Muscle coordination retraining inspired by musculoskeletal simulations reduces knee contact force

Abstract

Humans typically coordinate their muscles to meet movement objectives like minimizing energy expenditure. In the presence of pathology, new objectives gain importance, like reducing loading in an osteoarthritic joint, but people often do not change their muscle coordination patterns to meet these new objectives. Here we use musculoskeletal simulations to identify simple changes in coordination that can be taught using electromyographic biofeedback, achieving the therapeutic goal of reducing joint loading. Our simulations predicted that changing the relative activation of two redundant ankle plantarflexor muscles-the gastrocnemius and soleus-could reduce knee contact force during walking, but it was unclear whether humans could re-coordinate redundant muscles during a complex task like walking. Our experiments showed that after a single session of walking with biofeedback of summary measures of plantarflexor muscle activation, healthy individuals reduced the ratio of gastrocnemius-to-soleus muscle activation by 25 ± 15% (p = 0.004, paired t test, n = 10). Participants who walked with this "gastrocnemius avoidance" gait pattern reduced late-stance knee contact force by 12 ± 12% (p = 0.029, paired t test, n = 8). Simulation-informed coordination retraining could be a promising treatment for knee osteoarthritis and a powerful tool for optimizing coordination for a variety of rehabilitation and performance applications.

Conflict of interest statement

Stanford University has applied for a patent on behalf of S.D.U. and S.L.D. describing the muscle coordination retraining technique, entitled “Real-time electromyography feedback to change muscle activity during complex movements.” The patent is pending at the time of manuscript submission. S.D.U., S.L.D., R.W.J., A.S., and J.A.K. have no other competing interests to disclose.

© 2022. The Author(s).

Figures

Figure 1

Contributions to knee contact force during walking. Muscle force contributions to knee contact force (red) exceed the intersegmental force contribution (blue), which is the reaction force from inverse dynamics. Muscle force contributions are dominated by the hamstrings during the first 10% of stance, the quadriceps from 10–40% stance, and the gastrocnemius from 40 to 90% stance. Forces are shown in terms of bodyweight (BW) for the 10 healthy subjects in this study, and muscle activations were estimated by minimizing the sum of squared activations using static optimization.

Figure 2

Simulation-inspired biofeedback design process. (a) To design the biofeedback for teaching individuals to walk without activating their gastrocnemius (gastroc), we simulated walking using both a natural and a gastrocnemius avoidance objective function (Eqs. 1 and 2) in the static optimization muscle redundancy solver. (b) Based on the simulation results, we provided individuals with visual biofeedback of their muscle activation, measured with electromyography (EMG), that instructed them to change the coordination of their ankle plantarflexor muscles. Participants performed five walking trials: a baseline trial (natural walking), three trials with visual biofeedback, and a retention trial without feedback. During all three feedback (FB) trials, visual biofeedback instructed participants to reduce their gastrocnemius-to-soleus activation ratio (bar magnitude). During the final two feedback sessions, additional feedback was provided, instructing participants to also reduce their average gastrocnemius (gastroc) activity (bar color). (c) EMG-informed static optimization simulations were used to assess the effect of a gastrocnemius avoidance coordination pattern on knee contact force. The simulated gastrocnemius-to-soleus activation ratio was constrained to match the ratio measured with EMG.

Figure 3

Simulation of a gastrocnemius avoidance coordination pattern. Identical joint kinetics were simulated (n = 1) with a natural (Eq. 1) and a gastrocnemius (gastroc) avoidance (Eq. 2) static optimization objective function. During late stance, the muscles generate ankle plantarflexion, knee flexion, and hip flexion moments. The gastrocnemius avoidance coordination pattern requires increased soleus and hamstrings force to compensate for the ankle and knee moments normally generated by the gastrocnemius. The larger knee flexion moment arm (r) of the hamstrings compared to the gastrocnemius allows the hamstrings to generate the same moment with less force than the gastrocnemius, reducing knee contact force. A weighted average of the moment arms (hamstrings: biceps femoris long head and short head, semitendinosus, semimembranosus; gastrocnemius: medial and lateral heads) was computed at 75% of the stance phase, weighted by each muscle’s optimal force in the musculoskeletal model.

Figure 4

Changes in muscle activity and ankle mechanics following coordination retraining. (a) The mean (bar), and standard deviation (error bar) of changes in muscle activation measured with electromyography (n = 10). Participants reduced their gastrocnemius-to-soleus activation ratio and average gastrocnemius (gastroc) activation during training. They retained these reductions following the retention trial (*p < 0.05, paired t test, p-values reported after controlling for the false detection rate). (b–e) The mean (line) and standard deviation (shading) of ankle plantarflexor muscle activity and ankle mechanics for the baseline (base.) and retention (ret.) trials. Despite the 25 ± 15% change in activation ratio from the baseline to retention trial, the stance-phase-averaged ankle moment only changed by 3 ± 14%, which trended towards being equivalent to baseline within one baseline standard deviation (p = 0.063, two-one-sided t tests for equivalence, n = 10).

Figure 5

Simulated and measured muscle activation. Electromyography (EMG) linear envelopes averaged across all participants (n = 10) with a 40 ms electromechanical delay are compared to simulated muscle activations for the baseline and retention trials. Despite magnitude differences between EMG and simulated activations of the gastrocnemius and soleus, the relative changes between trials are consistent, indicating that our EMG-informed static optimization technique captured the changes in ankle plantarflexor muscle activity measured with EMG. The activations of the muscles in the top row were not informed by EMG in the simulation. The shape, timing, and between-trial changes in simulated activation matched EMG for these muscles, with the exception of the rectus femoris.

Figure 6

The effect of a gastrocnemius avoidance coordination pattern on knee contact force. (a) The mean (line) and standard deviation (shading) of knee contact force for the participants who reduced late-stance gastrocnemius activation during the retention trial (n = 8). These participants reduced their second peak of knee contact force compared to baseline (*p = 0.029, paired t test). (b) Changes in the first and second peak contact force between baseline (base.) and retention (ret.) are shown for all participants (n = 10), with the two participants who did not retain a reduction in late stance gastrocnemius activation represented with dashed lines. Six of the eight individuals who reduced late-stance gastrocnemius activation reduced their second peak knee contact force, but five increased their first peak. (c) For the eight subjects who reduced late-stance gastrocnemius activation, the change in knee contact force is decomposed into the intersegmental and muscle force components. Reductions in the second peak of knee contact force are primarily driven by reductions in gastrocnemius force.

Figure 7

Changes in knee and hip mechanics. The mean (line) and standard deviation (shading) of knee and hip kinematics and kinetics for the participants who retained a reduction in their late-stance gastrocnemius activation during the retention trial (n = 8). During the retention trial, participants walked with a smaller late-stance knee flexion moment (*p = 0.015, paired t test).