. Journal of Sports Sciences, 18(3), 141–152). However, a computer simulation analysis of human walking and running found similar force sharing between the soleus and gastrocnemius despite higher ground reaction forces in running compared to walking (Pandy, M. G., & Andriacchi, T. P. 2010. Muscle and joint function in human locomotion. Annual review of biomedical engineering, 12, 401–433.). In our study, over the more than three-fold variation in peak ground reaction force across bodyweight support levels, we found that greater bodyweight support lead to a greater relative peak soleus force compared to peak gastrocnemius forces (Fig. 7). There were differences in muscle fiber length at peak muscle force for both the soleus and gastrocnemius muscles, but the lack of soleus activation reduction with bodyweight support contributed to the greater reliance on soleus muscle force for ankle joint torque compared to gastrocnemius muscle force at higher bodyweight support levels.

On the other hand, the tibialis anterior showed no significant changes in muscle force, activation or fiber length range with bodyweight support levels. The tibialis anterior showed minor inverse trends as bodyweight support increased across all muscle parameters but we found the trends to be statistically not significant. The literature has provided conflicting data with some researchers finding decreases, increases or no changes in tibialis anterior with bodyweight support. Studies by Ferris et al. [30] and Dietz & Colombo [31] found no change in tibialis anterior activation. The lack of significant changes in tibialis anterior dynamics during stance is not surprising given its primary function as to assist swing dorsiflexion and limit plantarflexion during initial foot contact during the beginning of stance.

The biceps femoris showed an increase in both muscle activation and force during mid-stance with increased bodyweight support. This finding is congruent with previous findings by MacLean & Ferris [20] that found an increase in hip extension moment throughout stance phase with increased bodyweight support during overground walking. We suggest that the increase in muscle activity and force may be caused by the biarticular nature of the muscle which does both knee flexion and hip extension. We did not track other hip flexors/extensors during our study, such as the semitendinosus, semimembranosus, iliacus or psoas. Our lack of information on hip flexor and extensor muscles leaves an incomplete picture on hip actuation during bodyweight supported walking. The hip is known for aiding in forward propulsion during gait (Pandy, M. G., & Andriacchi, T. P. 2010. Muscle and joint function in human locomotion. Annual review of biomedical engineering, 12, 401–433.). The changes in hip extension moment could be attributed to the idea that it is compensating for the large decrease in ankle plantarflexion during stance phase (Lewis, C. L., & Ferris, D. P. 2008. Walking with increased ankle pushoff decreases hip muscle moments. Journal of biomechanics, 41(10), 2082–2089). Future experiments would need to measure electromyography on a greater number of hip muscles (e.g. gluteus maximus, semitendinosus, semimembranosus, or adductor magnus) to determine how muscle activation and force was affected by bodyweight support.

The limitations of our study include the assumptions made in our neuromuscular model, the experimental data recorded, and our exclusion of the swing phase. In our model, we made general assumptions about musculotendon behavior, muscle force-length relationship and neural to muscle activation transformation. These assumptions are widely accepted methods to simulate muscle dynamics through a modified-Hill muscle model. However, these assumptions do not consider any changes in fundamental musculotendon behavior that may occur with increased bodyweight support. In our experimental data collection, we collected EMG signals from only eight muscles of one leg. The eight muscles recorded consisted of the main ankle plantarflexors, ankle dorsiflexors, knee extensors, and knee flexors. As such, more muscles could be added to our EMG-driven model to increase insight onto the hip joint. Despite this we are confident in our results because of the success of the EMG-driven model to estimate forces in other studies [6], [7], [32]. These studies used a sixteen EMG muscle set to capture accurate hip joint movement and muscle activation. Future works could include a full lower body EMG signal collection along with more emphasis on muscular behavior at different walking speeds. Prior research has shown that musculotendon stiffness shifts with different walking speeds [32], [33].

We chose to analyze the stance phase exclusively because it would capture the majority of weight bearing muscles and increase the accuracy of our neuromusculoskeletal model. However, the swing phase is an integral part of walking and the effects of bodyweight support on muscle parameters in swing would be beneficial to know. The nature of our bodyweight support system is such that the legs are not supported in swing phase and therefore we expect bodyweight support to have a much smaller effect on muscle parameters in swing than stance phase.

Our results have shown that for some muscles (soleus, medial and lateral gastrocnemius, rectus femoris and vastii, lateralis & medialis), the use of bodyweight support could be used in a clinical setting to retrain muscular mobility in a reduced load environment. These muscles exhibit an inversely linear relationship between bodyweight support and exerted muscle force. On the other hand, the use of bodyweight support to rehabilitate patients could unintentionally burden the bicep femoris. We have shown that increased bodyweight support in overground walking increased bicep femoris force and activation. As such, it is possible that overground bodyweight support training may be improved by physical therapists or exoskeletons providing additional assistance to hip extension in the stance phase. We also found that the tibialis anterior was not responsive to bodyweight support in stance phase. Bodyweight supported overground walking therefore may not be a suitable environment to regain strength in the tibialis anterior as the force and activation required by the tibialis anterior in stance phase is similar to non-bodyweight supported walking.

Conclusion

Previous research on muscle activity patterns during walking with bodyweight support had found that not all muscles reduce muscle activation in proportion to effective bodyweight. In particular, soleus EMG amplitude was not significantly different across a three-fold range of bodyweight levels for overground walking. Using neuromusculoskeletal modeling, we found that soleus muscle force decreased substantially with increasing bodyweight support. The constant muscle activation combined with changes in muscle fiber length, resulting in lower soleus muscle force at lower effective bodyweights. In contrast, the biceps femoris saw a substantial increase in activation and muscle force during mid-stance at increased bodyweight support. This increase in activation and force can be attributed to its role as a hip extensor. These findings provide evidence that clinicians and biomechanics should not universally expect that muscle activation and muscle force will be reduced when using bodyweight support to assist gait during rehabilitation.

Fig 1:

Experimental design protocol, characterized. RF-Rectus Femoris, VL/VM-Vastus Lateralis/Vastus Medialis, BF-Bicep Femoris, LG/MG-Lateral Gastrocnemius/Medial Gastrocnemius, TA-Tibialis Anterior, SOL-Soleus, MOCAP- Motion Capture, EMG-Electromyogram