TABLE 1.
Overview of studies investigating the neurophysiological mechanisms underlying RAGT using EEG and fNIRS.
| Author(s) | Title | Robotic system | Method | EEG/fNIRS device | Parameter | Sample | Outcome | Conclusion |
| Calabrò et al.,2018 | Shaping neuroplasticity by using powered exoskeletons in patients with stroke: a randomized clinical trial | EksoTM | Prospective, pre-post, randomized clinical trial; 1st group: EGT and OGT; 2nd group: OGT; Study design: 45 min per session, five times a week over 8 weeks | EEG Brain Quick SystemPLUS (Micro-med; Mogliano Veneto, Italy); 21 electrodes Sampling rate: 512 Hz; Electrode placement: 10–20 system; Additional: sEMG & TMS | EEG: FPEC EMG: gait performance based on the 10 m walking test, gait cycle, muscle activation muscles; TMS: CSE and SMI from both M1 | Stroke patients; Hemiparesis caused by stroke in a chronic phase; Number of participants: 40 patients, 20 each in a group; Age: ≥55 years | FPEC: significant strengthening of EGT group; CSE: significant improvement in EGT group on the affected side; SMI: significant improvement in EGT group on affected side; 10 m walk test: significant improvement in EGT group; General walking quality: significant improvement; Hip and knee muscle activation: significant improvement | EGT seems, in addition to OGT, also promising in gait rehabilitation for patients after a stroke. The study suggests that Ekso could be useful to promote the mobility of people with stroke, thanks to the mechanisms of brain plasticity and remodulation of connectivity that are specifically carried along by the robotic system compared to conventional OGT. |
| Contreras-Vidal et al.,2018 | Neural Decoding of Robot-Assisted Gait during Rehabilitation after Stroke | H2 Exoskelett | Study design: 12 training sessions, 4 weeks Meanwhile, no additional therapy; Walking speed: preset and comfortable for the test persons, could be adjusted while walking | EEG BrainAmpDC, Brain Products, Germany; 64 electrodes Sampling rate: 1000 Hz | Changes in power spectra for the 0.1–3 Hz (delta) band; Additional: walking speed, walking distance | Stroke patients Chronic hemiparesis after a stroke Number of test persons: 5; Exclusion: 1 | Suppression of the delta frequency when walking in the occipital scalp area. Increase in delta frequency suppression in the frontal and central scalp regions. Improvement of walking distance and walking speed, which correlated with increased accuracy of offline decoding. | Proof of the feasibility of neuronal decoding of gait kinematics from the EEG during RAGT in chronic stroke patients. Since motor intention recognition from EEG signals is synchronized with motion feedback generated by exoskeleton-assisted movements of the lower extremities, the BMI-H2 system can promote brain reorganization through motor learning, presumably due to activity-dependent brain plasticity. First step in the development of a brain-machine interface to control driven exoskeletons. |
| Knaepen et al.,2015 | Human-Robot Interaction: Does Robotic Guidance Force Affect Gait-Related Brain Dynamics during Robot-Assisted Treadmill Walking? | Lokomat | Stratified randomization 4 gait conditions of 5 min each: 1st condition at the beginning without Lokomat. Condition 2–4 with Lokomat with 3 levels of managers. Speed: 2 km/h; Conditions: Unassisted treadmill walking as well as during robot-assisted treadmill walking GF: 30, 60, and 100% BWS: 0 | EEG BrainAmp DC & BrainVision Recorder, Brain Products GmbH, Germany; 32 electrodes Sampling rate: 1000 Hz Band pass: 0.5–100 Hz; Electrode placement: 10–20 system | Examination of ERSPs and PSDs during RATW at 30, 60, and 100% GF | No health impairments; Number of test persons: 11; Exclusion: 7; Male: 3; Female: 9; Age: 28.2 ± 4.0 years; Weight: 64,7 ± 7,7 Kg | Gait-related spectral modulations in the mu-, beta- and lower gamma bands above the SMC, related to certain phases of the gait cycle. Mu and beta rhythms were suppressed in the right primary sensory cortex during treadmill walking compared to 100% robot-assisted treadmill training, indicating significantly greater involvement of the sensorimotor area during treadmill walking compared to robot-assisted treadmill walking. Minor differences in the spectral performance of mu, beta and lower gamma bands between robotic treadmill walking with different guidance strengths. | High leadership strength and thus less active participation in the movement should be avoided during robot-supported treadmill training. This will optimize the participation of the sensorimotor cortex, which is known to be essential for motor learning. |
| Lapitskaya et al.,2011 | Robotic gait training in patients with impaired consciousness due to severe traumatic brain injury | Lokomat | Prospective, controlled, non-randomized study. Single training session: Speed: 1.5 ± 0.1 km/h (patients), 1.61 ± 0.08 km/h (healthy subjects); Training time: 17.1 ± 1.3 min (patients), 17.15 ± 0.11 min (healthy volunteers); Walking distance: 427.3 ± 38.6m (patients), 451.36 ± 15.02m (healthy volunteers); conditions: Recovery phases and RAGT; GF: 100%; EEG recording in sitting position | EEG Nervous system (Taugagreining hf, Reykjavik, Island) 19 electrodes placement: Fz, Cz and Pz; ECI Electro-Cap SystemTM, International, Inc., Eaton, OH, United States; SEP measurement: ’VikingQuest’ (Viasys Healthcare, San Diego, CA, United States) | Sensory nerve tracts were evaluated with the help of sensory ERPs. Global DAR and the latency of the P300 component of the event-related potentials before and after a training session. | Patients with TBI disturbances of consciousness; Sample size: 12; Male: 9; Female: 3; Age: 40.8 ± 18.2 years; Control group: 14 healthy male subjects; Age: 47.3 ± 14.5 years | Basic measurement: Impaired SEPs in most patients and a significantly larger DAR in patients compared to healthy ones; After RAGT: Reduction of DAR in healthy subjects, but not in patients. No changes in P300 latency after training in patients or healthy subjects. | The study showed that robotic gait training induces measurable changes in the EEG performance spectrum in healthy individuals, while no changes were observed in patients with severe TBI. The absence of changes in the EEG power spectrum after RAGT in the patient may be an indicator of the severity of the injury. |
| Nakanishi et al.,2014 | Rapid changes in arousal states of healthy volunteers during robot-assisted gait training: a quantitative time-series electroencephalography study | GAR | Conditions: Standing versus passive RAGT Standing: 30 s with eyes closed and 30 s with eyes open; RAGT: 6 min. at 3 conditions: (1) sinus wave noise stimulation, (2) verbal noise stimulation, (3) no noise stimulation; Speed: 0.11 m/s | EEG Polymate II AP216, TEAC, Tokyo, Japan Sampling rate: 1000 Hz; Electrode placement: 10–20 system | The PSD of the theta, alpha-1 and alpha-2 bands were calculated as indicators of objective drowsiness. | No health impairments; Sample size: 12; All male Age: 39.3 ± 1.8 years; Weight: 64.9 ± 2.3 kg; Body height: 168.4 ± 0.8 cm | Increase power density in theta (4.0–7.9 Hz) and alpha bands (ERS). | EEG-measured excitation values during RAGT decreased within a short time but can be restored and maintained by intermittent warning tone stimulation. |
| Seeber et al.,2013 | Spatial-Spectral Identification of μ and β EEG Rhythm Sources During Robot-Assisted Walking | Lokomat | Conditions: 3 runs standing upright, 3 min each; 4 runs active walking, 6 min each | EEG 120 electrodes Sampling rate: 2.5 kHz; High pass: 0.1 Hz; Low pass: 1000 Hz; Electrode placement: 10–20 system | Power spectra (ERD); Functional brain topography | No health impairments; Sample size: 8; Male: 5; Female: 3; Age: 26.3 ± 3.5 years | Individual mu and beta ERD activities in SMC. The beta-ERD is more focal and consistent among the test persons in the foot area than the mu-ERD. | A method capable of considering individual slight differences in the rhythms mu and beta and locating the ERD activity of these rhythms at the cortical level. Maximum frequencies of ERD were successfully identified for each subject in the frequency range mu and beta. The resulting spectral peaks lead to mu and beta topographies for these frequencies. |
| Seeber et al.,2014 | EEG beta suppression and low gamma modulation are different elements of human upright walking | Lokomat | Conditions: 4 runs active walking 6 min each, 3 runs upright standing 3 min each. Speed: 1.8 km/h - 2.2 km/h (constant, adapted to the test persons) BWS: <30% GF: 100% | EEG BrainAmp, Brain-products, Germany; 4× 32-channel amplifiers combined to 120 channels; Sampling rate: 2.5 kHz Band pass: 0.1–1000 Hz; Electrode placement: 10–20 system (EasyCap, Germany) | Amplitude Modulation & Power Spectra (ERD) | No health impairments; Sample size: 10; Male: 5; Female: 5; Age: 25.6 ± 3.5 years | During active walking, mu (10–12 Hz) and beta (18–30 Hz) oscillations were suppressed (ERD) compared to standing upright. Significant beta ERD was visible in 9/10 subjects in central sensomotoric areas. Low gamma (24–40 Hz) amplitude were modulated in relation to the gait cycle phase. | Persistent mu and beta ERD reflect a motion-related state change in cortical excitability, while gait-phase modulations in the lower gamma represent the motion sequence time during walking. |
| Seeber et al.,2015 | High and low gamma EEG oscillations in central sensorimotor areas are conversely modulated during the human gait cycle | Lokomat | Randomized 8 runs RAGT 6 min each (4 with active walking and 4 with passive walking), 3 runs standing upright. 3 min rest between the runs. Speed: 1.8 km/h - 2.2 km/h; GF: 100%; BWS: <30% | BrainAmp amplifier, Brain-products, Germany; 4× 32 channel amplifiers combined to 120 channels. Sampling rate: 2.5 kHz; Band pass: 0.1–1000 Hz; Electrode placement: 10–20 system (EasyCap, Germany); Additional: EMG | Temporal dynamics of EEG oscillations in the source space by using time-frequency decomposition; Amplitude differences between walking and standing in mu (10–12 Hz), beta (18–30 Hz, gamma (60–80 Hz) and low gamma 24–40 Hz), high gamma (70–90 Hz) | No health impairments; Sample size: 10; Male: 5; Female: 5; Age: 25.6 ± 3.5 years right-handed | Increased gamma (60–80 Hz) amplitudes in central SMC and modulation of high gamma during walking compared to standing; High gamma and low gamma amplitudes are both modulated in relation to the gait cycle, but conversely to each other | Altered synchrony state during walking compared to standing due to static increase of gamma amplitudes; Dynamic, amplitude modulations at 70–90 Hz during gait cycle may reflect gait phase dependent interactions in locomotor network; Distinction of high and low gamma amplitudes in walking experiments due to its negative correlation |
| Villa-Parra et al.,2015 | Toward a robotic knee exoskeleton control based on human motion intention through EEG and sEMGsignals | H2 Exoskelett Additional: UFES’s Smart Walker | Conditions: Sit down/stand up; Knee flexion/extension; 60 trials per 10 s, 3 min rest between conditions | EEG BrainNet BNT 36 electrodes; Additional sEMG | EEG: HMI analysis using: ERD/ERS & ERP (slow cortical potentials); sEMG: myoelectric pattern classification with regard to the lower limb | No health impairments; Sample size: 4; All male | Highest beta ERD in the range from 20 to 24 Hz; Highest beta ERS in the range from 16 to 22 Hz | Combination of EEG/sEMG signals can be used to define a control strategy for the robot system |
| Wagner et al.,2012 | Level of participation in robotic-assisted treadmill walking modulates midline sensorimotor EEG rhythms in able-bodied subjects | Lokomat | Randomized 8 runs RAGT 6 min each (4 with active walking and 4 with passive walking), 3 runs standing upright. 3 min rest between the runs. Speed: 1.8 km/h - 2.2 km/h; GF: 100%; BWS: <30% | EEG BrainAmp DC and MR plus amplifier, Brain-products, Germany; 4×32 channel amplifiers combined to 120 channels. Sampling rate: 2.5 kHz; Band pass: 0.1–1000 Hz; Electrode placement: 10–20 system (EasyCap, Germany); Additional: EMG | Power spectra relating to the active and passive robot-supported gait | No health impairments; Sample size: 14; Exclusion: 1; Male: 8; Female: 7; Age: 22 to 28 (average: 24.3 ± 2.7) right-handed | Mu (8–12 Hz) and beta (18–21 Hz) rhythms are suppressed during active walking compared to passive walking. | Significant differences in cortical activation between active and passive robotic walking support the evaluation of brain monitoring techniques and brain-computer interface technologies to improve gait restoration therapies in a top-down approach. |
| Youssofzadeh et al.,2014 | Directed neural connectivity changes in robot-assisted gait training: A partial Granger causality analysis | ALEX II on non-dominant leg (left) | Speed: 0.87 ± 0.15 m/s. 10 training paradigms Rest: 2 - 4 min. between trials Haptic and visual guidance: 100%. | EEG; g.tec’s; g.USBamp; 16 electrodes; Electrode placement: 10–20 system | PGC to elucidate the functional connectivity of EEG signals in RAGT; PSD for validity check | No health impairments; Sample size: 6 male subjects Age: 26.5 ± 6.5 years; Weight: 77.8 ± 9.7 kg; Body height: 1.79 ± 0.04 m | The results showed a strong causal interaction between the lateral motor cortical areas. A front-parietal connection was found in all robotic training units. After training a causal “top-down” cognitive control was found. | Causal ”top-down” cognition control indicates plasticity in connectivity in the respective brain regions. |
| Youssofzadeh et al.,2016 | Directed Functional Connectivity in Fronto-Centroparietal Circuit Correlates with Motor Adaptation in Gait Training | ALEX II on dominant leg (right) | Conditions: Standing upright, walking unsupported on a treadmill, robot-assisted walking with and without the task of adjusting the changed footpath on a screen. 10 training paradigms Speed: 0.87 km/h ± 0.15 m/s | EEG; g.tec’s; g.USBamp; 16 electrodes; Sampling rate: 512 Hz; Electrode placement: 10–20 system; Additional: EMG | PGC to elucidate the functional connectivity of EEG signals in RAGT; PSD for validity check | No health impairments; Sample Size: 6 male subjects; Age: 26.5 ± 6.5 years; Weight: 77.8 ± 9.7 kg; Body height: 1.79 ± 0.04 m | PGC analysis showed improved connectivity near sensorimotor areas (C3, CP3) during standing while additional connectivity near central (CPz) and frontal (Fz) areas during walking compared to standing. Significant fronto-centroparietal causal effects both in training and after training. Strong correlations between kinematic errors and fronto-centroparietal connectivity during and after training. PSD analysis showed increase in α rhythms during standing, and theta and γ during walking. | Fronto-centroparietal connectivity is a potential neuromarker for motor learning and adaptation in RAGT. |
| Kim et al.,2016 | Best facilitated cortical activation during different stepping, treadmill, and robot-assisted walking training paradigms and speeds: A functional near-infrared spectroscopy neuroimaging study | Lokomat | Randomized, based on a block design; 3 Conditions: Conventional walking, TW, RAGT; Speed: (1) self-selected, (2+3) 1,5, 2.0, 2.5, 3,0 km/h; GF: 100%. BWS: 50% | fNIRS 31-channels | Cerebral hemodynamic changes associated with cortical movement network regions in the primary SMC, PMC, SMA, PFC and SAC | No health impairments; Sample size: 14; Men: 8; Women: 6; Age: 30.06 ± 4.53 years right-handers | More global activation of the motion network (SMC, PMC, SMA) during RAGT compared to conventional and treadmill walking. Positive correlation of speed and activity of the movement network. | RAGT provides the best cortical activation associated with motor control. |
| Simis et al.,2018 | Using Functional near Infrared Spectroscopy (fNIRS) to assess brain activity of spinal cord injury patient, during robot-assisted gait | Lokomat | Conditions: Standing (resting position) and walking in the Lokomat | fNIRS 32 optodes: 16 emitters, 16 detectors; Placement: 10–20 system | Cerebral hemodynamic changes in the motor cortex of both hemispheres. Relative change in concentration of oxy- and deoxyhemoglobin | Patients with spinal cord injury; Number of test persons: 3 patients | Two of the patients had an increased activation in M1 during the RAGT, compared to the standing position. One of the patients showed no changes in M1 brain activity. | fNIRS is suitable for measuring the brain activity of SCI patients during robotic walking. Results indicate an increased involvement of the motor cortical areas during walking. |
ALEX, active leg exoskeleton; BWS, body weight support; CSE, cortical spinal excitability; DAR, delta-alpha EEG power ratio; EGT, ekso-gait training; ERD, event-related desynchronization; ERS, event-related synchronization; ERSPs, event-related spectral perturbation; FPEC, frontoparietal effective connectivity; GAR, gait assistance robot; GF, guidance force; HMI, human-motion-intention; M1, primary motor cortex; OGT, over ground gait training; PFC, prefrontal cortex; PGC, time-domain partial Granger causality; PMC, primary motor cortex; PSD, power spectral density; RAGT, robot-assisted gait training; SAC, sensory association area; sEMG, surface electromyography; SEP, sensory evoked potentials; SMA, supplementary motor area; SMC, sensorimotor cortex; SMI, sensorimotor integration; TBI, traumatic brain injury; TMS, transcranial magnetic stimulation.