The purpose of this study was to demonstrate that simulation results of plantar flexion activation and function for individuals pre- and post-intervention were consistent with (1) the purpose of the intervention and (2) expected muscle function during gait based on previous literature. In addition, we identified correlations between simulation results and clinical measures of walking function. Sixteen subject-specific models were created to represent pre- and post-training gait for 8 individuals post-stroke. To our knowledge, this is the first study to use musculoskeletal simulations of persons with chronic hemiparesis to simulate changes in muscle activation and function before and after participation in a gait retraining intervention.
Significant increases were seen in simulated plantar flexor activation during double support with no changes in total predicted activation over the entire gait cycle, suggesting selective increases in predicted plantar flexion activation during the phase of gait targeted by the intervention, i.e. double support. This new simulated activation pattern post-intervention is consistent with the timing of PF FES used during the intervention, which stimulated PF muscles only during pre-swing. This post-intervention activation pattern also agrees with healthy PF muscle coordination where the plantar flexors are activated most during late single leg stance and pre-swing. Thus, this new simulated pattern of activation for the plantar flexor muscles suggests that the subjects activated these muscles with more appropriate timing following the intervention.
The soleus and gastrocnemius play key roles in forward progression and swing initiation, respectively, primarily during the double support phase of gait . Post-stroke muscle weakness of the PF muscles can greatly limit force generation by these muscles during gait, leading to decreased walking speed, limited swing phase knee flexion and poor foot clearance during swing [26, 27]. Functional electrical stimulation during the FastFES intervention targeted the plantar flexor muscles pre-swing with the goal of enhancing the contribution of these muscles to their respective subtasks and improving gait.
In hemiparetic individuals, enhancing the contribution of the soleus towards forward COM acceleration has been identified as an important mechanism for increasing walking speed . Interestingly, our models simulated that for 7 of 8 subjects post-stroke pre-training, the soleus decelerated the COM during double support, which is contrary to what is expected for healthy walking [11, 28]. This suggests that the subjects’ paretic limb was in a poor biomechanical position for the soleus to accelerate the body forward. Post-training, the simulated PF muscles decelerated the COM less than pre-training for seven of eight subjects. In addition, simulations of 3 subjects began to accelerate the COM forward with the PF muscles post-intervention. Not surprisingly, these three subjects were three of the four fastest walkers post-training. Interestingly, the subject who showed the largest simulated increase in plantar flexor induced forward COM acceleration with training also showed the largest increase in self-selected walking speed (0.3 to 0.9 m/s), and walked at the fastest speed of the eight subjects post-intervention. It is important to note that the subjects in this study were walking at very slow speeds, ranging from 0.3 to 0.5 m/s pre-intervention, and the deceleration caused by the PF muscles pre-swing was likely a limiting factor to walking speed. A similar result was seen in a recent simulation study on a single individual post-stroke walking at a slightly faster speed of 0.6 m/s.  Peterson et al. highlighted forward propulsion as a limiting factor for walking speed, and also reported that the gastrocnemius contributes to negative acceleration of the pelvis during pre-swing , similar to the negative COM acceleration by the medial gastrocnemius found in our study.
Neptune and colleagues have shown that the gastrocnemius muscle is critical for swing initiation during pre-swing, increasing swing phase knee flexion and enabling foot clearance during swing by accelerating the knee into flexion . For seven out of eight of our subjects, the simulated medial gastrocnemius exhibited an increased contribution to knee flexion acceleration post-intervention. This increase in knee flexion acceleration is associated with improved trailing limb posture combined with simulated activation of the gastrocnemius during double support, resulting in a change in muscle timing relative to joint posture. Additionally, the increase in simulated knee flexion acceleration was concurrent with increased walking speed and is consistent with previous simulation studies [15, 29] that suggested that increased contribution to swing initiation by the gastrocnemius is required to increase walking speed. However, this is the first study to actually demonstrate that improved simulated plantar flexor activation is related to improved simulated knee flexion acceleration and walking speed, as was predicted by the previous cross-sectional analysis [15, 29]. Moreover, the simulations allowed an analysis of the contribution of changes in plantar flexor activation to changes in knee flexion acceleration, something that cannot be examined through experimental data. This serves as an example of how muscle-actuated simulations can enhance our understanding of changes with intervention beyond what can be ascertained from experimental data alone.
In a previous experimental study investigating immediate effects of FES, peak knee flexion during swing was hypothesized to increase with greater forward propulsive forces as a result of increases in simulated PF activation during pre-swing . In our simulations, PF activation correlated positively with peak knee flexion. This relationship was partially explained by gastrocnemius function (Figure 2), which can achieve increases in knee flexion directly through the bi-articulation at the ankle and knee, and by increased trailing limb angle. Leg extension by the trailing limb has been suggested to be important for achieving propulsion in persons with stroke [2, 18] and was one goal of the fast treadmill training intervention used in this study. By using fast walking to increase the trailing limb angle, the ground reaction force exerted at push-off was directed in a more horizontal orientation, providing a greater propulsive force in the forward direction. This greater propulsive force probably allowed for increased forward acceleration of the body and limbs and enabled faster walking speeds and greater knee flexion during swing. Also, greater forward COM acceleration simulated by our model was predictive of greater self-selected walking speed post-training, a relationship that is consistent with previous cross-sectional study predictions .
This study examined changes in simulated post-stroke muscle activation and function at self-selected speed after a gait retraining intervention. Due to the limited number of subjects and the high variability in individuals post-stroke, it is not known if the findings of this study apply to the population of individuals post-stroke as a whole. Also, all subjects walked at a faster self-selected walking speed post-intervention. While the use of greater walking speeds post-intervention may have confounding effects on variables such as forward COM acceleration, we believe this analysis is important because it allows us to determine what changes in model predicted muscle activation and function were necessary to achieve the increases seen in walking speed post-intervention, a common goal of gait is retraining. In fact, this the first study to show concurrent improvements in speed, propulsion, and predicted PF activation after a targeted training intervention with individuals post-stroke. Due to the absence of a control group, it is not clear if the changes seen from pre- to post-intervention are a specific result of either FES or fast treadmill walking. Future work will include groups trained with FES or at fast walking speeds to assess the individual components of the intervention.
The analysis of muscle function is somewhat limited since activation was simulated without the use of EMG for this study. However, our modeling does account for the effect of limb posture on muscle function, which is not accounted for by EMG. For some of the muscles included in this model, surface EMG cannot be obtained. When used longitudinally with an intervention, EMG magnitude can be insensitive to hypertrophy of muscle, which makes comparisons difficult over time. Additionally, relying on EMG signal amplitudes can be difficult, as signal magnitude can vary based on electrode placement and tissue conductivity . However, EMG can be useful for constraining timing of muscle activity, and future studies should consider its use. Additionally, the cost function used in the model minimizes the sum of the squares of the muscle activations, and generic muscle properties were used. Maximum isometric force parameters were not changed pre- to post-intervention, so it is possible that some of the increase in model predicted activation was due to strength increases elicited by the intervention. The selection of the cost function in particular could have an impact on the muscle activations predicted, as it is possible that the use of a different cost function could result in a different pattern of muscle activations which also reproduce the experimental kinematics and kinetics.
Subjects were allowed to use handrails during walking trials for safety purposes. Although the subjects were instructed to use the handrails with a ‘light touch,’ some subjects may have applied larger forces to the handrails. While the forces applied to the handrails were not accounted for explicitly, residual forces were calculated during the simulation and applied to the COM to account for kinetic imbalances due to external forces (i.e. handrail force). These residual forces were found to agree closely with the forces applied on the instrumented handrails during data collection.