Background
Safe and independent locomotion represents a fundamental motor function for humans that is essential for self-contained living and good quality of life [
1‐
5]. Locomotion requires the ability to coordinate a number of different muscles acting on different joints [
6‐
8], which are guided by cortical and subcortical brain structures within the locomotor network [
9]. Structural and functional changes within the locomotor network are often accompanied by gait and balance impairments which are frequently considered to be the most significant concerns in individuals suffering from brain injuries or neurological diseases [
5,
10,
11]. Reduced walking speeds and step lengths [
12] as well as non-optimal amount of gait variability [
13‐
15] are common symptoms associated with gait impairments that increase the risk of falling [
16].
In addition to manual-assisted therapy, robotic neurorehabilitation has often been applied in recent years [
17,
18] because it provides early, intensive, task-specific and multi-sensory training which is thought to be effective for balance and gait recovery [
17,
19,
20]. Depending on the severity of the disease, movements can be completely guided or assisted, tailored to individual needs [
17], using either stationary robotic systems or wearable powered exoskeletons.
Previous studies investigated the effectiveness of robot-assisted gait training (RAGT) in patients suffering from stroke [
21,
22], multiple sclerosis [
23‐
26], Parkinson’s disease [
27,
28], traumatic brain injury [
29] or spinal cord injury [
30‐
32]. Positive effects of RAGT on walking speed [
33,
34], leg muscle force [
23] step length, and gait symmetry [
29,
35] were reported. However, the results of different studies are difficult to summarize due to the lack of consistency in protocols and settings of robotic-assisted treatments (e.g., amount and frequency of training sessions, amount and type of provided robotic support) as well as fragmentary knowledge of the effects on functional brain reorganization, motor recovery and their relation [
36,
37]. Therefore, it is currently a huge challenge to draw guidelines for robotic rehabilitation protocols [
22,
36‐
38]. To design prologned personalized training protocols in robotic rehabilitation to maximize individual treatment effects [
37], it is crucial to increase the understanding of changes in locomotor patterns [
39] and brain signals [
40] underlying RAGT and how they are related [
36,
41].
A series of studies investigated the effects of robotic assistance (RA) on biomechanical gait patterns in healthy people [
39,
42‐
44]. On one side, altered gait patterns were reported during robot-assisted walking (RAW) compared to unassisted walking (UAW), in particular, substantially higher muscle activity in the quadriceps, gluteus and adductor longus leg muscles and lower muscle activity in the gastrocnemius and tibialis anterior ankle muscles [
39,
42] as well as reduced lower-body joint angles due to the little medial-lateral hip movements [
45‐
47]. On the other side, similar muscle activation patterns were observed during RAW compared to UAW [
44,
48,
49], indicating that robotic devices allow physiological muscle activation patterns during gait [
48]. However, it is hypothesized that the ability to execute a physiological gait pattern depends on how the training parameters such as body weight support (BWS), guidance force (GF) or kinematic restrictions in the robotic devices are set [
44,
48,
50]. For example, Aurich-Schuler et al. [
48] reported that the movements of the trunk and pelvis are more similar to UAW on a treadmill when the pelvis is not fixed during RAW, indicating that differences in musle activity and kinematic gait characteristics between RAW and UAW are due to the reduction in degrees of freedom that user’s experience while walking in the robotic device [
45]. In line with this, a clinical concern that is often raised with respect to RAW is the lack of gait variability [
45,
48,
50]. It is assumed that since the robotic systems are often operated with 100% GF, which means that the devices attempt to force a particular gait pattern regardless of the user’s intentions, the user lacks the ability to vary and adapt his gait patterns [
45]. Contrary to this, Hidler et al. [
45] observed differences in kinematic gait patterns between subsequent steps during RAW, as demonstrated by variability in relative knee and hip movements. Nevertheless, Gizzi et al. [
49] showed that the muscular activity during RAW was clearly more stereotyped and similar among individuals compared to UAW. They concluded that RAW provides a therapeutic approach to restore and improve walking that is more repeatable and standardized than approaches based on exercising during UAW [
49].
In addition to biomechanical gait changes, insights into brain activity and intervention-related changes in brain activity that relate to gait responses, will contribute to the optimization of therapy interventions [
41,
51]. Whereas the application of functional magnetic resonance imaging (fMRI), considered as gold standard for the assessment of activity in cortical and subcortical structures, is restricted due to the vulnerability for movement artifacts and the range of motion in the scanner [
52], functional near infrared spectroscopy (fNIRS) is affordable and easily implementable in a portable system, less susceptible to motion artifacts, thus facilitation a wider range of application with special cohorts (e.g., children, patients) and in everyday environments (e.g., during a therapeutic session of RAW or UAW) [
53,
54]. Although with lower resolution compared to fMRI [
55], fNIRS also relies on the principle of neurovascular coupling and allows the indirect evaluation of cortical activation [
56,
57] based on hemodynamic changes which are analogous to the blood-oxygenation-level-dependent responses measured by fMRI [
56]. Despite limited depth sensitivity, which restricts the measurement of brain activity to cortical layers, it is a promising tool to investigate the contribution of cortical areas to the neuromotor control of gross motor skills, such as walking [
53]. Regarding the cortical correlates of walking, numerous studies identified either increaesed oxygenated hemoglobin (Hboxy) concentration changes in the sensorimotor cortex (SMC) by using fNIRS [
53,
57‐
59] or suppressed alpha and beta power in sensorimotor areas by using electroencephalography (EEG) [
60‐
62] demonstrating that motor cortex and corticospinal tract contribute directly to the muscle activity of locomotion [
63]. However, brain activity during RAW [
36,
61,
64‐
68], especially in patients [
69,
70] or by using fNIRS [
68,
69], is rarely studied [
71].
Analyzing the effects of RA on brain activity in healthy volunteers, Knaepen et al. [
36] reported significantly suppressed alpha and beta rhythms in the right sensory cortex during UAW compared to RAW with 100% GF and 0% BWS. Thus, significantly larger involvement of the SMC during UAW compared to RAW were concluded [
36]. In contrast, increases of Hboxy were observed in motor areas during RAW compared UAW, leading to the conclusion that RA facilitated increased cortical activation within locomotor control systems [
68]. Furthermore, Simis et al. [
69] demonstrated the feasibility of fNIRS to evaluate the real-time activation of the primary motor cortex (M1) in both hemispheres during RAW in patients suffering from spinal cord injury. Two out of three patients exhibited enhanced M1 activation during RAW compared with standing which indicate the enhanced involvement of motor cortical areas in walking with RA [
69].
To summarize, previous studies mostly focused the effects of RA on either gait characteristics or brain activity. Combined measurements investigating the effects of RA on both biomechanical and hemodynamic patterns might help for a better understanding of the neurophysiological mechanisms underlying gait and gait disorders as well as the effectiveness of robotic rehabilitation on motor recovery [
37,
71]. Up to now, no consensus exists regarding how robotic devices should be designed, controlled or adjusted (i.e., device settings, such as the level of support) for synergistic interactions with the human body to achieve optimal neurorehabilitation [
37,
72]. Therefore, further research concerning behavioral and neurophysiological mechanisms underlying RAW as well as the modulatory effect of RAGT on neuroplasticy and gait recovery are required giving the fact that such knowledge is of clinical relevance for the development of gait rehabilitation strategies.
Consequently, the central purpose of this study was to investigate both gait characteristics and hemodynamic activity during RAW to identify RAW-related changes in brain activity and their relationship to gait responses. Assuming that sensorimotor areas play a pivotal role within the cortical network of automatic gait [
9,
53] and that RA affects gait and brain patterns in young, healthy volunteers [
39,
42,
45,
68], we hypothesized that RA result in both altered gait and brain activity patterns. Based on previous studies, more stereotypical gait characteristics with less inter- and intraindividual variability are expected during RAW due to 100% GF and the fixed pelvis compared to UAW [
45,
48], wheares brain activity in SMC can be either decreased [
36] or increased [
68].
Discussion
In this study, the effects of RA on cortical activity during TW and the relationship to changes in gait characteristics were investigated. We identified a classical double bump in the GRF, throughout the stance phase during both UAW and RAW, which was accompanied by significantly increased brain activity in the SMC compared to premotor/supplementary motor areas. However, individual analyses showed significantly higher inter- and intraindividual gait variability due to RA that correlated with increased hemodynamic activity in the SMC (p = 0.052; r = 0.570).
In both conditions, shape characteristics of the mean GRF curves during the stance phase were observed. This in not in line with the results of Neckel et al. [
46] who did not report a classical double bump during the stance phase during RAW, which could be due to the age differences of our samples. Furthermore, significantly altered kinematic patterns (lower GRF values and earlier and later appearances for the first and second vertical GRF peak values, respectively) as well as large inter- and intraindividual gait variability were observed during RAW compared to UAW. Results of the kinematic patterns are consistent with other biomechanical studies showing altered muscle activity [
39,
42] or kinematic patterns [
45‐
47] due to RA. The results of greater inter- and intraindividual gait variability during RAW do not agree with the more stereotypical and similar patterns of Gizzi et al. [
49], nor with the assumption that the user lacks the ability to vary and adapt gait patterns during RAW [
45,
48,
50].
Regarding brain activity during UAW, Hboxy concentration changes were significantly increased in sensorimotor areaes compared to areas of the SMA/PMC which is in line with other neurophysiological studies that showed increased Hboxy concentrations during walking [
57,
58]. This is further confirmed by EEG studies reporting suppressed alpha and beta oscillations within the SMC [
60‐
62] during active walking. This also demonstrates that the SMC and the corticospinal tract contribute directly to muscle activity in locomotion [
9,
53,
63] representing a general marker of an active movemet-related neuronal state [
61].
Analyzing the effects of RA on cortical patterns, significantly increased Hboxy concentration changes were also observed in SMC compared to frontal areas. Whereas Kim et al. [
68] observed more global network activation during RAW compared to UAW, Knaepen et al. [
36] reported significantly suppressed alpha and beta power during UAW compared to RAW with the conclusion that walking with 100% GF leads to less active participation and little activation of the SMC, which should be avoided during RAGT.
However, during RAW, we observed a positive correlation between ΔHboxy concentrations in the SMC and intraindividual gait variability. Thus, individuals with larger gait variability showed higher sensorimotor brain activity, which is similar to the results reported of Vitorio et al. [
41]. In this study, positive correlations between gait variability and ΔHboxy in the PMC and M1 were found in young healthy adults when walking with rhythmic auditory cueing [
41]. The following two possible explanations are suggested.
On one side, robotic guidance might induce additional and new sensory feedback that promotes active participation, resulting in high gait variability and increased brain activity. This possibility is supported by previous observations that muscles exhibited marked and structurally phased activity, even under full guidance conditions [
39,
42,
86‐
88]. Van Kammen et al. [
88] found muscle activity in the vastus lateralis, suggesting that the leg muscles are still activated during RAW as opposed to the muscles related to stability and propulsion, in which activity is reduced under guidance conditions. This finding is remarkable because, in this state, the exoskeleton is responsible for walking control, and theoretically, no voluntary activity from the performer is required [
87,
89]. However, the instructions used in the present study (i.e., ‘actively move along with the device’) may have affected activity, as previous studies have shown that encouraging active involvement increases muscle activity [
86,
87] as well as brain activity significantly during RAW [
64]. More specifically, Wagner et al. [
64] showed significantly suppressed alpha and beta power during active compared to passive RAW. Dobkin (1994) also showed that passive stepping can lead to task-specific sensory information that induces and modulates step-like electromyography activity [
90]. Thus, high guidance might also promote active contribution. Particularly in patients who are not able to walk unassisted, successful stepping induces task-specific sensory information that may trigger plastic changes in the central nervous system [
88,
91]. Since active participation and the production of variable movement patterns are prerequisites for activity-dependent neuroplasticity [
7,
20,
89,
92‐
94], it is important to determine whether the activation of the SMC can be triggered by changes in the levels of GF, BWS and kinematic freedom in order to specifically provoke gait variability due to active participation of the patient [
45,
48,
50]. High gait variability may indicate that people use multiple combinations of gait variables to walk more effectively [
45,
95], resulting in better and faster improvements during robotic rehabilitation.
On other side, the sensory feedback from robot guidance could also disturb the brain network underlying automatic walking, leading to increased gait variability and sensorimotor activity. According to Vitorio et al. [
41], the requirement to adapt to external stimuli leads to disturbances in automatic walking in young healthy people, resulting in higher gait variability and higher cortical costs. As previous study have shown, the ability to execute a physiological gait pattern depends on how the training parameters such as BWS, GF or kinematic freedom in the robotic devices are set. During RAW with fixed pelvis, significantly altered muscle activity [
39,
42,
45] and kinematic patterns [
48,
50] were found. In addition to GF, BWS and kinematic freedom, the presence of foot support may also contribute to altered patterns. The safety procedures of the therapy institution required that all subjects wear straps around the front foot to assist with ankle dorsiflexion, which is known to reduce activity in the ankle dorsiflexors [
39,
42].
In summary, increased gait variability and sensorimotor activity during RAW could be the result of active participation or disrupted automatic locomotor control. However, the generalization of these results to other populations is not intended or recommended. Healthy elderly individuals [
41] and patients with stroke [
22], multiple sclerosis [
23,
25,
26], Parkinson’s disease [
27,
28], brain injuries [
29] or spinal cord injuries [
30,
31] who suffer from gait and balance disorders react differently to robotic support than healthy young people, which may lead to different gait and brain activation patterns [
44]. In addition to high inter- and intraindividual variability within one sample, the heterogeneity of methodological procedures between studies appears to pose another challenge [
71].
Therefore, one future goal should be to understand the mechanisms underlying RAGT and which parameters determine the effectiveness of a single treatment in the heterogenuous population of patients suffering from neurological diseases [
37]. For this purpose, objective biomarkers for motor recovery and neuroplastic changes have to be identified [
37]. Then, specific training protocols and further interventions, such as augmented feedback with virtual reality, brain-machine interface or non-invasive brain stimulation, can be developed to deliver sustainable therapies for individualized rehabilitation that optimizes the outcome and efficacy of gait recovery, which together can foster independent living and improve the quality of life for neurological patients [
37,
71].
Methodological limitations
Two methodological limitations that emerged using the present approach should be mentioned. First, the ability to walk is guided by an optimal interaction between cortical and subcortical brain structures within the locomotor network [
53]. Using our NIRSport system, we were only able to report brain activity patterns in motor cortical areas and were unable to monitor the activities of subcortical areas or other cortical involvements. Various studies have reported that patients with gait disorders recruit additional cortical regions to manage the demands of UAW and RAW, due to structural and/or functional changes in the brain. Measuring the entire cortical network underlying locomotion may be necessary to investigate neuronal compensations and cognitive resources used for neuroplastic processes during gait rehabilitation. Therefore, we must be careful when discussing brain activity associated with other regions involved in locomotor control [
9].
Secondly, we must take into account the small sample size of our healthy volunteers and their young age (mean: 25 ± 4 years), which also had no gait pathologies. Thus, RA guidance of gait movement might have different effects in elderly subjects or patients who are not able to walk without restrictions [
96]. Therefore, the findings from our study are difficult to apply to other age or patient groups, as neurological patients often suffer from movement disorders and therefore use different control strategies during RAW. Although the available results provide relevant insights into the mobile applications of neurophysiological measurements during RAW, with approaches for further therapeutic interventions during robotic rehabilitation, the effects of RAW must also be investigated in other groups and in patients with gait disorders in the future.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.