Skip to main content
Download PDF
Download ePub

Can muscle synergies shed light on the mechanisms underlying motor gains in response to robot-assisted gait training in children with cerebral palsy?

Journal of NeuroEngineering and Rehabilitation volume 22, Article number: 23 (2025) Cite this article

Abstract

Background

Children with cerebral palsy (CP) often experience gait impairments. Robot-assisted gait training (RGT) has been shown to have beneficial effects in this patient population. However, clinical outcomes of RGT vary substantially from patient to patient. This study explored the hypothesis that clinical outcomes are associated with changes in muscle synergies in response to RGT.

Methods

Thirteen children with CP and Gross Motor Function Classification Scale (GMFCS) levels I–IV were recruited in the study. Children participated in a 6 week-RGT intervention and underwent clinical evaluations and gait studies—with focus on the analysis of electromyographic (EMG) data—pre- and post-training. Lower-limb muscle synergies were derived from the EMG recordings. Pre- vs. post-RGT clinical outcomes and muscle synergies were compared to explore potential relationships.

Results

Three and, less often, two muscle synergies were detected in study participants pre-RGT. Linear mixed effect models showed that composition of the muscle synergies and their temporal activation coefficients present deviations from normative data proportional to the severity of functional limitations (i.e., GMFCS levels, p < 0.01). At a group level, changes in muscle synergies pre- vs. post-RGT did not significantly correlate with changes in clinical outcomes (p > 0.05). However, it was observed that participants who displayed prominent changes in muscle synergies also displayed large improvements in clinical scores.

Conclusions

Gait impairments in children with CP were associated with muscle synergies that deviated from normative. Participants who demonstrated the most substantial improvements in clinical scores following RGT exhibited multiple changes in the muscle synergies. However, no statistically significant correlations were identified at the group level. Future studies relying on larger datasets are needed to further investigate this observation and potential underlying mechanisms.

Introduction

Cerebral palsy (CP) is formed by a group of disorders affecting movement, development, and posture that cause significant functional limitations, and it is the most common cause of motor disability in children [1]. Its estimated prevalence has been reported in the range between 1.5 and 3 per 1000 live births [1]. Its clinical features include muscle weakness, spasticity, and impairment of selective motor control [2].

Children with CP often experience gait impairments with a negative impact on mobility and quality of life [3,4,5,6]. Gait deviations in CP display large variability in their characteristics, are often complex, and can evolve during development, sometimes causing muscle contractures and joint deformities due to an imbalance in the forces generated by agonist–antagonist muscles and in muscle tone [7]. Interventions to improve gait are an important component of the treatment of children with CP [5]. Rehabilitation can minimize complications and improve gait, thus enhancing function [5].

Robot-assisted gait training (RGT) has been used as an efficient way to deliver high-dosage, high-intensity, task-specific interventions. These are “training ingredients” believed to maximize motor gains. While several studies have shown positive clinical outcomes in response to RGT at the group level [8,9,10,11,12,13,14,15,16,17], careful examination of the results of studies that provided data on a subject-by-subject basis reveals that significant variability marks the response to RGT in children with CP [8, 18,19,20,21,22]. It is unclear why children with CP display such a variability in their response to RGT. Are there individual patient characteristics that make RGT suitable for some children with CP and not for others? If clinicians knew why some children respond prominently to RGT whereas others do not, they would be able to prescribe RGT when a clinically important response is expected (based on the patient’s characteristics) and consider alternative treatments otherwise.

Muscle synergies represent muscle coordination patterns used to generate motor outputs. Their analysis provides a window of observation on how the nervous system generates movement patterns [23]. Several studies have shown that muscle synergies during gait are altered in children with CP [24]. Compared to typically developing children, children with CP display fewer muscle synergies marked by co-activations of agonist and antagonist muscles [25,26,27]. Changes in muscle synergies have been studied in children with CP in response to selective dorsal rhizotomy [28,29,30,31], orthopedic surgery [28, 30,31,32], botulinum toxin type A injections [28, 30, 33], and conservative treatment (i.e., physical therapy) [30], but changes in lower limb muscle synergies in response to RGT have only been reported in a study by Conner et al. [34] focused on assessing the effects of a robotic system developed by the authors that was used to deploy resistance training in children with CP. Because gait deviations in children with CP are associated with a variety of aberrant patterns of muscle activation [35, 36], we decided to explore the relationship between the characteristics of such patterns and the response to RGT in children with CP hoping to shed light on what causes the above-mentioned variability in the response to the intervention. To achieve this goal, we relied on the analysis of muscle synergies [37,38,39].

In the study herein presented, we performed an exploratory investigation in a group of children with CP who underwent RGT using a robotic system (LokomatPro by Hocoma AG, Volketswil, Switzerland) most often used in clinical sites that provide pediatric patients with access to RGT. We evaluated their muscle synergies pre- and post-training with the overall objective of investigating the relationship between changes in muscle synergies and clinical outcomes (including functional measures and quality of movement as described in the Methods section of the manuscript). Metrics derived to quantify changes in muscle synergy characteristics in response to RGT served as the primary outcomes of the study. More specifically, the cosine similarity was used to compare the weights of the muscle synergies and the zero-lag cross-correlation (ZLCC) was used to compare their temporal coefficients. Functional measures and measures of quality of movement pre- vs post-training served as the secondary outcomes. To achieve the above-stated overall objective of the study, we tested three hypotheses: (1) there is a correlation between the degree of disarrangement of muscle synergies and the functional ability of children with CP; (2) muscle synergies change in response to RGT in a way that makes them more similar to normative synergies (i.e., positive changes); and (3) changes in muscle synergies correlate with changes in clinical outcomes. Furthermore, we performed a qualitative analysis of the results to identify instances in which prominent changes in muscle synergy characteristics were accompanied by large motor gains.

Methods

Participants

A convenience sample of 13 children with CP and gait impairments were enrolled in the study. This sample size is comparable with the sample size used in previous exploratory studies with focus on muscle synergies [27, 40, 41]. All parents or guardians signed a consent form, and children signed an assent form. Both were approved by the Spaulding Rehabilitation Hospital Institutional Review Board (protocol # 2015P001482, clinical trial # NCT06156969). Children were eligible to participate in the study if they were diagnosed with spastic CP, were between 6 and 18 years of age, were classified between levels I and IV of the Gross Motor Function Classification System (GMFCS) [42], had femoral length between 210 and 350 mm (as required to use the RGT system utilized in the study), and had the ability to communicate pain and discomfort. Individuals were excluded from the study if they had received RGT during the last 3 months, reported any contraindication to RGT (such us thromboembolic disease, progressive neurological disorder, cardiovascular or pulmonary contraindications), displayed aggressive behavior, had severe cognitive deficits, joint instabilities, bone fractures, osteoporosis, lower-limb joint fusion (i.e., arthrodesis), or skin ulcers affecting the trunk or lower limbs.

Experimental procedures

Figure 1 shows a schematic representation of the study procedures. Details are provided in the following.

Fig. 1
figure 1

Overview of the study procedures. Children underwent a 6-week Robot-assisted Gait Training (RGT) intervention (3 sessions per week) and pre- and post-RGT evaluations including sections D and E of the Gross Motor Function Measure (GMFM) scale, the 10-m walk test (10 MWT), the 6-min walk test (6 MWT), and a gait evaluation to collect surface electromyographic (EMG) data as well as perform a visual assessment of patterns of motion using the Edinburgh Visual Gait Score (EVGS)

Full size image

Robot-assisted gait training protocol

Children were asked to undergo 18 gait training sessions using the LokomatPro (Hocoma AG, Volketswil, Switzerland) over a period of approximately 6 weeks. Each session included 30 min of walking assisted by the robot, allowing for rest breaks as needed. Training sessions were overseen by a clinician (a physical therapist or a physiatrist) trained in the use of the robotic system. Bodyweight support, walking speed, and guidance force provided by the robot were adjusted during the training session as deemed appropriate by the clinician overseeing the session. Bodyweight support and guidance force were adjusted in a way that allowed the child to achieve nearly normative gait kinematics while maximizing the level of participants’ engagement. In other words, the child was encouraged to put maximum effort toward generating the gait movements while the clinician overseeing the session decreased the bodyweight support and guidance force to the maximum extent possible without compromising the quality of the lower-body kinematics. This approach was adopted with the intention of achieving maximum engagement of the child during the training session (which is deemed relevant to motor learning) while avoiding aberrant movements (which are deemed detrimental to improving gait).

Surface EMG data during gait

Surface electromyographic (EMG) data was gathered pre- and post-RGT during overground gait using the Wave system (Cometa srl, Bareggio, Milan, Italy). The data was collected at 1800 Hz using wireless probes from the following muscles in both lower limbs: gastrocnemius (lateral head), soleus, tibialis anterior, peroneus longus, rectus femoris, vastus medialis, vastus lateralis, biceps femoris, semitendinosus, and gluteus maximus. We followed the SENIAM 8 (Surface Electromyography for the Non-Invasive Assessment of Muscles) guidelines to place the electrodes. The guidelines provide instructions about how to position the EMG electrodes for each muscle as a point on a line between two anatomical landmarks [43]. The pre-amplification units of the EMG system were attached using double-sided tape and secured with Coban. During the baseline session, after the electrodes were placed, we measured the distance from each of the electrodes to two specific body landmarks according to the above-referenced guidelines (e.g., for the electrodes used to record the activity of the gastrocnemius muscle, we measured the distance between the two electrodes and the head of the fibula and the lateral malleolus, respectively). These measures were used to place the EMG electrodes in the same position during the post-RGT assessment session. The system used to collect the EMG data was integrated with a motion capture system (Vicon, Oxford Metrics, Yarnton, Oxfordshire UK) used in the study to segment the EMG data according to the gait cycles. A minimum of five walking trials were collected for each side (i.e., leg). Participants were tested in barefoot conditions. They were allowed to use assistive devices if they could not walk safely without.

Normative muscle synergies were derived from recordings gathered (using the same experimental setup) from nine adult healthy controls as part of a study approved by the Spaulding Rehabilitation Hospital Institutional Review Board (protocol # 2019P002419). The normative muscle synergies we obtained from the control group were consistent with those reported in previous studies in children of age comparable with our group [28]. Specifically, we identified four muscle synergies. The first synergy was active during early stance (load acceptance) and marked by a prominent activation of the gluteus maximus, vastus lateralis, vastus medialis, and rectus femoris muscles. The second muscle synergy was active during the push off phase of the gait cycle and was marked by a prominent activation of the gastrocnemius, soleus, and peroneus muscles. The tibialis anterior and rectus femoris muscles were the primary contributors to the third synergy, which was active during the swing phase and allowed for foot clearance. Finally, the fourth muscle synergy was primarily marked by the activation of the biceps femoris and semitendinosus muscles and accounted for the deceleration of the leg during the terminal swing phase of the gait cycle.

Clinical data

Demographic, anthropometric and clinical data including sex, age, race, ethnicity, dominant side, weight, height, characteristics of CP (i.e., topographic classification and GMFCS level [42]), other diagnoses, orthoses or assisted devices used for walking were collected at baseline. Assessments were performed before and after study participants underwent RGT. The clinical tests performed by a clinician during the evaluation sessions included the standing (section D) and the walking, running and jumping (section E) sections of the Gross Motor Function Measure (GMFM-88) scale [44], walking speed measured by the 10-m walk test (10 MWT), and endurance using the 6-min walk test (6 MWT). Video recordings were collected during the gait trials for the EMG studies. Standard cameras were utilized to provide coronal and sagittal views and derive Edinburgh Visual Gait Scores (EVGS) [45].

Data analysis

Analysis of muscle synergies

Figure 2 shows a schematic representation of how muscle synergies contribute to generating patterns of muscle activity [46]. In this schematic example, the EMG envelopes of four muscles are shown to be the result of combining three muscle synergies. The weights of each muscle synergy determine how much each of the four muscles contribute to that specific synergy. The temporal coefficients (often referred to as “temporal activations” or “temporal activation coefficients”) determine the level of activation of each synergy over time. This model provides the basis for the analysis approach utilized in the study as described below.

Fig. 2
figure 2

Generation of muscle synergies. Muscle synergies are “modules” utilized by the central nervous system (CNS) to reduce the dimensionality of motor control during the performance of complex movements. The “modules” consist of weights (that determine the level of contribution of each muscle to a given synergy) and temporal coefficients (that determine the level of activation of each synergy over time). The combination of weights and temporal coefficients leads to the patterns of activation (and hence the EMG envelopes) of all monitored muscles

Full size image

Raw surface EMG data were filtered using a 4th-order high-pass Chebyshev filter, with a cut-off frequency of 20 Hz, to attenuate movement artifacts. The output data were rectified and filtered using a 4th-order low-pass Chebyshev filter, with a cut-off frequency of 5 Hz, to obtain the EMG envelope for each EMG channel, which was normalized by the peak value observed for each subject for that given channel [47, 48]. The resulting time-series were segmented according to the gait cycles (i.e., intervals from foot-contact to foot-contact). The EMG envelope for each gait cycle was then resampled to 100 points. The first and last gait cycles of each trial were discarded.

Muscle synergies were extracted separately for each study participant [46]. The EMG data for a total of 20 gait cycles per study participant were used for the analyses herein described. The EMG envelope data for each subject were stored in a m × t matrix, “m” being the number of muscles (i.e., ten) and “t” being the total number of samples (20 gait cycles × 100 points per cycle = 2,000 samples). Data were analyzed separately for each leg. We used the non-negative matrix factorization (NNMF) [47, 48] function provided by MATLAB (MathWorks, Natick MA, USA). The algorithm factors the initial “A” matrix (m × t) into two non-negative matrices denoted “W” (m × n) and “T” (n × t) by minimizing the root mean square value of the residual “D” defined by the following equation:

$$D = \frac{{||\left( { A - W*T } \right)||_{F} }}{{\surd \left( {n*m} \right)}}$$

where “n” is the number of synergies selected by the user, “m” the number of EMG channels, and F the Frobenius norm.

The W matrix identifies the contribution of each muscle to each synergy (i.e., weights). The T matrix defines the temporal activation of each synergy (i.e., temporal coefficients). The multiplicative update algorithm was utilized for these analyses. The number of replicates (number of times the algorithm is initialized to derive weights and temporal coefficients) was set to 5000 and the maximum number of iterations was set to 500 [49]. We identified the number of synergies by evaluating the difference, computed as R-squared (R2) value, between the EMG envelopes and their synergy-based reconstructions [49]. We used a double-threshold approach to select the number of synergies by requiring an average R2 > 75% and a minimum R2 per channel > 55% [50].

Because prior work suggested that EMG data collected from children with CP display fewer muscle synergies than data collected from a normative sample [26, 28, 35, 41, 51], we evaluated if the synergies observed in children with CP could be considered the result of merging normative synergies. We applied the technique proposed by Cheung et al. [39, 48]. Accordingly, we modeled the weights of each muscle synergy observed in children with CP using our healthy control reference synergies according to the following equation:

$$w_{{CP_{i} }} \approx \mathop \sum \limits_{k = 1}^{Kref} p_{k }^{i} w_{{REF_{k} }}$$

where the “i-th” CP synergy “\({w}_{{CP}_{i}}\)” was modeled, using a least square fit, as a linear combination of the four normative synergies “\({w}_{{REF}_{k}}\)”. This generated four coefficients “\({p}_{k }^{i}\)”, which represent how much each normative synergy contributed to “\({w}_{{CP}_{i}}\)”.

The results of the above-described analyses allowed us to test the hypothesis that the degree of disarrangement of muscle synergies is correlated with the functional ability of children with CP (as captured by GMFCS levels). We used two metrics: (1) the cosine similarity to compare weights (i.e., to compare the composition of the muscle synergies in healthy controls with that observed in children with CP); and (2) the zero-lag cross-correlation (ZLCC) value to compare temporal coefficients (i.e., to compare the temporal activation of the muscle synergies in healthy controls with that observed in children with CP). Both the cosine similarity and the ZLCC parameters span the interval between 0 and 1, where the maximum value represents a perfect match with the reference (i.e., average observation in healthy controls) [23, 52, 53]. For these analyses, we considered all the outputs of the muscle synergy algorithm that satisfied the above-stated R2 criteria. It is worth noting that the cosine similarity has been used to compare the composition of different muscle synergies in previous studies as it is an effective metrics when one wants to capture changes in the weights that determine the contribution of different muscles to a synergy [54]. Also, the ZLCC has been extensively used to compare waveforms of a nature similar to the temporal coefficients that we compared in this study as it provides an effective way to measure differences in the shape of waveforms of interest [52]. To test the hypothesis that there is a correlation between the degree of disarrangement of muscle synergies and the functional ability of children with CP, we used mixed-effect regression models (for each individual synergy) with GMFCS levels as fixed effect and random effects to account for between-subjects and within-subject differences (i.e., “SynergySimilarity ~ 1 + GMFM_Level + (1 | SubID)”).

Then, we focused on longitudinal analyses and investigated if muscle synergies change in response to RGT in a way that makes them more similar to normative synergies. We estimated changes in response to RGT in cosine similarity and ZLCC values (using normative muscle synergies as reference). For these analyses, we computed the mean and standard deviation of the distribution of similarity values obtained by comparing each synergy with the normative synergies. The results were examined for both the most affected and the contralateral leg. The range spanned by the observed changes in cosine similarity and ZLCC values was divided into intervals based on their magnitude. Results were reported in table format displaying participant-by-participant data. To assess the relevance of changes in muscle synergy characteristics, we used as reference the variability observed in our normative dataset (with typical standard deviation ≅ 0.015). We reasoned that, if a change (pre- vs post-training) was comparable with the variability that marks the normative dataset, it should not be considered significant. Accordingly, we considered positive changes < 0.02 as not significant; changes ≥ 0.02 but < 0.04 as moderate improvements and highlighted them in light green; changes ≥ 0.04 but < 0.06 as large improvements and highlighted them using a dark green color; and changes ≥ 0.06 as very large improvements and highlighted them using a darker green color. Negative changes of the same magnitudes were considered moderate, large, and very large worsening and highlighted in orange, light red, and dark red, respectively. A Chi Square test was performed on these data to test if the changes in muscle synergies in response to RGT were random changes.

Clinical outcomes and muscle synergies

Because we anticipated significant differences across study participants in both clinical outcomes and muscle synergies in response to RGT, we sought to investigate if the results of the muscle synergy analysis could be used to shed light on the response to RGT on a participant-by-participant basis.

Paired t-tests were first performed on the GMFM, 10 MWT, 6 MWT, and EVGS scores to assess if data collected in the study showed significant changes in clinical scores in response to RGT as previously observed in other studies [10, 16, 18, 19, 55, 56]. A 5% significance level was used for these analyses. Then, we considered the changes in response to RGT observed on a participant-by-participant basis and examined potential associations with changes in muscle synergies.

Changes in clinical scores on a participant-by-participant basis were identified as follows. Changes in sections D and E of the GMFM-88 that were ≥ 2 points were highlighted as exceeding the minimum clinically important difference (MCID) in standing (1.2 points) and walking (1.6 points) function as suggested by Oeffinger et al. [57]. The range of motor gains exceeding this value was divided into three equal intervals. Data falling in these three intervals was labeled as “moderate”, “large”, and “very large” motor gains, respectively.

The choice of MCID values for the 10 MWT and 6 MWT in children with CP has been a matter of debate [58]. In absence of well-established reference values, criteria consistent with clinical experience as reported by Storm et al. [59] were used in this study. Accordingly, changes in 10 MWT and 6 MWT values in response to RGT that fell between 10 and 30% of the baseline value were labeled as “moderate”; changes between 30 and 50% of the baseline value were labeled as “large”; and changes exceeding 50% of the baseline value were labeled as “very large”.

The EVGS scores of the two legs were used to determine the most affected leg of each participant. The EVGS scores were then averaged to generate a total EVGS score (i.e., combining the scores of both legs). Changes exceeding the MCID threshold value of 1.9 points [60] were highlighted and the range of observed EVGS scores exceeding the MCID threshold value was divided into three equal intervals as described above for the GMFM scores.

Finally, we tested the hypothesis that positive changes in muscle synergies correlate with positive clinical outcomes. We used the above-stated definitions of positive and negative changes in clinical outcomes and muscle synergies and derived accordingly the Kendall correlation and its significance. In addition, qualitative observations were made based on visual observation of the summary tables of changes in muscle synergies and clincial outcomes in response to RGT.

Results

Table 1 provides a detailed description of the participants’ characteristics. Participants were 13.1 ± 3.4 years old (mean ± standard deviation). Eight out of the thirteen participants were females. Most participants had diplegia (9 children), two had quadriplegia, one had triplegia, and one had hemiplegia. Participants used different types of ankle–foot orthoses. Some of them used a mobility assistive device (e.g., one used a cane, three of them crutches, and three a walker). Eight subjects participated in 18 RGT sessions, four completed 17 sessions, and one completed 12 sessions.

Table 1 Subject-by-subject demographic and clinical data at baseline
Full size table

Muscle synergies in children with CP

The composition of the muscle synergies pre- and post-RGT in all study participants is shown in Fig. 3. Herein we show the composition of the muscle synergies as the linear combination of the four normative synergies. The color-coded graphical representation indicates the percentage contributions of the normative synergies to each of the synergies observed in children with CP.

Fig. 3
figure 3

Muscle synergies Pre- vs Post-RGT. Left panel: the gait cycle of healthy volunteers (top), the weights of the normative (i.e., observed in healthy volunteers) muscle synergies (middle), and the gait cycle of children with CP (bottom) showing an example of merging of muscle synergies. The main muscles active in each phase of the gait cycle are highlighted in the representation of both the healthy child and the child with CP. Right panel: a representation of the percentage contribution of each normative synergy to the muscle synergies observed in children with CP in the most affected leg (top) and contralateral leg (bottom). Color code: red—load acceptance synergy, blue—push off synergy, black—foot clearance synergy, and purple—leg deceleration synergy. CP cerebral palsy, RGT Robot-assisted Gait Training

Full size image

Almost all the participants showed three muscle synergies, both pre- and post-RGT. Two participants displayed two muscle synergies pre-RGT. When we identified three muscle synergies, their composition was typically characterized as follows. The first muscle synergy (Syn 1) most often displayed the “load acceptance” and “leg deceleration” normative synergies as its dominant components. The “push-off” normative synergy also contributed to this synergy to a degree that varied across study participants. The second muscle synergy (Syn 2) mostly resembled the “push-off” normative synergy but included components from other normative synergies (most often the “load acceptance” and “leg deceleration” normative synergies). Finally, the third muscle synergy (Syn 3) displayed high similarity with the “foot clearance” normative muscle synergy but included components from other normative synergies (most prominently the “push-off” normative synergy) to a variable degree across participants.

A qualitative observation of the temporal activation coefficients (not shown in Fig. 3) highlighted a less prominent modulation of the muscle synergies than typically observed in control subjects. In a clinical context, this is often referred to as “non-phasic activity”. The lack of prominent modulation was more apparent in children with more severe functional limitations (i.e., GMFCS levels III and IV compared to levels I and II).

Muscle synergies and GMFCS levels

To quantify the relationship between the characteristics of the muscle synergies and the GMFCS levels, we used the cosine similarity value to compare the weights of the normative muscle synergies and the muscle synergies of children with CP [23, 52, 53]. We used the zero-lag cross-correlation (ZLCC) value as the metric for comparison of the normative and the participants’ temporal coefficients. Figure 4 shows the cosine similarity and the ZLCC for both legs combined and grouped by GMFCS levels [23, 52, 53]. Data for the control group (which also displayed variability across individuals) is shown in the figure to provide a reference value. GMFCS levels III and IV were associated with muscle synergies that appeared to deviate from the normative muscle synergies more than for GMFCS levels I and II in both their composition (i.e., weights) and temporal activations (i.e., temporal coefficients). We performed a statistical analyses using mixed-effects regression models with GMFCS levels as fixed effect. Separate models were fitted for the cosine similarity values and for the ZLCC values of each synergy. Significance was achieved for the cosine similarity values and ZLCC values of all three synergies (p < 0.01). These results imply that the correlation between GMFMCS levels and cosine similarity values as well as ZLCC values observed by visual inspection of the results (Fig. 4) is statistically significant. Visual observation of the muscle synergies in study participants suggested that this result was due to a higher incidence of co-activations of agonist and antagonist muscles in the composition of the muscle synergies with an increase in GMFCS levels. Besides, a less prominent modulation of the temporal coefficients was observed for higher GMFCS levels. The differences between muscle synergies for level II and level III participants appear to be mostly captured by the weights rather than the temporal coefficients of the muscle synergies.

Fig. 4
figure 4

Cosine similarity and ZLCC of muscle synergies. Boxplots are shown for controls and for different GMFCS levels (Level I–IV). Boxplots combine values for both legs and for pre- and-post-RGT recordings. GMFCS Gross motor function classification system, Syn Synergy, ZLCC zero-lag cross-correlation, RGT Robot-assisted Gait Training. Marked with an asterisk symbol the groups that shown significant difference in the Dunn’s post hoc comparisons (p < 0.05)

Full size image

Changes in muscle synergies in response to RGT

Tables 2 and 3 show the cosine similarity and ZLCC values pre- and post-RGT for the most affected and the contralateral leg, respectively. The cells of these tables are highlighted in different colors according to the magnitude of the change observed in response to RGT as explained in the Methods section. We performed Chi Square tests and despite the small sample size, this analysis highlighted significant changes in response to RGT in the cosine similarity values for Syn 1 which displayed patterns closer to normative post-RGT (p = 0.042). Visual inspection of these tables suggested that changes in cosine similarity values for the most affected leg (Table 2) is marked by a more consistent pattern of improvement for the first muscle synergy (Syn 1) compared to the second (Syn 2) and third (Syn 3) synergies. In fact, when a change was observed for Syn 1, it was always an improvement in similarity with the normative data. The largest improvements in cosine similarity for Syn 1 were observed for two of the GMCFS level III participants (i.e., Sub 12 and 13). More variability was observed in changes in cosine similarity for Syn 2, with some participants displaying very large improvements (such as Sub 11 and 12) and others displaying a considerable worsening in cosine similarity (such as Sub 7). Highly variable results were observed for Syn 3. Changes in cosine similarity for the contralateral leg (Table 3) appeared to be less consistent across participants for all three muscle synergies. Changes in ZLCC values for the most-affected (Table 2) and the contralateral leg (Table 3) in response to RGT displayed patterns similar to the ones displayed by the changes in cosine similarity values. More consistent improvements in ZLCC values were generally observed for Syn 1. In contrast, Syn 2 and 3 showed a less consistent ZLCC change in response to the intervention.

Table 2 Cosine similarity and ZLCC for each muscle synergy of most affected leg pre- and post-RGT
Full size table
Table 3 Cosine similarity and ZLCC for each muscle synergy of the contralateral leg pre- and post-RGT
Full size table

Clinical outcomes of RGT

To explore a potential association between changes in muscle synergies in response to RGT and clinical outcomes, we examined the clinical outcome data pre- and post-RGT shown in Table 4 and then compared changes in the outcomes shown in this table with those observed in the above-discussed tables summarizing the results of the muscle synergy analysis (i.e., Tables 2 and 3).

Table 4 Clinical outcomes and EVGS scores, Pre- and Post-RGT
Full size table

Group-level analysis of pre- vs post-RGT clinical outcomes showed improvements in sections D (p ≅ 0.02) and E (p < 0.01) of the GMFM-88, and in EVGS scores (p < 0.01). Nearly significant improvements were observed in the 6 MWT scores (p ≅ 0.06), whereas no group change was observed in the 10 MWT scores (p ≅ 0.46). These results appear to be consistent with previous reports on the effects of RGT in children with CP [16, 19, 55, 61].

A closer look at Table 4 shows a large variability in the response to RGT across subjects. Elements of the table are highlighted in different colors according to the magnitude of the observed change and its direction (i.e., improvement vs. worsening in clinical scores according to the criteria described in the Methods section). Participants displayed improvements across different clinical dimensions: eight participants showed an improvement in section D of the GMFM-88, ten showed an improvement in section E of the GMFM-88, four in gait speed, seven in gait endurance, and seven in gait quality (i.e., EVGS scores). Some participants displayed an improvement only in one clinical dimension, such as Sub 05, who displayed an improvement in section E of the GMFM-88 from 36 to 49 points and a worsening in walking speed (as captured by the 10 MWT). In contrast, others showed large improvements in several clinical outcomes. For instance, Sub 12 showed improvements in sections D and E of the GMFM-88, in the 6 MWT score, and in the EVGS score.

Exploring the relationship between muscle synergies and clinical outcomes of RGT

We estimated the Kendall correlation between the changes in muscle synergies in response to RGT (Tables 2 and 3) and clinical outcomes (Table 4) and obtained p-values ranging between 0.11 and 0.45 (i.e., not significant) and correlation values ranging between 0.06 and 0.12 (i.e., low correlation) for different clinical outcomes. However, visual inspection of the tables showed that some participants displayed a large change in muscle synergy characteristics in response to RGT as well as large motor gains.

For instance, Sub 12 showed two muscle synergies pre-RGT and three muscle synergies post-RGT for both the most affected and the contralateral leg. The two synergies that were present pre-RGT showed positive changes in cosine similarity for both the most affected and the contralateral leg. This participant displayed motor gains across four clinical dimensions. Sub 05 also displayed two muscle synergies pre-RGT. However, the number of muscle synergies did not change post-RGT. No noticeable changes in cosine similarity and ZLCC values for the synergies of the most affected leg were observed. This participant showed an improvement only in one clinical outcome (i.e., section E of the GMFM scale). These observations suggest that, at least in some patients, a prominent change in muscle synergy characteristics leads to large motor gains. Vice versa, small changes or no change in muscle synergy characteristics appear to be associated with marginal motor gains.

When we attempted to extend such a consideration to the rest of the study group, we observed that great variability across individuals in the relationship between changes in muscle synergies in response to RGT and motor gains. All the remaining study participants displayed three muscle synergies, both pre- and post-RGT. All of them showed clinical improvements in at least two of the clinical outcomes tracked in the study. Because we had observed different muscle synergy characteristics for different GMFCS levels (Fig. 4), we explored potential relationships between changes in muscle synergy characteristics in response to RGT and clinical outcomes separately for each GMFCS level. Also, because we observed more consistent changes in response to RGT for the most-affected leg, we focused on the data for this leg.

The two level I study participants (i.e., Sub 09 and 10) displayed different changes in muscle synergies in response to RGT. Sub 09 showed improvements in cosine similarity and ZLCC values for both Syn 1 and Syn 2. In contrast, Sub 10 did not show improvements in cosine similarity in response to RGT and showed a moderate improvement in ZLCC for Syn 2. Sub 09 showed improvements in three clinical scale scores, whereas Sub 10 displayed improvements in only two clinical scale scores.

Five of the study participants (i.e., Sub 01, 03, 04. 06, and 08) were classified as level II according to the GMFCS. Sub 01, 06, and 08 showed improvements in response to RGT in cosine similarity for two synergies, whereas Sub 04 did not show any improvements in cosine similarity but displayed improvements in ZLCC for two synergies. Sub 03 displayed an improvement in cosine similarity for one synergy and no improvements in ZLCC values. Changes in the GMFM scale results were fairly consistent in these study participants (i.e., all five participants displayed an improvement in at least one of the two GMFM sections considered in the study, with Sub 03, 04, and 06 showing improvements in both). Sub 06 showed gains across the greatest number of clinical dimensions (four out of five) and was the only one who displayed improvements in cosine similarity for both Syn 1 and Syn 2.

Three of the study participants classified as level III according to the GMFCS displayed three muscle synergies both pre- and post-RGT. In response to RGT, only Sub 13 showed an improvement in cosine similarity (for Syn 1), whereas Sub 02 showed improvements in ZLCC for all three synergies, and Sub 07 showed improvements in ZLCC for two out of three synergies. All three participants showed improvements in at least one of the sections of the GMFM tracked in the study and in EVGS. Changes in 10 MWT and 6 MWT were less consistent across level III participants, with Sub 07 displaying a worsening in these outcomes (possibly because of difficulties experienced by the child with following instructions during the post-RGT session). Sub 02 showed improvements in both. Sub 13 showed an improvement in 6 MWT, but not in 10 MWT.

Finally, we had one level IV participant (i.e., Sub 11). Improvements in response to RGT were observed in three clinical scales (sections D and E of the GMFM scale and 6 MWT). The clinical response was accompanied by changes in cosine similarity for all three synergies and ZLCC values for two out of three synergies of the most affected leg.

Discussion

We carried out this study to investigate if changes in the characteristics of the muscle synergies pre- vs. post-RGT could account for the variability in the outcomes of RGT that our research group and others observed in previous studies [8, 18,19,20,21,22]. The great majority of the literature has largely neglected consideration of the variability in the response to RGT among children with CP. Most studies have been focused on comparing different intervention modalities based on the average effects observed in samples of the target patient population [62]. Herein we took a radically different approach, namely we focused on a single intervention modality (i.e., RGT) and attempted to analyze differences in the response to RGT across study participants. We performed group analyses to assess if a group response could be detected. This provided us with confidence that our sample of children with CP was representative of the CP population. Then we explored individual differences in the response to RGT.

First, we assessed if we could identify relationships between the characteristics of the muscle synergies in children with CP and the severity of their functional limitations as captured by the GMFCS levels. Such a relationship was previously suggested by Tang et al. [26] based on a qualitative analysis of the muscle synergies. In this study, we took a quantitative approach based on the cosine similarity between the composition of normative muscle synergies and the synergies observed in children with CP. In addition, we estimated the ZLCC between the temporal coefficients of normative muscle synergies and those of the synergies observed in children with CP. We observed a decrease in cosine similarity and in ZLCC with an increase in the severity of functional limitations, namely an increase in GMFCS level. By fitting a mixed-effects regression model to the data of each synergy, we showed a statistically significant correlation between muscle synergy characteristics and GMFCS level for all the parameters considered in the study. In other words, we showed that the severity of functional limitations is associated with the degree of disarrangement of the muscle synergies, both in their composition and in their temporal activation. This finding is consistent with previous studies by Steele’s group based on the Walking Dynamic Motor Control (Walk-DMC) index [31, 35]. However, our work shows that functional limitations affect individual muscle synergies. In contrast, the Walk-DMC index is meant to account for the characteristics of all the muscle synergies at once. In line with previous work by Safavynia et al. [63], we argue that the analysis of individual synergies could be highly relevant to the clinical decision process concerning the design of RGT interventions. Such analysis could allow clinicians to target specific muscle groups and the timing of their activation during the gait cycle.

Subsequently, we analyzed the changes in muscle synergies observed in response to RGT. Changes in muscle synergies (although often modest in magnitude) were previously observed in children with CP pre vs. post orthopedic surgery [28, 30,31,32], botulinum toxin type A injections [28, 30, 33], selective dorsal rhizotomy [28,29,30,31], and conservative treatment (physical therapy) [30]. In our study with focus on RGT, we observed different patterns of change in muscle synergies among study participants. In one participant, we observed a change in the number of muscle synergies from two to three and a dramatic improvement in clinical outcomes pre- vs. post-intervention. In most cases, we observed three muscle synergies at baseline and changes in the synergy composition and/or temporal coefficients. These changes varied from participant to participant showing highly complex response across individuals and GMFCS levels. However, our analysis of the results based on a Chi Square test highlighted a statistically significant improvement in the composition of the Syn 1 pre vs post-RGT.

Albeit no significant association was detected via Kendall correlation tests, we observed several instances in which motor gains across multiple clinical outcomes occurred together with changes in several muscle synergies that displayed an improvement either in cosine similarity with the normative synergies or in ZLCC values. To further explore the relationship between changes in muscle synergies and clinical outcomes, we suggest carrying out future studies using a sample size of sufficient magnitude to allow one to explore if clusters of individuals displaying a similar association between muscle synergies and clinical outcomes could be identified. In other words, we suggest that cluster analysis techniques might be able to identify stereotypic responses to RGT in subsets of children with CP. Furthermore, we suggest carrying out future studies by monitoring the muscle synergies recruited by study participants during RGT and encouraging the use of synergies that are as close as possible to normative synergies. This could be achieved by processing EMG recordings collected during RGT, estimating the muscle synergies recruited by each participant, estimating the cosine similarity and ZLCC using normative synergies as reference, and generating feedback accordingly. Feedback could be provided to therapists with oversight of the RGT session so that training parameters (e.g., level of bodyweight support provided by the robotic system [64]) could be adjusted accordingly. Also, feedback could be provided to participants by generating visual and/or auditory feedback or forces generated by the robot to resist movement when participants recruit synergies with low cosine similarity and ZLCC values and to facilitate movement when participants recruit synergies with high cosine similarity and ZLCC values. This could be particularly suitable in children with CP displaying severe selective motor control impairment. The above-mentioned feedback modalities could help “break” aberrant synergies (associated with severe selective motor control impairment) while encouraging the recruitment of “physiological” (i.e., normative) synergies.

The main limitations of this study include the small sample size, the heterogeneity of the participants, and the fact that the normative muscle synergies were not derived from a sample of individuals matching the age and gender of the children with CP undergoing RGT. Furthermore, due to the limited sample size, the heterogeneity of the participants, and the large number of tests required to investigate the characteristics of the muscle synergies, we opted for not adjusting for multiple comparisons the p-values obtained from the performed statistical tests. These limitations should be considered when designing future clinical trials to further explore the relationship between muscle synergy characteristics and clinical outcomes of RGT. For instance, it is possible that the heterogeneity of the sample used in our study might have played a confounding factor. However, it should be noted that the inclusion of a heterogeneous population provided us with some advantages in the context of our exploratory study. Whereas it made more difficult to achieve statistical significance on a group basis (e.g., for the Kendall correlation analyses), it provided access to a larger variety of muscle synergy characteristics and hence maximized the likelihood of identifying individual cases in which changes in muscle synergies were associated with large motor gains. Future studies will need to rely on a larger sample size to expand upon the analyses performed in this preliminary trial and enable the analysis of covariates as well as the use of the clustering techniques mentioned above. Furthermore, whereas we believe that the use of data collected from healthy adults was appropriate in the context of the preliminary study herein reported, an appropriate sample of pediatric data should be used to generate the normative synergies in future studies.

Conclusions

The results of our study showed a significant variability in motor gains observed in the response to RGT among children with CP. Our findings also highlighted that, at least in a subset of children with CP, motor gains in response to RGT are associated with changes in muscle synergies leading to an increase in their similarity to normative synergies. Future studies should be performed to explore the relationship between motor gains and changes in muscle synergies in response to RGT using a large sample size hence enabling statistical analyses that explore covariates and the application of clustering techniques to identify subgroups of children with CP that display a similar response to RGT hence enabling within-cluster statistical analyses. The identification of baseline characteristics of the patient’s muscle synergies that are predictive of a large response to RGT could be enabled by such analyses and lead to personalized intervention strategies, which we hope would lead to better motor gains.

Availability of data and materials

The datasets analyzed during the current study are available from the corresponding author on reasonable request.

References

  1. Graham HK, Rosenbaum P, Paneth N, Dan B, Lin JP, Damiano DL, et al. Cerebral palsy. Nat Rev Dis Primers. 2016;2:15082.

    Article  PubMed  Google Scholar 

  2. Rose J. Selective motor control in spastic cerebral palsy. Dev Med Child Neurol. 2009;51(8):578–9.

    Article  PubMed  Google Scholar 

  3. Varni JW, Burwinkle TM, Sherman SA, Hanna K, Berrin SJ, Malcarne VL, et al. Health-related quality of life of children and adolescents with cerebral palsy: hearing the voices of the children. Dev Med Child Neurol. 2005;47(09):592.

    PubMed  Google Scholar 

  4. Maher CA, Toohey M, Ferguson M. Physical activity predicts quality of life and happiness in children and adolescents with cerebral palsy. Disabil Rehabil. 2016;38(9):865–9.

    Article  PubMed  Google Scholar 

  5. Sheu J, Cohen D, Sousa T, Pham KLD. Cerebral palsy: current concepts and practices in musculoskeletal care. Pediatr Rev. 2022;43(10):572–81.

    Article  PubMed  Google Scholar 

  6. Rethlefsen SA, Blumstein G, Kay RM, Dorey F, Wren TAL. Prevalence of specific gait abnormalities in children with cerebral palsy revisited: influence of age, prior surgery, and gross motor function classification system level. Dev Med Child Neurol. 2017;59(1):79–88.

    Article  PubMed  Google Scholar 

  7. Armand S, Decoulon G, Bonnefoy-Mazure A. Gait analysis in children with cerebral palsy. EFORT Open Rev. 2016;1(12):448–60.

    Article  PubMed  PubMed Central  Google Scholar 

  8. Meyer-Heim A, Ammann-Reiffer C, Schmartz A, Schäfer J, Sennhauser FH, Heinen F, et al. Improvement of walking abilities after robotic-assisted locomotion training in children with cerebral palsy. Arch Dis Child. 2009;94(8):615–20.

    Article  CAS  PubMed  Google Scholar 

  9. Borggraefe I, Schaefer JS, Klaiber M, Dabrowski E, Ammann-Reiffer C, Knecht B, et al. Robotic-assisted treadmill therapy improves walking and standing performance in children and adolescents with cerebral palsy. Eur J Paediatr Neurol. 2010;14(6):496–502.

    Article  PubMed  Google Scholar 

  10. Schroeder AS, Homburg M, Warken B, Auffermann H, Koerte I, Berweck S, et al. Prospective controlled cohort study to evaluate changes of function, activity and participation in patients with bilateral spastic cerebral palsy after Robot-enhanced repetitive treadmill therapy. Eur J Paediatr Neurol. 2014;18(4):502–10.

    Article  CAS  PubMed  Google Scholar 

  11. van Hedel HJA, Meyer-Heim A, Rüsch-Bohtz C. Robot-assisted gait training might be beneficial for more severely affected children with cerebral palsy. Dev Neurorehabil. 2016;19(6):410–5.

    Article  PubMed  Google Scholar 

  12. Bayon C, Raya R. Robotic therapies for children with cerebral palsy: a systematic review. Transl Biomed. 2016;7(1):1–10.

    Article  Google Scholar 

  13. Carvalho I, Pinto SM, das Chagas DV, Praxedes Santos JL, de Sousa Oliveira T, Batista LA. Robotic gait training for individuals with cerebral palsy: a systematic review and meta-analysis. Arch Phys Med Rehabil. 2017;98(11):2332–44.

    Article  PubMed  Google Scholar 

  14. Bonanno M, Militi A, La Fauci BF, De Luca R, Leonetti D, Quartarone A, et al. Rehabilitation of gait and balance in cerebral palsy: a scoping review on the use of robotics with biomechanical implications. J Clin Med. 2023;12:3278.

    Article  PubMed  PubMed Central  Google Scholar 

  15. Peri E, Turconi AC, Biffi E, Maghini C, Panzeri D, Morganti R, et al. Effects of dose and duration of Robot-Assisted Gait Training on walking ability of children affected by cerebral palsy. Technol Health Care. 2017;25(4):671–81.

    Article  PubMed  Google Scholar 

  16. Wallard L, Dietrich G, Kerlirzin Y, Bredin J. Robotic-assisted gait training improves walking abilities in diplegic children with cerebral palsy. Eur J Paediatr Neurol. 2017;21(3):557–64.

    Article  CAS  PubMed  Google Scholar 

  17. Klobucká S, Klobucký R, Kollár B. Effect of robot-assisted gait training on motor functions in adolescent and young adult patients with bilateral spastic cerebral palsy: a randomized controlled trial. NeuroRehabilitation. 2020;47(4):495–508.

    PubMed  Google Scholar 

  18. Borggraefe I, Meyer-Heim A, Kumar A, Schaefer JS, Berweck S, Heinen F. Improved gait parameters after robotic-assisted locomotor treadmill therapy in a 6-year-old child with cerebral palsy. Mov Disord. 2008;23(2):280–3.

    Article  PubMed  Google Scholar 

  19. Sicari M, Patritti BL, Deming LC, Romaguera F, Pelliccio M, Nimec D, et al. Robotic gait training in children with cerebral palsy: a case series. Gait Posture. 2009;30:S2.

    Article  Google Scholar 

  20. Patritti BL, Sicari M, Deming LC, Romaguera F, Pelliccio MM, Kasi P, et al. The role of augmented feedback in pediatric robotic-assisted gait training: a case series. Technol Disabil. 2010;22(4):215–27.

    Article  Google Scholar 

  21. Smania N, Bonetti P, Gandolfi M, Cosentino A, Waldner A, Hesse S, et al. Improved gait after repetitive locomotor training in children with cerebral palsy. Am J Phys Med Rehabil. 2011;90(2):137–49.

    Article  PubMed  Google Scholar 

  22. Bayon C, Ramirez O, Velasco M, Serrano JI, Lara SL, Martinez-Caballero I, et al. Pilot study of a novel robotic platform for gait rehabilitation in children with cerebral palsy. In: Proceedings of the IEEE RAS and EMBS international conference on biomedical robotics and biomechatronics. 2016; 2016-July. p. 882–7.

  23. d’Avella A, Saltiel P, Bizzi E. Combinations of muscle synergies in the construction of a natural motor behavior. Nat Neurosci. 2003;6(3):300–8.

    Article  PubMed  Google Scholar 

  24. Bekius A, Bach MM, van der Krogt MM, de Vries R, Buizer AI, Dominici N. Muscle synergies during walking in children with cerebral palsy: a systematic review. Front Physiol. 2020;11:632.

    Article  PubMed  PubMed Central  Google Scholar 

  25. Zwaan E, Becher JG, Harlaar J. Synergy of EMG patterns in gait as an objective measure of muscle selectivity in children with spastic cerebral palsy. Gait Posture. 2012;35(1):111–5.

    Article  PubMed  Google Scholar 

  26. Tang L, Li F, Cao S, Zhang X, Wu D, Chen X. Muscle synergy analysis in children with cerebral palsy. J Neural Eng. 2015;12(4): 046017.

    Article  PubMed  Google Scholar 

  27. Hashiguchi Y, Ohata K, Osako S, Kitatani R, Aga Y, Masaki M, et al. Number of synergies is dependent on spasticity and gait kinetics in children with cerebral palsy. Pediatr Phys Ther. 2018;30(1):34–8.

    Article  PubMed  Google Scholar 

  28. Shuman BR, Goudriaan M, Desloovere K, Schwartz MH, Steele KM. Muscle synergies demonstrate only minimal changes after treatment in cerebral palsy. J Neuroeng Rehabil. 2019;16(1):46.

    Article  PubMed  PubMed Central  Google Scholar 

  29. Oudenhoven L, Van Der KM, Buizer A, Dominici N, Harlaar J. Selective motor control before and after selective dorsal rhizotomy in ambulant children with cerebral palsy. Gait Posture. 2016;49:29.

    Article  Google Scholar 

  30. Shuman BR, Goudriaan M, Desloovere K, Schwartz MH, Steele KM. Associations between muscle synergies and treatment outcomes in cerebral palsy are robust across clinical centers. Arch Phys Med Rehabil. 2018;99(11):2175–82.

    Article  PubMed  PubMed Central  Google Scholar 

  31. Schwartz MH, Rozumalski A, Steele KM. Dynamic motor control is associated with treatment outcomes for children with cerebral palsy. Dev Med Child Neurol. 2016;58(11):1139–45.

    Article  PubMed  PubMed Central  Google Scholar 

  32. Loma-Ossorio García M, Torricelli D, Moral Saiz B, Parra Mussin EM, Martín Lorenzo T, Barroso F, et al. Changes in modular control of gait following SEMLS in children with cerebral palsy. Gait Posture. 2015;42:S56.

    Article  Google Scholar 

  33. Van Der KM, Oudenhoven L, Buizer A, Dallmeijer A, Dominici N, Harlaar J. The effect of EMG processing choices on muscle synergies before and after BoNT-A treatment in cerebral palsy. Gait Posture. 2016;2016(49):31.

    Google Scholar 

  34. Conner BC, Schwartz MH, Lerner ZF. Pilot evaluation of changes in motor control after wearable robotic resistance training in children with cerebral palsy. J Biomech. 2021;126: 110601.

    Article  PubMed  PubMed Central  Google Scholar 

  35. Steele KM, Rozumalski A, Schwartz MH. Muscle synergies and complexity of neuromuscular control during gait in cerebral palsy. Dev Med Child Neurol. 2015;57(12):1176–82.

    Article  PubMed  PubMed Central  Google Scholar 

  36. Goudriaan M, Papageorgiou E, Shuman BR, Steele KM, Dominici N, Van Campenhout A, et al. Muscle synergy structure and gait patterns in children with spastic cerebral palsy. Dev Med Child Neurol. 2022;64(4):462–8.

    Article  PubMed  Google Scholar 

  37. Bizzi E, Cheung VCK. The neural origin of muscle synergies. Front Comput Neurosci. 2013;7:1–6.

    Article  Google Scholar 

  38. Lacquaniti F, Ivanenko YP, Zago M. Development of human locomotion. Curr Opin Neurobiol. 2012;22(5):822–8.

    Article  CAS  PubMed  Google Scholar 

  39. Cheung VCK, Turolla A, Agostini M, Silvoni S, Bennis C, Kasi P, et al. Muscle synergy patterns as physiological markers of motor cortical damage. Proc Natl Acad Sci. 2012;109(36):14652–6.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  40. Cheung VCK, Piron L, Agostini M, Silvoni S, Turolla A, Bizzi E. Stability of muscle synergies for voluntary actions after cortical stroke in humans. Proc Natl Acad Sci. 2009;106(46):19563–8.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  41. Shuman B, Goudriaan M, Bar-On L, Schwartz MH, Desloovere K, Steele KM. Repeatability of muscle synergies within and between days for typically developing children and children with cerebral palsy. Gait Posture. 2016;2016(45):127–32.

    Article  Google Scholar 

  42. Palisano R, Rosenbaum P, Walter S, Russell D, Wood E, Galuppi B. Development and reliability of a system to classify gross motor function in children with cerebral palsy. Dev Med Child Neurol. 1997;39(4):214–23.

    Article  CAS  PubMed  Google Scholar 

  43. Hermens HJ, Freriks B, Disselhorst-Klug C, Rau G. Development of recommendations for SEMG sensors and sensor placement procedures. J Electromyogr Kinesiol. 2000;10(5):361–74.

    Article  CAS  PubMed  Google Scholar 

  44. Russell DJ, Rosenbaum PL, Marilyn WAL. Gross motor function measure (GMFM-66 and GMFM-88) user’s manual. 2nd ed. London: Mac Keith Press; 2013.

    Google Scholar 

  45. Read HS, Hazlewood ME, Hillman SJ, Prescott RJ, Robb JE. Edinburgh visual gait score for use in cerebral palsy. J Pediatr Orthopaed. 2003;23(3):296–301.

    Article  Google Scholar 

  46. d’Avella A. Muscle synergies. In: Encyclopedia of neuroscience. Berlin: Springer, Berlin Heidelberg; 2009. p. 2509–12.

    Chapter  Google Scholar 

  47. Rabbi MF, Diamond LE, Carty CP, Lloyd DG, Davico G, Pizzolato C. A muscle synergy-based method to estimate muscle activation patterns of children with cerebral palsy using data collected from typically developing children. Sci Rep. 2022;12(1):3599.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  48. Cheung VCK, Cheung BMF, Zhang JH, Chan ZYS, Ha SCW, Chen CY, et al. Plasticity of muscle synergies through fractionation and merging during development and training of human runners. Nat Commun. 2020;11(1):4356.

    Article  PubMed  PubMed Central  Google Scholar 

  49. Devarajan K, Cheung VCK. On nonnegative matrix factorization algorithms for signal-dependent noise with application to electromyography data. Neural Comput. 2014;26(6):1128–68.

    Article  PubMed  PubMed Central  Google Scholar 

  50. Roh J, Rymer WZ, Perreault EJ, Yoo SB, Beer RF. Alterations in upper limb muscle synergy structure in chronic stroke survivors. J Neurophysiol. 2013;109(3):768–81.

    Article  PubMed  Google Scholar 

  51. Li F, Wang Q, Cao S, Wu D, Wang Q, Chen X. Lower-limb muscle synergies in children with cerebral palsy. In: 2013 6th international IEEE/EMBS conference on neural engineering (NER). IEEE; 2013. p. 1226–9.

  52. Rimini D, Agostini V, Knaflitz M. Intra-subject consistency during locomotion: similarity in shared and subject-specific muscle synergies. Front Hum Neurosci. 2017;11: 292485.

    Article  Google Scholar 

  53. d’Avella A, Portone A, Fernandez L, Lacquaniti F. Control of fast-reaching movements by muscle synergy combinations. J Neurosci. 2006;26(30):7791–810.

    Article  PubMed  PubMed Central  Google Scholar 

  54. Steele KM, Munger ME, Peters KM, Shuman BR, Schwartz MH. Repeatability of electromyography recordings and muscle synergies during gait among children with cerebral palsy. Gait Posture. 2019;67:290–5.

    Article  PubMed  Google Scholar 

  55. Patritti B, Sicari M, Deming L, Romaguera F, Pelliccio M, Benedetti MG, et al. Enhancing robotic gait training via augmented feedback. In: 2010 annual international conference of the IEEE engineering in medicine and biology society, EMBC’10; 2010. p. 2271–4.

  56. Meyer-Heim A, Borggraefe I, Ammann-Reiffer C, Berweck S, Sennhauser FH, Colombo G, et al. Feasibility of robotic-assisted locomotor training in children with central gait impairment. Dev Med Child Neurol. 2007;49(12):900–6.

    Article  CAS  PubMed  Google Scholar 

  57. Oeffinger D, Bagley A, Rogers S, Gorton G, Kryscio R, Abel M, et al. Outcome tools used for ambulatory children with cerebral palsy: responsiveness and minimum clinically important differences. Dev Med Child Neurol. 2008;50(12):918–25.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  58. Moll F, Kessel A, Bonetto A, Stresow J, Herten M, Dudda M, et al. Use of robot-assisted gait training in pediatric patients with cerebral palsy in an inpatient setting—a randomized controlled trial. Sensors. 2022;22(24):9946.

    Article  PubMed  PubMed Central  Google Scholar 

  59. Storm FA, Petrarca M, Beretta E, Strazzer S, Piccinini L, Maghini C, et al. Minimum clinically important difference of gross motor function and gait endurance in children with motor impairment: a comparison of distribution-based approaches. Biomed Res Int. 2020;2020:1–9.

    Article  Google Scholar 

  60. del Pilar Orozco M, Abousamra O, Church C, Lennon N, Henley J, Rogers KJ, et al. Reliability and validity of Edinburgh visual gait score as an evaluation tool for children with cerebral palsy. Gait Posture. 2016;2016(49):14–8.

    Article  Google Scholar 

  61. Schroeder AS, Von Kries R, Riedel C, Homburg M, Auffermann H, Blaschek A, et al. Patient-specific determinants of responsiveness to robot-enhanced treadmill therapy in children and adolescents with cerebral palsy. Dev Med Child Neurol. 2014;56(12):1172–9.

    Article  PubMed  Google Scholar 

  62. Conner BC, Remec NM, Lerner ZF. Is robotic gait training effective for individuals with cerebral palsy? A systematic review and meta-analysis of randomized controlled trials. Clin Rehabil. 2022;36(7):873–82.

    Article  PubMed  PubMed Central  Google Scholar 

  63. Safavynia S, Torres-Oviedo G, Ting L. Muscle synergies: Implications for clinical evaluation and rehabilitation of movement. Top Spinal Cord Inj Rehabil. 2011;17(1):16–24.

    Article  PubMed  Google Scholar 

  64. Kerkman JN, Zandvoort CS, Daffertshofer A, Dominici N. Body weight control is a key element of motor control for toddlers’ walking. Front Netw Physiol. 2022;2:1–12.

    Article  Google Scholar 

Download references

Acknowledgements

We would like to thank Stefano Righi, Ana Paula Quixadá, Gregory Schiurring, Anne O’Brien, and Benito Lorenzo Pugliese for their help with the study.

Funding

This work was supported by fellowships from the Fundación Alfonso Martin Escudero and the Real Colegio Complutense at Harvard (RCC), and by the Peabody Foundation and the Foundation for Physical Medicine and Rehabilitation (PM&R). Vincent C. K. Cheung was supported by grants from the Research Grants Council of the Hong Kong Special Administrative Region, China (Project Nos. R4022-18, N_CUHK456/21, 14114721, and 14119022 to VCKC).

Author information

Authors and Affiliations

  1. Department of Physical Medicine and Rehabilitation, Harvard Medical School, Spaulding Rehabilitation Hospital, Boston, MA, USA

    Gloria P. Vergara-Diaz, Stefano Sapienza, Eric Fabara, Catherine Adans-Dester, Donna Nimec & Paolo Bonato

  2. Department of Physical Medicine and Rehabilitation, Hospital Universitario Virgen del Rocío, Seville, Spain

    Gloria P. Vergara-Diaz & Carmen Echevarría Ruiz de Vargas

  3. Luxembourg Centre for Systems Biomedicine (LCSB), University of Luxembourg, Esch-Sur-Alzette, Luxembourg

    Stefano Sapienza

  4. Luxembourg Institute of Health (LIH), Strassen, Luxembourg

    Stefano Sapienza

  5. Department of Rehabilitation and Movement Sciences, Rutgers University, Newark, NJ, USA

    Jean-François Daneault

  6. School of Electrical and Electronic Engineering, University College Dublin, Dublin, Ireland

    Giacomo Severini

  7. School of Biomedical Sciences, and The Gerald Choa Neuroscience Institute, The Chinese University of Hong Kong, Hong Kong, China

    Vincent C. K. Cheung

Authors
  1. Gloria P. Vergara-Diaz
    View author publications

    You can also search for this author inPubMed Google Scholar

  2. Stefano Sapienza
    View author publications

    You can also search for this author inPubMed Google Scholar

  3. Jean-François Daneault
    View author publications

    You can also search for this author inPubMed Google Scholar

  4. Eric Fabara
    View author publications

    You can also search for this author inPubMed Google Scholar

  5. Catherine Adans-Dester
    View author publications

    You can also search for this author inPubMed Google Scholar

  6. Giacomo Severini
    View author publications

    You can also search for this author inPubMed Google Scholar

  7. Vincent C. K. Cheung
    View author publications

    You can also search for this author inPubMed Google Scholar

  8. Carmen Echevarría Ruiz de Vargas
    View author publications

    You can also search for this author inPubMed Google Scholar

  9. Donna Nimec
    View author publications

    You can also search for this author inPubMed Google Scholar

  10. Paolo Bonato
    View author publications

    You can also search for this author inPubMed Google Scholar

Contributions

G.V.D., S.S., J.F.D., G.S., and P.B. contributed to the design of the study. G.V.D., J.F.D., E.F., C.A.D., D.N., and P.B. contributed to the implementation of the research protocols. S.S., G.V.D., E.F., and P.B. curated and analyzed the data. G.V.D., S.S., J.F.D., G.S., V.C., C.E.R.V., D.N., and P.B. contributed to the interpretation of the data. G.V.D., S.S., and P.B. wrote the manuscript. All the authors contributed substantially to the critical revision of the manuscript and approved its final version.

Corresponding author

Correspondence to Paolo Bonato.

Ethics declarations

Ethics approval and consent to participate

The research study presented in the manuscript was approved by the Spaulding Rehabilitation Hospital Institutional Review Board (protocol # 2015P001482, clinical trial # NCT06156969). All parents or guardians signed a consent form, and children signed an assent form.

The research study carried out to gather normative muscle synergies was also approved by the Spaulding Rehabilitation Hospital Institutional Review Board (protocol # 2019P002419). All participants signed a consent form.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Vergara-Diaz, G.P., Sapienza, S., Daneault, JF. et al. Can muscle synergies shed light on the mechanisms underlying motor gains in response to robot-assisted gait training in children with cerebral palsy?. J NeuroEngineering Rehabil 22, 23 (2025). https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s12984-025-01550-x

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s12984-025-01550-x

Keywords

Journal of NeuroEngineering and Rehabilitation

ISSN: 1743-0003

Contact us