Comparative effects of passive and active mode robot-assisted gait training on brain and muscular activities in sub-acute and chronic stroke 1

Abstract

BACKGROUND:

Robot-assisted gait training (RAGT) was initially developed based on the passive controlled (PC) mode, where the target or ideal locomotor kinematic trajectory is predefined and a patient basically ‘rides’ the robot instead of actively participating in the actual locomotor relearning process. A new insightful contemporary neuroscience and mechatronic evidence suggest that robotic-based locomotor relearning can be best achieved through active interactive (AI) mode rather than PC mode.

OBJECTIVE:

The purpose of this study was to compare the pattern of gait-related cortical activity, specifically gait event-related spectral perturbations (ERSPs), and muscle activity from the tibialis anterior (TA) and clinical functional tests in subacute and chronic stroke patients during robot-assisted gait training (RAGT) in passive controlled (PC) and active interactive (AI) modes.

METHODS:

The present study involves a two-group pretest-posttest design in which two groups (i.e., PC-RAGT group and AI-RAGT group) of 14 stroke subjects were measured to assess changes in ERSPs, the muscle activation of TA, and the clinical functional tests, following 15– 18 sessions of intervention according to the protocol of each group.

RESULTS:

Our preliminary results demonstrated that the power in the μ band (8– 12 Hz) was increased in the leg area of sensorimotor cortex (SMC) and supplementary motor area (SMA) at post-intervention as compared to pre-intervention in both groups. Such cortical neuroplasticity change was associated with TA muscle activity during gait and functional independence in functional ambulation category (FAC) and motor coordination in Fugl– Meyer Assessment for lower extremity (FMA-LE) test as well as spasticity in the modified Ashworth scale (MAS) measures.

CONCLUSIONS:

We have first developed a novel neuroimaging experimental paradigm which distinguished gait event related cortical involvement between pre- and post-intervention with PC-RAGT and AI-RAGT in individuals with subacute and chronic hemiparetic stroke.

Keywords

Walkbot Robotic assisted gait training gait event-related spectral perturbations electroencephalography

1 Introduction

Over the past two decades, the robot-assisted gait training (RAGT) strategy has evolved from a passive to a more active or interactive challenge mode as mechatronic science and neuro-engineering technologies advance (Park et al., 2020; 2021;). RAGT was initially developed based on the passive control (PC) mode, where the target or ideal locomotor kinematic trajectory is predefined and a patient basically ‘rides’ the robot instead of actively participating in the actual locomotor relearning process (Marchal-Crespo & Reinkensmeyer, 2009). A new insightful contemporary neuroscience and mechatronic evidence suggest that robotic-based locomotor relearning can be best achieved through active interactive (AI) mode rather than PC mode (Cao et al., 2014; Fleerkotte et al., 2014; Marchal-Crespo & Reinkensmeyer, 2009). The main drawback of PC mode is that because a robot passively or adaptively guides the limb, the patient tends to ride on the robot, which considerably reduces his or her use of the involved limb force output (Wolbrecht et al., 2007) and associated neuroplasticity. The contemporary motor learning approach suggests that such a passive guidance learning strategy may be only recommended during the initial (cognitive) stage of motor learning (Dohring & Daly, 2008), at which stage the patient learns “to understand what to do”. However, as the patient progresses to the associate and automatic stage of motor learning, to maximize long-term motor control learning and neuroplasticity, more variable, with less guidance AI modes that are in some ways contrary to PC mode are recommended. Advances in sensors and control algorithms have used adaptive controller designed for robotic actuators to intrinsically detect patient’s movement capability in real time and to actively involve patients in the gait training process (Hussain et al., 2016; Veneman et al., 2007). The greater spatiotemporal gait variability and vastus medialis muscle activity in chronic patients with incomplete spinal cord injury were significantly increased during AI mode compared to PC mode, indicating more active gait participation during AI mode (Duschau-Wicke et al., 2010).

Building on the theoretical frameworks of motor learning and neurophysiology, as well as mechatronic neurorehabilitation evidence, we recently developed an innovative locomotor control algorithm of “impedance-based assistance” for the Walkbot robotic gait training system (Van Tran et al., 2015). An impedance-based control mode is an interactive resistive-assistance that enables the patient to move with or against some variable form of impedance-based resistive-assistance forces (Hussain et al., 2011). For example, the lower the impedance is set, the more resistive-assistance force or velocity is interactively or variably changed, as the patient actively attempts to reach a certain target in the ankle-knee-hip interlimb segmental kinematic and kinetic coordination level, which is concurrently provided in the real-time presentation mode in the system (Marchal-Crespo & Reinkensmeyer, 2009). Thus, low impedance control mode means AI mode that elicits greater active participation of the patients through human-robot interaction. In contrast, high impedance control mode means PC mode designed to guide the lower limbs of patients to follow a certain predefined trajectory. Nevertheless, the therapeutic effect of PC and AI modes, based on the impedance-control mode during recovery processes after neurological impairments such as stroke, have not been explored for locomotor training despite the important clinical ramifications in locomotor relearning and neuroplasticity in neurorehabilitation.

Several neuroimaging methods have been developed to measure neuroplasticity and activation of the cortical area in the brain. Specifically, the novel, sophisticated gait event-related spectral perturbations (ERSPs)– electromyography (EMG) computational algorithms were used to analyze and compare cortical and muscle activation patterns during PC and AI modes, respectively. Advanced gait ERSPs– EMG has been widely utilized to examine cortico-muscular activation patterns in previous literature because of its superior time resolution and accessibility over functional magnetic resonance imaging (fMRI) and functional near infrared spectroscopy (fNIRS) (Gwin et al., 2011; Moreno et al., 2013; Severens et al., 2012; van Kammen et al., 2016). As one cardinal feature of motor-related cortico-muscular activity between electro-encephalography (EEG) in the corresponding primary motor cortex of the contralateral hemisphere and EMG in a contracting muscle has been studied and reported to indicate synchronized oscillatory discharges derived from the neuronal cells in the descending corticospinal pathways (Kilner et al., 1999). EEG– EMG evidence demonstrated that EEG in the left motor cortex (C3) and EMG in the right first dorsal interosseous muscle were activated during the abduction of the index finger motor task (Johnson & Shinohara, 2012). Wagner et al. (2012) compared active to passive RAGT (i.e., both at 100% assistance and < 30% body weight support) and revealed that the power spectral density in the μ (mu) and β (beta) bands in the foot/leg area of the sensory cortex was significantly reduced during active compared to passive RAGT, indicating an increased activation of this area during active gait participation (Wagner et al., 2012). EMG evidence demonstrated that tibialis anterior (TA) muscle was deactivated during initial contact and swing phases while gastrocnemius muscle was overactivated resulting from cortical disinhibition and associated spasticity in individuals with hemiparetic stroke (Jacquelin Perry, 2010).

Therefore, the purpose of this study was to compare the patterns of gait-related cortical activity, specifically ERSPs, and EMG activity from TA and clinical functional tests in patients with subacute and chronic stroke during RAGT in AI mode and PC mode. We hypothesized that the AI-RAGT group would demonstrate greater cortical and TA muscle activation as well as clinical functions in the lower limb in the region of interests (ROIs) than the PC-RAGT group in the Walkbot system. The ROIs, which reflects the Brodmann area, were included according to previous findings of event-related desynchronization (ERD)/event-related synchronization (ERS) pattern of the sensorimotor cortex (SMC) and supplementary area (SMA). Imaging studies using NIRS demonstrated that walking is bilaterally associated with activity in the primary SMC and SMA (Miyai et al., 2001). The SMC area reflects imagined, planned/prepared, and executed movements as well as the TA muscle activations according to advanced EEG-EMG measurement (Gwin et al., 2011; Knaepen et al., 2015; Petersen et al., 2012; Seeber et al., 2014; Severens et al., 2012; Wagner et al., 2012).

2 Methods

2.1 Patients

A total of 20 subacute to chronic stroke patients recruited from the department of physiotherapy and rehabilitation, Myongji-choonhey Rehabilitation Hospital and Cheongdam Hospital in Seoul, and Yonsei Goodwellnes center. The study was approved by the Research Ethics Committee of the Yonsei University Mirae Campus (1041849-201906-BM-083-02) Institutional Review Board and the Myongji-choonhey Rehabilitation Hospital (CR217019) Institutional Review Board in the Republic of Korea. Written informed consent was obtained from each subject after screening for eligibility.

Inclusion criteria entailed: (1) diagnosis of a first-ever stroke (3 months to 5 years); (2) hemiplegic motor problems in lower extremity; (3) 20 years or older; (4) significant ambulation deficits (Functional Ambulation Category [FAC] < 5); (5) able to follow a three-step command; and (6) able to perform rehabilitation intervention at least 5 days per week for 3 weeks (Park et al., 2018, 2021;). Exclusion criteria included the following: (1) implanted metallic material or medical devices in the body; (2) severe spasticity (Modified Ashworth Scale [MAS] > 4); (3) severe cognitive disorder (Mini-Mental State Examination < 10); (4) difficulty in walking due to orthopedic problems (Kellegren-Lawrence grade > 3); (5) ejection fraction less than 30% due to severe heart disease or diagnosed with unstable angina; and (6) body mass index higher than 30. The patients were stratified into PC-RAGT group and AI-RAGT group based on age, sex, height, weight, onset period of stroke, and affected side.

2.2 Experimental design and procedure

The present study employed a two-group pretest-posttest design where 20 subjects were randomly assigned either into the PC- and AI- RAGT groups. A procedural checklist was followed to ensure consistent experimental protocols. Standardized clinical testing procedures, including the clinical function tests included FAC, MAS, and Fugl– Meyer Assessment for the lower extremity (FMA-LE), were consistently implemented before the testing conditions. All subjects completed the demographic and health questionnaire.

2.3 Muscle activation of Tibialis Anterior (TA)

Surface EMG (g.Nautilus, g.tec medical engineering GmbH, Schiedlberg, Austria) was used to record activity in the TA. To measure activation of the TA muscle, surface EMG electrodes were placed parallel to and medial or lateral to the medial shaft of the tibia, at approximately 1/4 to 1/3 the distance between the knee and the ankle (Criswell, 2010). EMG signals were recorded bipolarly using self-adhesive, disposable Ag/AgCl electrodes with a 10 mm diameter and a minimum electrode distance of 25 mm, placed according to the SENIAM protocol (Hermens et al., 1999). The electrode sites were prepared by removing body hair, and by abrading and cleaning the skin with alcohol. All electrode impedance was less than 50 kΩ. The average epochs were divided into seven phases according to Perry (2010) in the same way as ERSPs (Jacquelin Perry, 2010). We focused on the activation of TA muscle during swing phase. The muscle activation of TA at each time point was divided by the maximal activation in condition and normalized by multiplying by 100.

2.4 Gait Event-Related Spectral Perturbations (Gait ERSPs)

Single-trial ERSPs were determined by averaging the difference between each single-trial log spectrogram and the baseline (i.e., the mean of log spectrum over all gait cycles per condition) (Pfurtscheller & Da Silva, 1999). To illustrate event-related perturbations, significant differences from the baseline average gait cycle power spectrum were computed with a permutation method (Delorme & Makeig, 2004). To visualize intra-stride changes between PC and AI conditions, ERSPs were calculated in three frequency bands: μ (sensorimotor rhythm, SMR; 8– 12 Hz), β (12– 30 Hz) and low γ (gamma) (30– 45 Hz); using a common baseline (i.e., the average over all gait cycles) for each condition (Makeig, 1993). Typically, for n trials, if F_k (f, t) is the spectral estimate of trial k at frequency f and time t, the calculation of the ERSP can be formalized as follows: $ERSP (f, t) = \frac{1}{n} \sum_{k = 1}^{n} {| F_{k} (f, t) |}^{2}$ (1)

The relative changes in the ERSP between the PC and AI conditions were computed where the ERSP trial average is normalized for each frequency band based on the gait cycle events. The average power spectrum is first determined at each PC condition (the subject was instructed to relax while ‘riding’ on Walkbot) and the AI condition (the subject was asked to exert leg muscles to match against the impedance resistance or stiffness). The relative changes in the ERSP formula can be expressed as follows: $\begin{matrix} Relative ERSP (f, t) = \\ PC_ERSP (f, t) - AI_ERSP (f, t) \end{matrix}$ (2)

Hence, in the interpretation, the greater the relative change in the ERSP represents the greater improvement in neuroplastic changes or brain activation, which is likely to occur during AI condition than PC condition. The ERSP indicates changes in power (in dB) as a function of frequency over the time-course of the event-related potentials. Calculating an ERSP requires computing the power spectrum over a sliding latency window then averaging across data trials.

2.5 Experimental tests for neuroplasticity and muscular activation via the EEG-EMG measurement system

All stroke subjects concurrently underwent the EEG-EMG unit measurement to assess intervention-related changes in neuroplasticity and muscle activation before and after 15– 18 sessions of intervention, respectively. We used the wireless 32-channel EEG-EMG measurement unit system (g.Nautilus, g.tec medical engineering GmbH, Schiedlberg, Austria) which comprises of 24 channels of EEG and 8 channels of EMG, and active electrodes on the cap. The 10– 20 system was used to place the 24 EEG active electrode placements on the scalp (Bulea et al., 2015; Klem et al., 1999). Neurophysiological (i.e., EEG and EMG) signals were recorded by wireless that contained 24 channels of EEG (Fp1, Fp2, Fz, F3, F4, F7, F8, FC1, FC2, FC6, T7, T8, Cz, C1, C2, C3, C4, C5, CP1, CP2, P2, P3, Pz, POz) and 8 channels of EMG and the sampling rate was 500 Hz. The reference electrode was placed on the right earlobe and the ground electrode was placed at the AFz region of brain. An electrode gel was injected for every electrode, and the resistance between the skin and electrode was dropped below 50 kω for each channel. The subjects were fitted into the exoskeleton of the Walkbot_G (Walkbot_G, P&S Mechanics, Seoul, South Korea), which was adjusted to the length of subjects’ legs and fastened with three cuffs at thigh and shank. The axes of the subjects’ hip and knee joints were aligned with the axis of the Walkbot. The amount of body weight support was set at 20% for each subject, preventing excessive knee flexion during stance phase or toe dragging during swing phase (Park et al., 2020). All subjects used foot-lifters at both ankles to ensure foot clearance during swing phase. Then, wire surface EMG electrodes connected to 8 channels in the wireless cap were attached to the TA muscle of the subjects. Finally, custom-built mechanical foot-witches (3104 Interlink Electronics 0.5” circular FSR, SeedTech Inc., Bucheon, South Korea) were attached and secured under the bilateral heel pads to synchronize the 24 channel EEG– 8 channel EMG signals as the patient’s heel contacted with the treadmill surface to mark each gait cycle. The gait cycle was defined as the interval between two consecutive right initial contact (IC) of affected side and the subdivided events: affected side IC (AS-IC), unaffected side toe-off (US-TO), unaffected side IC (US-IC), affected side toe-off (AS-TO), and the subsequent AS-IC (i.e., double support [1st DS], 0– 10%; affected side single support [AS-SS], 10– 50%; 2nd DS, 50– 60%; unaffected side single support [US-SS], 60– 100%) (Hidler & Wall, 2005; Jacquelin Perry, 2010; Mummolo et al., 2013). A Simulink-based customized software (Matlab version 2015a, Mathworks Inc., Massachusetts, USA) was used to acquire the synchronized EEG and EMG data (Park et al., 2018).

The EEG and EMG data were recorded during gait on a treadmill with the exoskeleton of Walkbot system. All subjects underwent a preliminary five-minute walking period, which allowed acclimation to RAGT experimental conditions: PC and AI modes. The velocity of the treadmill was set to 1.0– 1.4 km/h individually in consideration of the function of each patient. Afterwards, recordings for 60 seconds were repeated thrice consecutively for each condition. Subjects were instructed to follow the guidance of the robotic orthosis and to relax their arm muscles and rest them on the handrails of the Walkbot in every condition. Patients with an arm brace or severe spasticity of an affected upper limb kept the arm in a comfortable position, and only the unaffected side arm was induced to rest on the handrails. To minimize movement artifacts in the EEG and EMG signals, subjects were asked to look straight ahead toward a specified point on a screen, not to close their eyes for prolonged periods of time, and to blink normally to avoid eye artifacts (Park et al., 2018; Wagner et al., 2012). To minimize fatigue, subjects rested between each measurement of the 2∼3 min intervals. Overall experimental settings at the time of measurement are presented in Fig. 1.

Fig. 1

Experimental setup.

2.6 Intervention

The RAGT intervention was consisted of 15– 18 sessions with PC-RAGT or AI-RAGT. The physical therapist, who is robotic training specialist, performed the RAGT intervention according to the intervention protocol for 25 min/session combined with conventional physiotherapy (30 min/session, 1 session/day) 2– 3 days per week over 2 months.

Anthropometric data including height, weight, foot size, thigh length, shank length, and ankle height were measured and entered into the computer patient database of Walkbot, which were used to automatically adjust the exoskeleton legs length and optimal gait cycle so as to curtail to the individual patient condition. Each participant then wore a suspension vest secured with elastic straps which was later connected to the harness mounted on the counterweight system. Depending on the patient’s clinical conditions, approximately 10– 40% (adjustable range, 0– 100%) of the total body weight was supported and then gradually decreased in 5– 10% increments. Gait velocity started at 1.0– 1.1 km/h and gradually increased by 0.1 km/h per session depending on the condition of patients, as tolerated to 1.4 km/h. During and after each session, the subjects were provided with constant verbal encouragement using the knowledge of results derived from the real time gait kinematics (joint angles) and kinetics (resistive torque, stiffness) data on the ankle, knee, and hip interlimb joint movement.

Specifically, for the PC-RAGT intervention, high impedance, which is a form of guidance force (analogue to ‘passive or full guidance’), was provided with the stiffness of Walkbot controller set to ω_H = ω_K = 3π rad/s (Md = 2.0 kgm², B_d = 107 Nms/rad, and K_d = 711 Nm/rad). For the AI-RAGT intervention, low impedance, which involves an interactive resistive training with the stiffness of Walkbot controller set to ω_H = ω_K = 0.8π rad/s (Md = 2.0 kgm², B_d = 28 Nms/rad, and K_d = 51 Nm/rad), was provided to maximize the augmented effects by means of facilitating active participation of the interlimb coordinated ankle-knee-hip joint gait related strengthening.

2.7 EEG Analysis for neural plasticky estimation

Neural plasticity was estimated by computing the ERSP power (in dB) in the μ band (8– 12 Hz) over the region of the interests which represent the leg area of sensorimotor cortex (SMC) and supplementary motor area (SMA) before and after the intervention between groups (PC mode and AI mode). The ERSP is a function of frequency over the time-course of the event-related potentials, which was computed the power spectrum over a sliding latency window then averaging across data trials using the Matlab using EEGLAB. As illustrated in Fig. 3, EEG data analysis was performed in Matlab using EEGLAB (Delorme, and Makeig 2004). First, the EEG data set (sampling rate = 500 Hz) was retrieved, and the EMG channel was removed. The 24 channels would be left, and the data was band pass filtered (1– 60 Hz). Using the kurtosis method, we removed the bad channels at more than five standard deviations from the mean. The remaining EEG channels were then re-referenced to their common average. Next, artifact subspace reconstruction (ASR) was applied to remove high amplitude artifacts (e.g., eye blinks, muscle burst). ASR applied principal component analysis to EEG data in sliding windows and identifies channels that significantly deviate from the baseline data containing minimal movement artifact (Luu et al., 2017). In other words, it is an algorithm for data that contains many moving artifacts to reduce the moving artifacts and recover data or to remove channels with overly large moving artifacts. Next, independent component analysis (ICA) was used to separate the channels independently. Then, we remove automatically bad independent components (ICs) using Multiple Artifact Rejection Algorithm (MARA). We applied the moving average filter and proceeded with the segmentation considering one gait cycle an epoch. Finally, we performed the ERD/ERS pattern analysis.

Fig. 2

Flowchart of the study.

Fig. 3

Flowchart for EEG data analysis.

2.8 EMG analysis

EMG data analysis was performed in Matlab. First, we removed the offset of the EMG data set (sampling rate = 500 Hz). Then, EMG data were band pass filtered (20– 240 Hz). We eliminated direct current offset, motion-artifacts, and high frequency noise. Next EMG data were band-stop filtered (60 Hz) to eliminate the electric frequency. The EMG signals were then smoothed using 50 points moving average filter. The smoothed EMG signals were extracted for each gait cycle to obtain average gait cycles. We then computed the moving root mean square (RMS) envelopes of the EMG signals.

2.9 Statistical analysis

Statistical analyses were used the descriptive statistics for EEG and EMG due to the inadequate sample size. A power analysis using G-Power software (Franz Faul, Kiel, Germany) was conducted to assess the sample size requirement (N = 20) based on our previous study, which demonstrated effect size (eta squared, η² = 0.6) and power (1-β= 0.8). The inferential statistics was only executed for homogeneity of group and comparing the score of clinical function tests. The Chi-square test and Mann-Whitney U test were used for homogeneity between the PC-RAGT and AI-RAGT groups. The clinical function tests were analyzed using the Mann– Whitney U test for between-group pre-intervention and post-intervention comparisons and the Wilcoxon signed-rank test for within-group comparisons for each intervention group. SPSS version 21.0 software (SPSS software, SPSS Inc., Chicago, USA) was used for the analysis at p < 0.05.

3 Results

3.1 Patients characteristics

Of the 20 stroke patients enrolled, 14 patients were included in the analysis. The flow through the study and the description of dropouts are presented in Fig. 2. The demographic characteristics of 14 subjects are presented in Table 1. No significant differences were observed in baseline characteristics, including age, sex, height, weight, onset period of stroke, and affected side of hemiparesis between the two groups, satisfying the assumption of homogeneity.

Table 1
Baseline demographic characteristics (N = 14)

PC^a-RAGT^b group AI^c-RAGT group p

Age (years) 51.38±11.01^d 51.75±15.33 0.956

Sex (male/female) 5/2 6/1 0.554

Body height (cm) 165.46±12.02 167.25±13.87 0.876

Body weight (kg) 67.85±8.52 68.18±7.34 0.315

Months from stroke onset 28.38±23.21 24.75±27.49 0.780

Affected side of hemiparesis 0.554

Left 5 6

Right 2 1

	PC^a-RAGT^b group	AI^c-RAGT group	p
Age (years)	51.38±11.01^d	51.75±15.33	0.956
Sex (male/female)	5/2	6/1	0.554
Body height (cm)	165.46±12.02	167.25±13.87	0.876
Body weight (kg)	67.85±8.52	68.18±7.34	0.315
Months from stroke onset	28.38±23.21	24.75±27.49	0.780
Affected side of hemiparesis			0.554
Left	5	6
Right	2	1

^aPC: Passive-controlled. ^bRAGT: Robot-assisted gait training. ^cAI: Active-interactive. ^dMean±Standard deviation.

3.2 Clinical functional tests

Wilcoxon signed-rank tests demonstrated significant within-group improvements in FAC, MAS, FMA-LE for both groups (p < 0.05), but the between-group changes were not observed (Table 2).

Table 2
Results of clinical functional tests for PC- and AI-RAGT groups (N = 14)

Outcome measures Group Pre Post p (within-group)

FAC^a PC^d-RAGT^e group 3.5^g 4 0.046

AI^f-RAGT group 3 4 0.034

p (Between-group) 0.379 0.609

MAS^b PC-RAGT group 2 2 0.024

AI-RAGT group 2 1 0.038

p (Between-group) 0.081 0.185

FMA-LE^c PC-RAGT group 17.5 23.5 0.012

AI-RAGT group 19.5 26 0.011

p (Between-group) 1.000 0.874

Outcome measures	Group	Pre	Post	p (within-group)
FAC^a	PC^d-RAGT^e group	3.5^g	4	0.046
	AI^f-RAGT group	3	4	0.034
	p (Between-group)	0.379	0.609
MAS^b	PC-RAGT group	2	2	0.024
	AI-RAGT group	2	1	0.038
	p (Between-group)	0.081	0.185
FMA-LE^c	PC-RAGT group	17.5	23.5	0.012
	AI-RAGT group	19.5	26	0.011
	p (Between-group)	1.000	0.874

^aFAC: Functional ambulation category. ^bMAS: Modified Ashworth scale. ^cFMA: Fugl– Meyer Assessment for lower extremity. ^dPC: Passive-controlled. ^eRAGT: Robot-assisted gait training. ^fAI: Active-interactive. ^gMedian value.

3.3 Gait Event-Related Spectral Perturbations (Gait ERSPs)

Descriptive analysis revealed increased μ power in SMC area, including CP1 and CP2, showed after intervention in subjects PC-1,4,5, and 6 (Table 3). In subjects PC-3 and 5, the μ power predominantly increased after intervention in overall SMC area and SMA. In subject PC-1, the μ power increased after intervention in FC2 of SMA, although the power in FC1 and Fz decreased. The increased μ power showed after intervention in SMA, excluding FC2, in subject PC-6. In subject PC-4, the μ power decreased after intervention in SMA area, although the power in SMC increased. The μ power in subject PC-2 decreased throughout SMC area and SMA. The pretest data of subject PC-3 in CP2 was lost during pre-analysis due to contamination of the data.

Table 3
Results of gait ERSPs (dB) for PC-RAGT group (N = 7)

Subject no. SMC^a SMA^b

CP1 CP2 FC1 FC2 Fz

Pre Post Pre Post Pre Post Pre Post Pre Post

PC^c-1 1.559^c 3.221 1.828 2.653 0.939 2.108 1.487 1.353 1.529 2.039

PC-2 0.672 0.234 0.470 0.126 0.607 0.453 0.796 0.340 1.181 0.466

PC-3 1.233 35.285 36.122 0.133 37.606 0.315 37.562 0.341 38.312

PC-4 0.645 1.023 1.085 1.474 1.440 0.707 0.760 0.675 1.177 0.694

PC-5 5.144 35.380 5.498 34.342 5.568 34.496 5.183 34.936 5.289 37.261

PC-6 3.012 6.048 8.090 10.946 2.783 4.415 7.014 5.487 3.339 3.404

PC-7 0.705 1.019 0.861 0.579 1.319 2.021 0.758 0.467 1.282 0.949

Subject no.	SMC^a	SMA^b
	Pre	Post	Pre	Post	Pre	Post	Pre	Post	Pre	Post
PC^c-1	1.559^c	3.221	1.828	2.653	0.939	2.108	1.487	1.353	1.529	2.039
PC-2	0.672	0.234	0.470	0.126	0.607	0.453	0.796	0.340	1.181	0.466
PC-3	1.233	35.285		36.122	0.133	37.606	0.315	37.562	0.341	38.312
PC-4	0.645	1.023	1.085	1.474	1.440	0.707	0.760	0.675	1.177	0.694
PC-5	5.144	35.380	5.498	34.342	5.568	34.496	5.183	34.936	5.289	37.261
PC-6	3.012	6.048	8.090	10.946	2.783	4.415	7.014	5.487	3.339	3.404
PC-7	0.705	1.019	0.861	0.579	1.319	2.021	0.758	0.467	1.282	0.949

^aSMC: Somatosensory motor cortex. ^bSMA: Supplementary motor area. ^cPC: Passive-controlled.

The increased μ power in both SMC area and SMA showed after intervention in subjects AI-1, 6, 7 (Table 4). In subject AI-2, the μ power predominantly increased after intervention in overall SMC area and SMA. However, the μ power in subjects AI-3, 4 decreased throughout SMC area and SMA after intervention. In subject AI-5, the μ power increased after intervention in CP1, FC2, although the power in CP2, FC1, Fz decreased. The posttest data of subject AI-2 in CP2 and pretest data of subject AI-4 in FC2 were lost during pre-analysis due to contamination of the data.

Table 4

Results of gait ERSPs (dB) for AI-RAGT group (N = 7)

Subject no.	SMC^a				SMA^b
	CP1		CP2		FC1		FC2		Fz
	Pre	Post	Pre	Post	Pre	Post	Pre	Post	Pre	Post
AI^c-1	1.890^c	7.817	1.218	7.198	1.401	8.049	0.847	7.042	1.444	9.254
AI-2	3.200	43.361	4.240		2.912	46.515	2.276	47.021	2.703	48.967
AI-3	56.820	17.998	47.254	12.014	37.946	3.162	33.408	3.179	32.243	4.143
AI-4	1.524	0.207	2.345	0.505	1.835	0.254		0.588	2.867	0.230
AI-5	2.549	2.559	9.577	3.603	1.623	1.024	2.388	2.834	2.189	1.572
AI-6	1.728	3.512	3.940	4.242	1.368	4.358	1.340	2.785	1.538	3.621
AI-7	2.653	5.625	1.753	5.787	1.820	5.971	2.739	5.461	2.462	4.678

^aSMC: Somatosensory motor cortex. ^bSMA: Supplementary motor area. ^cAI: Active-interactive.

3.4 Muscle activation of Tibialis Anterior (TA)

The mean RMS EMG amplitude values of affected and unaffected TA muscles during swing phase were presented in Tables 5 and 6. TA muscle activation in subject PC-1 increased in both affected and unaffected side in the PC-RAGT group. The increased activation of TA showed in unaffected side of subject PC-2 and 3, although the activation in affected side decreased after the PC RAGT intervention. In contrast, the increased activation of TA showed in affected side of subject PC-4, 5 and 7, whereas the activation in unaffected side decreased after the intervention. The decreased muscle activation of TA in both affected and unaffected side showed after the PC-RAGT intervention in subject PC-6.

Table 5
Results of muscle activation (%) of TA during swing phase for PC-RAGT group (N = 7)

Subject no. Affected side Unaffected side

Pre Post Pre Post

PC^a -1 11.53 ±6.52^b 12.96±5.47 10.58±3.22 12.29±4.16

PC-2 19.8±6.84 5.82±3.32 13.51±4.55 20.32±9.17

PC-3 28.87±5.80 5.41±1.92 10.43±2.00 40.25±16.72

PC-4 7.21±2.48 15.15±4.93 24.16±8.98 11.42±3.78

PC-5 5.39±1.53 16.79±25.43 7.34±4.04 6.59±2.68

PC-6 26.48±6.12 9.2±4.28 18.33±11.65 11.84±3.43

PC-7 13.46±6.33 19.58±18.96 9.87±3.53 3.07±1.32

Subject no.	Affected side	Unaffected side
PC^a -1	11.53 ±6.52^b	12.96±5.47	10.58±3.22	12.29±4.16
PC-2	19.8±6.84	5.82±3.32	13.51±4.55	20.32±9.17
PC-3	28.87±5.80	5.41±1.92	10.43±2.00	40.25±16.72
PC-4	7.21±2.48	15.15±4.93	24.16±8.98	11.42±3.78
PC-5	5.39±1.53	16.79±25.43	7.34±4.04	6.59±2.68
PC-6	26.48±6.12	9.2±4.28	18.33±11.65	11.84±3.43
PC-7	13.46±6.33	19.58±18.96	9.87±3.53	3.07±1.32

^aPC: Passive-controlled. ^bMean±Standard deviation.

Table 6

Results of muscle activation (%) of TA during swing phase for AI-RAGT group (N = 7)

Subject no.	Affected side		Unaffected side
	Pre	Post	Pre	Post
AI^a-1	11.83±5.57^b	10.86±4.58	0.81±0.42	15.85±8.22
AI-2	11.78±7.46	24.25±17.30	4.72±1.57	14.86±7.79
AI-3	0.76±0.39	4.84±2.47	10.7±3.94	18.18±13.51
AI-4	2.68±0.60	10.2±1.93	19.32±9.44	7.15±3.93
AI-5	9.38±3.52	5.3±1.9	16.64±12.02	13.54±7.21
AI-6	29.51±10.01	22.68±9.6	17.21±9.19	31.03±13.53
AI-7	24.05±22.80	8.75±5.66	19.91±7.99	24.48±5.19

^aAI: Active-interactive. ^bMean±Standard deviation.

The muscle activation of TA in subjects AI-2 and 3 increased in both affected and unaffected side in the AI-RAGT group. The increased activation of TA showed in affected side of subject AI-4, although the activation in unaffected side decreased after the AI RAGT intervention. The increased activation of TA showed in unaffected side of subject AI-1, 5, 6, 7, although the activation in affected side decreased after the AI-RAGT intervention.

4 Discussion

The present clinical trial is a first comparative investigation highlighting important cortico-neuromuscular changes and associated functional gait recovery between the PC-RAGT and AI-RAGT modes in subacute to chronic hemiparetic stroke patients. As hypothesized, our preliminary results demonstrated that the power in the μ band was increased in the leg area of SMC (CP1 and CP2) and SMA (FC1 and FC2) at post-intervention as compared to pre-intervention. Such cortical neuroplasticity change was associated with TA muscle activity during gait and functional independence in ambulation in FAC and motor recovery and coordination FMA-LE test as well as spasticity in the MAS measures.

ERSP analysis underpinning cortical neuronal activity during RAGT seemingly revealed greater tendency of the increase of power in the μ band, indicating increased neuroplasticity in AI-RAGT group than in the PC-RAGT group. These novel findings further corroborate with Wagner and colleagues (2012) who reported more desynchronized μ (8– 12 Hz) and β (18– 21 Hz) rhythms in the central SMC area during phases of active robotic walking than passive robotic walking in 14 healthy participants, supporting the notion of cortical suppression during active robotic training mode (Wagner et al., 2012). Similarly, other EEG studies showed a significant suppression and correlated desynchronization in the μ band over the SMC during precision stepping (Presacco et al., 2011), robotic assisted stepping (Wieser et al., 2010), and walking with robotic gait orthosis (Seeber et al., 2014) in healthy subjects. fNIRS showed more increased cortical hemodynamic activation in the SMC and SMA during robot-assisted walking than during conventional stepping walking and treadmill walking condition in 14 healthy adults (Kim et al., 2016). Nevertheless, no previous clinical study has conducted to ascertain the robotic training induced neuroplastic changes and associated clinical locomotor recovery in stroke population, which made difficult to compare with previous evidence in the current literature.

Possible neuroplastic mechanisms supporting active RAGT may arise from synaptogenesis and unmasking. Contemporary neuroimaging studies demonstrated increased corticospinal excitability in the ipsilesional SMC and corticospinal tract following intensive stroke intervention, respectively indicating the theoretic assumption of ‘synaptogenetic’ neuroplasticity (Marshall et al., 2000; Rehme et al., 2011). Alternatively, other neuroimaging studies with fMRI, diffusion tensor imaging (DTI) and transcranial magnetic stimulation (TMS) showed therapy-induced cortical reorganization in the contralesional SMC and anterior corticospinal pathway, which further substantiates the notion of the ‘unmasking’ neuroplasticity (Hoyer & Celnik, 2011; Jang et al., 2005; Sun et al., 2013). Preclinical evidence demonstrated that synaptogenesis increases substantially, and that dendrite number and shape are altered following after experimental lesioning in brain of adult rat (Jones et al., 1996). Furthermore, both synaptogenesis and dendrite remodeling are associated with increases in neurological activity in cerebral cortical map (Brown et al., 2009; Jones et al., 1996). According to previous research involving cortical map rearrangements, injury to the motor cortex leads to the recruitment of motor areas that were not making significant contribution to the lost function before the injury (Pekna et al., 2012). The notion that the activity of cortical areas recruited after injury plays a role in functional recovery in humans is supported by a study showing that in well-recovered stroke patients, the ipsilesional dorsal motor cortex shows increase in activity (Gerloff et al., 2006).

Interestingly, some of subjects showed the decrease in μ band power value and muscle activation of TA at post-intervention compared to pre-intervention in both groups. Perhaps, regardless of the RAGT modes, the intensive, repetitive RAGT training overtime shifted cortically driven locomotor control to subcortical or spinal networks which regulate automatic locomotor movement with little or no conscious attention (Petersen et al., 2012). Luft and colleagues (2008) implemented a randomized controlled study to examine a long-term effect of a 6-month progressive task-repetitive treadmill exercise (T-EX) on changes in neural substrates using fMRI (Luft et al., 2008). Significant increases in the subcortical activation in the contralesional cerebellum and associated gait recovery function were evident in 12 subjects with chronic hemiparetic stroke (Luft et al., 2008). It is possible that sensorimotor stimuli via accurate, repetitive RAGT can help reorganize cortical or subcortical and spinal locomotor networks, which subsequently facilitates the recovery of functional ambulation (Macko et al., 2005). However, further research is warranted to unveil such brain plasticity underlying locomotor recovery after stroke.

Taken together, we have first developed a novel, advanced neuroimaging experimental paradigm which opened a new horizon and window of probing the RAGT induced neuroplastic changes in hemiparetic stroke. The present EEG experimental method distinguished gait event related cortical involvement and differentiated the positive, promising changes between pre- and post-intervention with PC- and AI-RAGT in individuals with subacute and chronic hemiparetic stroke.

A couple of study limitations should be considered in future research. One limitation is that sample size was relatively small due to patient attrition and data missing. A careful interpretation should be hence exercised when generalizing the current findings. Another limitation is that the important role of the subcortical and spinal control networks modulating the locomotor behavior and recovery mechanism were not measured in the present investigation during the RAGT. Therefore, it is difficult to decipher the exact neural substrates and potential neurophysiological mechanisms accounting for neuroplasticity, which invites for new development of dynamic neuroimaging during gait. Lastly, it would be of great interest to assess the cortico-neuromuscular coherence which is cardinal information to understand cortico-neuromuscular recovery processes after stroke underlying locomotor relearning and neuroplasticity in neurorehabilitation. Further research is needed to clarify the specific contributions of neuromuscular mechanisms and neuroplasticity of active RAGT for stroke patients.

5 Conclusion

We have first developed an advanced neuroimaging experimental paradigm which distinguished gait event related cortical involvement and differentiated the promising changes between the PC-RAGT and AI-RAGT modes in individuals with subacute and chronic hemiparetic stroke. Our results warrant future studies on specific contributions of cortico-muscular mechanisms and neuroplasticity of active RAGT in larger stroke patient populations.

Footnotes

Acknowledgments

This research received financial and administrative support from the This research received financial and administrative support from the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (No.2019014441) and Brain Korea 21 PLUS Project (grant no. 2019-51-0018) for the Department of Physical Therapy in Graduate School, Yonsei University. This manuscript has been submitted solely to this journal and has not been published or submitted elsewhere.

Conflict of interest

None to report.

References

Brown,

C. E.

, Aminoltejari,

, Erb,

, Winship,

I. R.

, & Murphy,

T. H.

(2009) In vivo voltage-sensitive dye imaging in adult mice reveals that somatosensory maps lost to stroke are replaced over weeks by new structural and functional circuits with prolonged modes of activation within both the peri-infarct zone and distant sites, Journal of Neuroscience 29(6), 1719–1734.

Bulea,

T. C.

, Kim,

, Damiano,

D. L.

, Stanley,

C. J.

, & Park,

H.-S.

(2015) Prefrontal, posterior parietal and sensorimotor network activity underlying speed control during walking, Frontiers in Human Neuroscience 9, 247.

Cao,

, Xie,

S. Q.

, Das,

, & Zhu,

G. L.

(2014) Control strategies for effective robot assisted gait rehabilitation: the state of art and future prospects, Med Eng Phys 36(12), 1555–1566. https://doi.org/10.1016/j.medengphy.2014.08.005

Criswell,

(2010) Cram’s introduction to surface electromyography. Jones & Bartlett Publishers.

Delorme,

, & Makeig,

(2004) EEGLAB: an open source toolbox for analysis of single-trial EEG dynamics including independent component analysis, Journal of Neuroscience Methods 134(1), 9–21.

Dohring,

M. E.

, & Daly,

J. J.

(2008) Automatic synchronization of functional electrical stimulation and robotic assisted treadmill training, IEEE Transactions on Neural Systems and Rehabilitation Engineering 16(3), 310–313.

Duschau-Wicke,

, Caprez,

, & Riener,

(2010) Patient-cooperative control increases active participation of individuals with SCI during robot-aided gait training, Journal of Neuroengineering and Rehabilitation 7(1), 1–13.

Fleerkotte,

B. M.

, Koopman,

, Buurke,

J. H.

, van Asseldonk,

E. H.

, van der Kooij,

, & Rietman,

J. S.

(2014) The effect of impedance-controlled robotic gait training on walking ability and quality in individuals with chronic incomplete spinal cord injury: an explorative study, Journal of Neuroengineering and Rehabilitation 11, 26. https://doi.org/10.1186/1743-0003-11-26

Gerloff,

, Bushara,

, Sailer,

, Wassermann,

E. M.

, Chen,

, Matsuoka,

, Waldvogel,

, Wittenberg,

G. F.

, Ishii,

, & Cohen,

L. G.

(2006) Multimodal imaging of brain reorganization in motor areas of the contralesional hemisphere of well recovered patients after capsular stroke, Brain 129(3), 791–808.

10.

Gwin,

J. T.

, Gramann,

, Makeig,

, & Ferris,

D. P.

(2011) Electrocortical activity is coupled to gait cycle phase during treadmill walking, Neuroimage 54(2), 1289–1296.

11.

Hermens,

H. J.

, Freriks,

, Merletti,

, Stegeman,

, Blok,

, Rau,

, Disselhorst-Klug,

, & Hägg,

(1999) European recommendations for surface electromyography, Roessingh Research and Development 8(2), 13–54.

12.

Hidler,

J. M.

, & Wall,

A. E.

(2005) Alterations in muscle activation patterns during robotic-assisted walking, Clinical Biomechanics 20(2), 184–193.

13.

Hoyer,

E. H.

, & Celnik,

P. A.

(2011) Understanding and enhancing motor recovery after stroke using transcranial magnetic stimulation, Restorative Neurology and Neuroscience 29(6), 395–409.

14.

Hussain,

, Jamwal,

P. K.

, Ghayesh,

M. H.

, & Xie,

S. Q.

(2016) Assist-as-needed control of an intrinsically compliant robotic gait training orthosis, IEEE Transactions on Industrial Electronics 64(2), 1675–1685.

15.

Hussain,

, Xie,

S. Q.

, & Liu,

(2011) Robot assisted treadmill training: mechanisms and training strategies, Medical Engineering and Physics 33(5), 527–533.

16.

Jacquelin Perry,

(2010) Gait analysis: normal and pathological function. New Jersey: SLACK.

17.

Jang,

S. H.

, Cho,

S.-H.

, Kim,

Y.-H.

, You,

S. H.

, Kim,

S. H.

, Kim,

, & Yang,

D. S.

(2005) Motor recovery mechanism of diffuse axonal injury: a combined study of transcranial magnetic stimulation and functional MRI, Restorative Neurology and Neuroscience 23(1), 51–56.

18.

Johnson,

A. N.

, & Shinohara,

(2012) Corticomuscular coherence with and without additional task in the elderly, Journal of Applied Physiology 112(6), 70–981.

19.

Jones,

T. A.

, Kleim,

J. A.

, & Greenough,

W. T.

(1996) Synaptogenesis and dendritic growth in the cortex opposite unilateral sensorimotor cortex damage in adult rats: a quantitative electron microscopic examination, Brain Research 733(1), 142–148.

20.

Kilner,

, Baker,

, Salenius,

, Jousmäki,

, Hari,

, & Lemon,

(1999) Task-dependent modulation of 15-30 Hz coherence between rectified EMGs from human hand and forearm muscles, The Journal of Physiology 516(2), 559–570.

21.

Kim,

H. Y.

, Yang,

S. P.

, Park,

G. L.

, Kim,

E. J.

, & You,

J. S. H.

(2016) Best facilitated cortical activation during different stepping, treadmill, and robot-assisted walking training paradigms and speeds: a functional near-infrared spectroscopy neuroimaging study, NeuroRehabilitation 38(2), 171–178.

22.

Klem,

G. H.

, Lüders,

H. O.

, Jasper,

H. H.

, & Elger,

(1999) The ten-twenty electrode system of the International Federation. The International Federation of Clinical Neurophysiology, Electroencephalography and Clinical Neurophysiology Supplement 52, 3–6.

23.

Knaepen,

, Mierau,

, Swinnen,

, Fernandez Tellez,

, Michielsen,

, Kerckhofs,

, Lefeber,

, & Meeusen,

(2015) Human-robot interaction: does robotic guidance force affect gait-related brain dynamics during robot-assisted treadmill walking? PloS One 10(10), e0140626.

24.

Luft,

A. R.

, Macko,

R. F.

, Forrester,

L. W.

, Villagra,

, Ivey,

, Sorkin,

J. D.

, Whitall,

, McCombe-Waller,

, Katzel,

, & Goldberg,

A. P.

(2008) Treadmill exercise activates subcortical neural networks and improves walking after stroke: a randomized controlled trial, Stroke 39(12), 3341–3350.

25.

Luu,

T. P.

, Brantley,

J. A.

, Zhu,

, & Contreras-Vidal,

J. L.

(2017) Electrocortical amplitudemodulations of human level-ground, slope, and stair walking. 2017 39th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC),

26.

Macko,

R. F.

, Ivey,

F. M.

, Forrester,

L. W.

, Hanley,

, Sorkin,

J. D.

, Katzel,

L. I.

, Silver,

K. H.

, & Goldberg,

A. P.

(2005) Treadmill exercise rehabilitation improves ambulatory function and cardiovascular fitness in patients with chronic stroke: a randomized, controlled trial, Stroke 36(10), 2206–2211.

27.

Makeig,

(1993) Auditory event-related dynamics of the EEG spectrum and effects of exposure to tones, Electroencephalography and Clinical Neurophysiology 86(4), 283–293.

28.

Marchal-Crespo,

, & Reinkensmeyer,

D. J.

(2009) Review of control strategies for robotic movement training after neurologic injury, Journal of Neuroengineering and Rehabilitation 6(1), 20. https://doi.org/10.1186/1743-0003-6-20

29.

Marshall,

R. S.

, Perera,

G. M.

, Lazar,

R. M.

, Krakauer,

J. W.

, Constantine,

R. C.

, & DeLaPaz,

R. L.

(2000) Evolution of cortical activation during recovery from corticospinal tract infarction, Stroke 31(3), 656–661.

30.

Miyai,

, Tanabe,

H. C.

, Sase,

, Eda,

, Oda,

, Konishi,

, Tsunazawa,

, Suzuki,

, Yanagida,

, & Kubota,

(2001) Cortical mapping of gait in humans: a near-infrared spectroscopic topography study, Neuroimage 14(5), 1186–1192.

31.

Moreno,

J. C.

, Barroso,

, Farina,

, Gizzi,

, Santos,

, Molinari,

, & Pons,

J. L.

(2013) Effects of robotic guidance on the coordination of locomotion, Journal of Neuroengineering and Rehabilitation 10(1), 1–15.

32.

Mummolo,

, Mangialardi,

, & Kim,

J. H.

(2013) Quantifying dynamic characteristics of human walking for comprehensive gait cycle, Journal of Biomechanical Engineering 135(9).

33.

Park,

, Oh-Park,

, Bialek,

, Friel,

, Edwards,

, & You,

J. S. H.

(2021) Abnormal synergistic gait mitigation in acute stroke using an innovative ankle– knee– hip interlimb humanoid robot: A preliminary randomized controlled trial, Scientific Reports 11(1), 1–13.

34.

Park,

, Oh-Park,

, Dohle,

, Bialek,

, Friel,

, Edwards,

, Krebs,

H. I.

, & You,

J. S. H.

(2020) Effects of innovative hip-knee-ankle interlimb coordinated robot training on ambulation, cardiopulmonary function, depression, and fall confidence in acute hemiplegia, NeuroRehabilitation 46(4), 577–587.

35.

Park,

J.H.

, Shin,

Y.I.

, You,

J. S. H.

, & Park,

M. S.

(2018) Comparative effects of robotic-assisted gait training combined with conventional physical therapy on paretic hip joint stiffness and kinematics between subacute and chronic hemiparetic stroke, NeuroRehabilitation 42(2), 181–190.

36.

Pekna,

, Pekny,

, & Nilsson,

(2012) Modulation of neural plasticity as a basis for stroke rehabilitation, Stroke 43(10), 2819–2828.

37.

Petersen,

T. H.

, Willerslev-Olsen,

, Conway,

B. A.

, & Nielsen,

J. B.

(2012) The motor cortex drives the muscles during walking in human subjects, The Journal of Physiology 590(10), 2443–2452.

38.

Pfurtscheller,

, & Da Silva,

F. L.

(1999) Event-related EEG/MEG synchronization and desynchronization: basic principles, Clinical Neurophysiology 110(11), 1842–1857.

39.

Presacco,

, Goodman,

, Forrester,

, & Contreras-Vidal,

J. L.

(2011) Neural decoding of treadmill walking from noninvasive electroencephalographic signals, Journal of Neurophysiology 106(4), 1875–1887.

40.

Rehme,

A. K.

, Eickhoff,

S. B.

, Wang,

L. E.

, Fink,

G. R.

, & Grefkes,

(2011) Dynamic causal modeling of cortical activity from the acute to the chronic stage after stroke, Neuroimage 55(3), 1147–1158.

41.

Seeber,

, Scherer,

, Wagner,

, Solis-Escalante,

, & Müller-Putz,

G. R.

(2014) EEG beta suppression and low gamma modulation are different elements of human upright walking, Frontiers in Human Neuroscience 8, 485.

42.

Severens,

, Nienhuis,

, Desain,

, & Duysens,

(2012) Feasibility of measuring event related desynchronization with electroencephalography during walking. 2012 Annual International Conference of the IEEE Engineering in Medicine and Biology Society,

43.

Sun,

, Yin,

, Zhu,

, Fan,

, Zang,

, Wu,

, Jia,

, Bai,

, Zhu,

, & Hu,

(2013) Cortical reorganization after motor imagery training in chronic stroke patients with severe motor impairment: a longitudinal fMRI study, Neuroradiology 55(7), 913–925.

44.

van Kammen,

, Boonstra,

A. M.

, van der Woude,

L. H.

, Reinders-Messelink,

H. A.

, & den Otter,

(2016) The combined effects of guidance force, bodyweight support and gait speed on muscle activity during able-bodied walking in the Lokomat, Clinical Biomechanics 36, 65–73.

45.

Van Tran,

, Kim,

, Lee,

, Kang,

, & Ryu,

(2015) Force/torque sensorless impedance control for indirect driven robot-aided gait rehabilitation system. 2015 IEEEInternational Conference on Advanced Intelligent Mechatronics (AIM),

46.

Veneman,

J. F.

, Kruidhof,

, Hekman,

E. E.

, Ekkelenkamp,

, Van Asseldonk,

E. H.

, & Van Der Kooij,

(2007) Design and evaluation of the LOPES exoskeleton robot for interactive gait rehabilitation, IEEE Transactions on Neural Systems and Rehabilitation Engineering 15(3), 379–386.

47.

Wagner,

, Solis-Escalante,

, Grieshofer,

, Neuper,

, Müller-Putz,

, & Scherer,

(2012) Level of participation in robotic-assisted treadmill walking modulates midline sensorimotor EEG rhythms in able-bodied subjects, Neuroimage 63(3), 1203–1211.

48.

Wieser,

, Haefeli,

, Bütler,

, Jäncke,

, Riener,

, & Koeneke,

(2010) Temporal and spatial patterns of cortical activation during assisted lower limb movement, Experimental Brain Research 203(1), 181–191.

49.

Wolbrecht,

E. T.

, Chan,

, Le,

, Cramer,

S. C.

, Reinkensmeyer,

D. J.

, & Bobrow,

J. E.

(2007) Real-time computer modeling of weakness following stroke optimizes robotic assistance for movement therapy. 2007 3rd International IEEE/EMBS Conference on Neural Engineering