EEG-based sensorimotor neurofeedback for motor neurorehabilitation in children and adults: A scoping review

(cid:1) EEG-based sensorimotor neurofeedback is sparsely explored in children with motor disorders and adult populations beyond stroke. (cid:1) Most brain-computer interfaces use upper limb motor imagery to trigger visual, haptic or electrical stimulation neurofeedback. (cid:1) Reporting of EEG neurofeedback parameters and outcomes varies widely: greater transparency is required to validate brain-behaviour changes.


Introduction
Dystonia and dystonic/dyskinetic cerebral palsy (CP) are neurological movement disorders, characterised by sustained or intermittent muscle contractions resulting in abnormal, often painful, twisting movements and postures (Albanese et al., 2013).Whilst adult-onset dystonia is commonly localised and not progressive, childhood-onset dystonia is more often severe and generalised (Bressman, 2004).Dystonia and dystonic/dyskinetic CP are lifelong conditions that adversely affect quality of life, causing physical and psychological challenges (Girach et al., 2019;Skogseid et al., 2007;Zurowski et al., 2013).There is no cure and management options are limited.Neuromodulation with Deep Brain Stimulation (DBS) of the globus pallidus internus has yielded considerable benefits in some individuals with severe, medically refractory dystonia (Gimeno et al., 2013;Marks et al., 2013;Romito et al., 2015;Vidailhet et al., 2009).However, this invasive technique presents its own set of challenges, such as risk of infection and patient anxiety surrounding surgery.Further, patient outcomes of DBS are variable: individuals with acquired dystonia (dystonic/dyskinetic CP) respond more modestly than those with genetic/idiopathic dystonia (Elkaim et al., 2019;Koy et al., 2013;Lumsden et al., 2022;Marks et al., 2013;Vidailhet et al., 2009) indicating a significant need for alternative therapies for this population.
Researchers have begun to explore the efficacy of other therapeutic strategies to augment the outcomes of DBS.For example, proof-of-concept for a rehabilitation intervention, the Cognitive Orientation to daily Occupational Performance (Polatajko & Mandich, 2004) (CO-OP) has been established for childhoodonset hyperkinetic movement disorders (Gimeno et al., 2019;Gimeno et al., 2021;Gimeno et al., 2020).CO-OP is a performance-based, client-centred intervention aimed at improving performance in self-identified functional goals, instead of solely focusing on reducing dystonic symptoms.Although this work is encouraging, non-pharmacological and non-invasive interventions are lacking, and there is a critical clinical need to develop innovative therapies.Further work to understand the pathophysiology underlying dystonia and dystonic/dyskinetic CP is key to informing the development of new data-driven therapeutic approaches.
Recent research using electroencephalography (EEG) demonstrates that cortical sensorimotor processing, specifically modulation of the brain rhythm 'mu', is abnormal in children and young people (henceforth referred to as children) with dystonia and dystonic/dyskinetic CP (McClelland et al., 2021).Arising from the central/midline fronto-parietal sensorimotor brain region, the mu rhythm, also known as the sensorimotor rhythm (SMR), comprises two components, a prominent alpha/mu (8-13 Hz) rhythm and a smaller contribution from a beta (13-30 Hz) rhythm, which have a near harmonic relationship (Wischnewski et al., 2022).Mu is strongly associated with cortical sensorimotor processing: in particular, mu oscillatory activity is reduced in power in response to movement or somatosensory stimulation.This phenomenon is termed an event-related desynchronisation (ERD) (Neuper et al., 2003;Pfurtscheller et al., 2000) and is considered to reflect activation of the sensorimotor cortex.This is usually followed by an event-related synchronisation (ERS) whereby the cortex is deactivated, which is associated with motor control and inhibition, movement outcome and error processing (Pfurtscheller, 2001;Torrecillos et al., 2015).Importantly, mu ERD and ERS can also be evoked by observed or imagined movement, a phenomenon which is exploited in the development of brain-computer interfaces (BCIs) (Broetz et al., 2010;Jeunet et al., 2019).
EEG-based BCI systems acquire and detect changes in cortical activity with high temporal resolution and translate EEG signals into output commands in real-time, allowing the participant to control an external device (such as a switch or a remotecontrolled wheelchair) or engage with computer systems.In addition to enabling device control, the closed-loop paradigm of a BCI can provide real-time neurofeedback of a specific brain rhythm via various modalities such as visual, auditory, haptic or electrical stimulation.The neurofeedback encourages the participant to gain voluntary control and self-regulation of the selected brain rhythm, usually the mu/SMR activity (Sitaram et al., 2017), through operant conditioning or associative learning, capitalising on the Hebbianassociated and long-term potentiation-like mechanisms of neuroplasticity (Mrachacz-Kersting et al., 2012;Sitaram et al., 2017).Thus, while some EEG-based BCI systems are developed as assistive devices for those with difficulties to communicate or perform motor activities, others are designed specifically for neurorehabilitation.The latter focus on the modulation of mu to enhance sensorimotor control, with the potential of improving motor capabilities and alleviating clinical symptoms (Young et al., 2021).Positive effects of neurofeedback have been demonstrated in adults with stroke (Broetz et al., 2010;Cervera et al., 2018;Jeunet et al., 2019;Remsik et al., 2019).However, the application of EEG-based BCIs as a neurorehabilitation technique in children (Kinney-Lang et al., 2016) and adults with dystonia or dystonic/ dyskinetic CP is relatively unexplored.
Motor imagery (MI) has been used as a cognitive strategy by people with Parkinson's Disease (Nonnekes et al., 2019), and also in children with hyperkinetic movement disorders including dystonia and dystonic/dyskinetic CP (Butchereit et al., 2022) who showed subsequent improvement in motor performance and skills acquisition (Gimeno et al., 2019;Gimeno et al., 2020).Other strategies include distraction, mental self-guidance, internally/externally focused attention, and emotional regulation (Butchereit et al., 2022).It is likely that the use of MI as a strategy engages mu modulation and activation of the sensorimotor network, relevant to many EEG-BCI systems, whilst other cognitive strategies such as mindfulness meditation can improve EEG-BCI control in healthy adults (Stieger et al., 2021;Tan et al., 2014).We were therefore interested in how such strategies have been used in EEGneurofeedback studies in populations with neurological disorders.

Aim of study
We planned to conduct a scoping review to establish (i) the extent of research investigating EEG-based sensorimotor neurofeedback techniques in children with dystonia and dystonic/dyskinetic CP, and (ii) whether any (cognitive) strategies have been used to augment neurofeedback.E. Cioffi, A. Hutber, R. Molloy et al. Clinical Neurophysiology 167 (2024) 143-166 From an initial search of the literature, we found no evidence of EEG-based sensorimotor neurofeedback research in children with dystonia or dystonic/dyskinetic CP.Further, our search identified only two studies exploring such techniques in adults with dystonia.Expanding the search to include both adults and children with CP (all types, rather than specifically dystonic/dyskinetic) still only yielded six articles.Therefore, the scoping review was broadened to include adult and paediatric populations and to span a wider range of neurological motor impairments.

Methods
This study followed the Joanna Briggs Institute guidelines (Peters et al., 2020), underpinned by the Arksey and O'Malley (Arksey & O'Malley, 2005) and Levac and colleagues (Levac et al., 2010) frameworks, and was conducted in line with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) Checklist (Tricco et al., 2018).The review protocol was registered prospectively in the Open Science Framework (OSF) database (DOI 10.17605/OSF.IO/ SKH85).

Development of the research question
The research question was developed using the Population, Concept, Context framework (Peters et al., 2020) and extended as outlined in the aims above.The primary research question which guided the review was: 'How has EEG-based sensorimotor neurofeedback been used in rehabilitation for children and adults with neurological motor impairments?'.The secondary question was, 'Have other techniques been used to augment this feedback?'.

Search strategy
Two reviewers independently conducted an initial search in MEDLINE and Cumulative Index to Nursing and Allied Health Literature (CINAHL) databases to gain an understanding of the breadth of relevant literature.Text words in the titles and abstracts were used to establish search terms and develop the full search strategy (Supplementary Material, Appendix A), which was subsequently reviewed by a librarian following the Peer Review of Electronic Search Strategies (PRESS) checklist (McGowan et al., 2016).The final, full search strategy was used to identify literature published up until August 2022 in MEDLINE, CINAHL and Web of Science, with the search strategy modified for each database where necessary.Due to the high volume of papers and time taken to screen articles, a second more recent search with identical search parameters was conducted in October 2023 to ensure findings remained current and incorporated the most recently published research.

Study selection
The inclusion and exclusion criteria were developed and refined by the team, based on an initial screening of a small sample of articles.One modification was to exclude articles whereby feedback related only to electromyography (EMG), whereas articles using combined EMG and EEG feedback were retained.
The final inclusion criteria included: (1) patients with neurological motor impairments (i.e., stroke, dystonia, cerebral palsy, Parkinson's Disease and multiple sclerosis); (2) Scalp EEG-BCI systems processing signals recorded over the sensorimotor cortex (encompasses mu, beta and alpha where this represented mu rhythms); (3) participants of all ages; (4) any type of experimental study designs reporting original data; (5) any publication year; (6) studies published in English.Although reviews were excluded, ref-erence lists were examined to identify further eligible studies not already captured.
Titles and abstracts of identified articles were exported and uploaded into Rayyan screening software, and duplicates were removed.Titles and abstracts were screened by reviewers (EC, RM, SM) independently against the eligibility criteria, with at least 20% screened by at least two reviewers to ensure consistency.Any disputes were discussed until consensus was reached or resolved by a third reviewer.The full texts of included articles were examined further to confirm eligibility and reviewed independently by two reviewers (EC and VM), with at least 25% screened by both.Disagreements were resolved through discussion.Reasons for exclusion at full-text screening stage were documented, including those listed above and a further category for articles providing insufficient methodological information (e.g.EEG neurofeedback parameters unclear).

Data extraction
A data extraction form was included in the scoping review protocol and uploaded to OSF before commencing the study.Data extraction included study design, participant information (motor impairment diagnosis, age), sensorimotor task parameters, neurofeedback mode, EEG-based sensorimotor neurofeedback parameters, outcome measures and augmented strategy use.
Three members of the research team (EC, VM, HG) independently extracted and compared five articles of different methodologies to ensure data extraction captured all relevant aspects, resulting in some minor refinements.Data extraction was completed independently by two reviewers (EC and AH), with at least ten percent of included articles extracted by both reviewers to ensure consistency.Throughout the extraction process, team members met regularly to discuss any uncertainties and ensure accuracy.Disagreements between reviewers were resolved through discussion or by senior authors (VM and HG) when necessary.

Results
In the initial search, 126 articles out of 4,373 (total retrieved from database searching and reference lists after de-duplication) were included, based on the screening steps and exclusion reasons outlined in Fig. 1.A further seven articles were included from the second search.Thus, 133 articles were included in total.Most commonly, articles were excluded because the EEG-based neurofeedback signal(s) was not recorded from the sensorimotor cortex or, in some cases, this area was included in the recording, but the feedback signal was not based on the SMR (n = 26).

Year of publication
The 133 included articles are listed in Table 1.All were published in the last 25 years (1998 -2023), with 101 (76%) published within the last 10 years (Fig. 2A).

Study design
The study design of included articles was categorised based on the Oxford Centre for Evidence-Based Medicine (OCEBM) (OCEBM, 2011).Most studies were case-series (n = 57, 43%) or case reports of three or less patients (n = 39, 29%), which we categorised as level 4. Of these level 4 studies, 32 reported single patient case reports and 43 studies enrolled between two and ten patients in an intervention condition.Although a proportion of articles reported randomised controlled trials (n = 37, 28%), these were mostly E. Cioffi, A. Hutber, R. Molloy et al. Clinical Neurophysiology 167 (2024) 143-166 unpowered studies without sample size estimations.Four articles reported enrolling sample sizes large enough to achieve a statistical power of 80-95% (Frolov et al., 2017;Norouzi & Vaezmousavi, 2019;Tsuchimoto et al., 2019;Zanona et al., 2023).However, the derivation of these power analyses was sometimes unclear.

Sensorimotor task parameters
The sensorimotor task parameters employed in each study are displayed in Fig. 3A, including limb(s) involved and sensorimotor task.Most commonly, participants were asked to perform motor imagery (MI) only (n = 75, 56%).Other studies involved participants attempting or executing actual movement (n = 38, 29%).A portion of studies explored both paradigms (n = 9, 7%).In two of these cases, the sensorimotor task depended on the severity of participants' motor impairment or individual preference (Hohne et al., 2014;Tan et al., 2010), and another three combined MI and motor attempt to trigger neurofeedback (Carrere et al., 2021;Cincotti et al., 2008;Zhang et al., 2018).The remaining four studies consisted of two parts, with the BCI controlled initially through MI followed by motor attempt, or vice versa (Daly et al., 2009;Hortal et al., 2015;Norman et al., 2018;Pitt & Brumberg, 2022).
The upper limb(s) was most frequently studied (n = 116, 87%) with tasks mainly involving reaching (elbow extension-flexion) and/or grasping (wrist and finger extension-flexion).In the 15 paradigms involving the lower limbs, the tasks included imagining or attempting foot or ankle dorsiflexion.In a proportion of studies (n = 11, 8%), participants were not instructed to perform or imagine movement, but rather were simply told to minimise movement and stay mentally focused.In one study, participants were asked to ''find a mental condition" that successfully reduced alpha power, which in turn played a video, although the strategies employed by participants were not reported (Alves-Pinto et al., 2017).However, in another study, despite not being specifically directed to, it was reported that participants tended to use motor imagery to modulate SMR and control a visual cursor (Wolpaw & McFarland, 2004).
Two of the studies enrolling children employed MI of the upper limb (Bobrov et al., 2020;Jadavji et al., 2023), whilst the third used MI or motor attempt of the upper or lower limbs (Cincotti et al., 2008).

Feedback mode
Fig. 3B summarises the feedback modes used.The most prevalent mode was visual only (n = 34, 26%) and involved feedback on a display screen, for example, vertical or horizontal movement of a cursor towards a target, colour change of a target, or variations in bar height.Visual feedback was also incorporated in most interventions delivering multimodal feedback (n = 57, 43%).In 13 studies, visual feedback was delivered in the form of virtual reality, whereby participants' sensorimotor EEG signals controlled the movement of virtual hands performing a motor task.
Haptic feedback was also commonly used, mostly delivered by a robotic device with the BCI using sensorimotor EEG signals to initiate movement.Robotic devices included exoskeletons, mechani-  cal orthoses and continuous passive motion machines attached to participants' impaired limbs.Other haptic feedback included vibrotactile (Vourvopoulos et al., 2019a) or brush (Hu et al., 2021) stimuli.A further mode of feedback involved functional electrical stimulation (FES) to facilitate movement.Robotic devices and FES were used solely (16 and 13 studies, respectively) or in combination with visual feedback on a display screen (23 and 15 studies, respectively).Four studies administered electro-tactile stimulation to the tongue, along with visual and FES feedback (Remsik et al., 2018;Remsik et al., 2019;Sinha et al., 2021;Young et al., 2014).Other less common feedback modes included auditory and neuromuscular electrical stimulation.
Two of the paediatric studies delivered visual feedback on a display screen (Cincotti et al., 2008;Bobrov et al., 2020), whilst the other delivered both visual feedback and FES (Jadavji et al., 2023).

EEG-based sensorimotor neurofeedback parameters
This review focused on studies that report the use of sensorimotor EEG signals for neurofeedback.Where these details were provided, the frequencies and EEG channels used for neurofeedback, as well as the signal processing tools and features extracted, are listed for each study in Table 1.The extent to which these parameters were reported varied widely, often depending on the BCI design.Within a given study, the parameters used for online and offline EEG analysis were not necessarily the same.Therefore, parameters for online and offline EEG analysis, where performed, are listed separately.Overall, 72% (n = 96) of studies reported details of both online and offline parameters, although often details were incomplete (Table 1).
Most studies (n = 90, 68%) used a BCI that incorporated machine learning algorithms, such as Common Spatial Patterns (CSP) and Support Vector Machines, and/or classifiers, such as Bayesian or Linear, in their signal processing pipeline to identify participantspecific spatial and/or frequency features from the EEG signal for neurofeedback (Table 1).Twenty-five of these studies used unspecified classifiers and algorithms within the BCI2000 software system.Often, studies were focused on BCI system feasibility and used classification accuracy (CA) to evaluate the performance of the BCI model (see section 3.8.1).The EEG parameters in these designs tended to comprise a broader frequency range (e.g., 5-30 Hz or 0.05-40 Hz), including the alpha and beta components of the sensorimotor rhythm.After calibration, this would then be refined to a participant-specific frequency band for training sessions.There was variability in the EEG channels used for neurofeedback: whilst some studies reported the specific channels used, such as C3, Cz and C4 in the sensorimotor region, often channels or spatial components were selected after calibration by classifiers that showed the strongest cortical activation or highest power associated with different mental states or tasks.
EEG features used for neurofeedback were mainly focused on detecting and processing ERD of mu, alpha, beta or an unspecified rhythm (n = 103, 77%).Five papers mention ERD in the introduction or results, but this was not specified within the methodology (Ang et al., 2014a;Ang et al., 2011;Broetz et al., 2010;Chen et al., 2021;Pfurtscheller et al., 2000).A further fourteen studies used SMR power, often without specifying the nature of the change, and two studies used functional connectivity in the alpha frequency band for EEG neurofeedback (Mottaz et al., 2018;Mottaz et al., 2015).The remaining studies did not report a specific EEG feature used for neurofeedback (n = 11, 8%).
All three paediatric studies used ERD as the neurofeedback feature.However, this was detected across varying channels and frequencies.Two studies used CSP and/or classifiers to detect ERD in a broad frequency range across many channels (Bobrov et al., 2020;Jadavji et al., 2023).The remaining study utilised a narrower frequency range (3-14 Hz) over a sub-set of 59 channels, using unspecified classifiers within the BCI2000 software system (Cincotti et al., 2008).

(Offline) EEG analysis
Of the 133 included studies, 96 (72%) performed offline EEG analysis, mostly focusing on spectral power measures such as ERD/event-related spectral perturbation -ERSP (n = 90, 68%).A subset of studies (n = 28, 21%) analysed other quantitative spectral measures such as entropy, EEG:EEG coherence and corticomuscular coherence, with one study analysing and comparing a large array of EEG features (Erickson-Davis et al., 2012).More recently published studies have focused on a range of functional and effective connectivity measures such as coherence (generalised partial directed coherence, magnitude squared coherence and the imaginary component of coherence), direct transfer function, global and local efficiency, clustering coefficient, node strength, network density and phase slope index (Chang et al., 2023;Kern et al., 2023;Mottaz et al., 2018;Mottaz et al., 2015;Remsik et al., 2021;Yuan et al., 2021;Zhan et al., 2022;Zhang et al., 2023).
A small number of studies additionally analysed measures of brain symmetry and lateralisation.Three studies looked at the brain symmetry index (BSI) to capture differences in spectral power between the cerebral hemispheres (Ang et al., 2014a;Kumari et al., 2022;Zhang et al., 2018).Other studies analysed ERD-based lateralisation index to assess the strength and/or timing of ERD in different brain regions in relation to specific cognitive or motor functions (Braun et al., 2017;Jia et al., 2022;Kumari et al., 2022;Remsik et al., 2019;Vourvopoulos et al., 2019a;Vourvopoulos et al., 2019b).

Outcome measures
Reported outcomes included measures of BCI system feasibility, BCI participant performance and clinical outcome scores.Usability was also documented.

BCI system feasibility
BCI classification accuracy (CA) was used in 20 studies to measure the feasibility and performance of the BCI system itself.CA reflects the classifier's ability to accurately detect and differentiate between different mental states and translate the participant's brain activity into commands (Yuan & He, 2014).All 20 studies reported that CA was better than chance level (50%).Fifteen reported accuracy was greater than the recommended minimum E. Cioffi, A. Hutber, R. Molloy et al. Clinical Neurophysiology 167 (2024) 143-166 accuracy level for a BCI system (70%) for at least one training session (Kübler et al., 2004).CA may also be reported as a measure of participant performance (see below).

BCI participant performance
Just over half of the studies reported one or more measures related to participants' performance in the BCI neurofeedback task (n = 73, 55%), as shown in Fig. 4. The enhancement of ERD from pre-to post-neurofeedback training -i.e., the performance measure most directly related to the targeted neurofeedback signal itself (ERD) -was reported in 41 articles (31%).However, 29 of these 41 articles displayed this result graphically without specifying numerical values in the text.A further stated there had been an enhancement of ERD but did not report values either graphically or in the text.For the nine studies that reported the value of ERD enhancement, this was reported differently across studies depending on their methodology.In some studies, ERD was expressed as the percentage change in power from baseline, (also described in some articles as the mu suppression score) while in others it was expressed as a ratio between the average band-power during the motor task and the reference period.Others reported the signed r-squared coefficient of determination value.In some articles, the pre-and post-intervention ERD values were reported; in others, only the change in ERD from pre-to post-intervention was stated and a few reported both.Some articles reported the group mean values of ERD (either percentage or ratio) for an experimental group versus a control group, or for the contralesional versus the ipsilesional hemisphere.The variability in reporting makes it difficult to compare values of ERD enhancement across studies.
For example, Chowdhury et al. ( 2020) calculated the ratio between the average band-power during the motor task and the reference period.They showed graphically that this measure decreased over the weekly neurofeedback sessions (i.e.there was a change in favour or mu ERD rather than ERS); they reported that the group-mean ratio changed from 1.03 to 0.74 over time, amounting to a change of À28.36%, which represents a statistically significant (p < 0.05) enhancement in ERD.In contrast Remisk et al. (2019) reported the signed r 2 coefficient of determination value, calculated from the absolute mu power during movement trials compared with rest trials, with negative values indicating a mu ERD.They demonstrated a statistically significant enhancement in mu ERD for the ipsilesional hemisphere following therapy (mean r squared value pre-and post-intervention À0.142 and À0.161 respectively, p = 0.039).
CA was also used in 44 studies as a measure of how successfully participants controlled the BCI system (separately from reflecting system feasibility).This depicted the percentage of times participants successfully operated the BCI to trigger neurofeedback.CAs ranged from over 50 to above 90%, with accuracy tending to increase over intervention periods, demonstrating improved participant performance over time.Factors reported to have potentially influenced CA included fatigue (Prasad et al., 2010;Resquin et al., 2016), type of MI task (Pfurtscheller et al., 2000) and medication (Irimia et al., 2018).Other performance measures included success rate, defined as the percentage of successful trials or targets attained, and SMR modulation (not specified as ERD).
All three paediatric studies used BCI CA to measure performance.Changes in the neurofeedback feature itself, the mu ERD, were not reported.

Clinical outcome scores
Most studies reported improvement in motor function using clinical outcome measures (n = 86, 65%), most commonly the Fugl-Meyer Assessment (FMA) (n = 59, 44%), Action Research Arm Test (n = 21, 16%) and/or (Modified) Ashworth Scale (n = 19, 14%).These measures reflect the large proportion of studies enrolling stroke participants.Twenty-two studies reported that improvements in clinical motor scores were sustained at followup, with 13 studies reporting statistically significant improvements.Follow-up time points ranged from one-to-twelve months post BCI therapy, with two studies reporting statistically signifi- cant improvement in upper limb FMA scores persisting at six (Zhang et al., 2023) and six-to-twelve (Biasiucci et al., 2018) months post-BCI training.For both of these studies, more BCI participants than controls achieved clinically significant scores.However, most studies did not investigate the relationship between clinical outcomes and neurophysiological outcome measures.Twenty-eight studies (21%) explored an association between motor outcomes and participant performance.Twenty-one of these conducted a formal correlation analysis, with 15 studies reporting a statistically significant positive correlation (Table 1).The remaining seven performed informal correlation analysis i.e., associations were commented on, but no statistical analyses were reported.
Sixteen of the 28 studies explored an association between motor improvement and a specific ERD-related outcome, i.e., the performance measure directly related to the neurofeedback signal (ERD).Two of these were single case reports (Mukaino et al., 2014;Naros & Gharabaghi, 2015).The other 14 are described in Table 2.All 14 studies reported improvements in at least one clinical measure, with nine reporting statistically significant improvements (Table 2).Eight of these studies reported some improvement in clinical scales surpassing the minimal clinically important difference (MCID).
Eight out of the 14 studies reported ERD enhancement from pre-to post-neurofeedback training.However, ERD enhancement was often shown graphically without numerical values noted in the text.The remaining six studies correlated changes in motor function with a single ERD measurement, rather than ERD modulation from pre to post training.Overall, six studies found a statistically significant correlation between motor improvements and an ERD-related outcome.
Thirteen studies conducted formal correlation analyses (and one informal) between motor improvement and additional performance measures, such as CA (n = 7) or coherence (n = 4).While these correlations are listed in Table 1, further details have not been examined in Table 2 as these outcomes are indirectly related to the neurofeedback signal.As these are indirect measures of neurofeedback performance, it is not possible to conclude that any observed motor improvements are attributable specifically to the neurofeedback training.
Only one paediatric study measured motor improvement clinically, reporting significant improvements in sub-scales of FMA, ARAT and Jebsen-Taylor function tests (Bobrov et al., 2020).However, associations between these improvements and neurofeedback performance were not explored.

Usability
Thirty-one articles commented on the usability of the BCI neurofeedback system.Standard usability assessment tools included NASA task load index (Kumari et al., 2022;Sakamaki et al., 2022;Zulauf-Czaja et al., 2021), visual analogue scale (Chowdhury et al., 2018;Prasad et al., 2010;Zich et al., 2017), simulator sickness questionnaire (Vourvopoulos et al., 2019b;Vourvopoulos et al., 2019c), Quebec user evaluation of satisfaction with assistive technology (Nishimoto et al., 2018;Zulauf-Czaja et al., 2021) and system usability scale (de Castro-Cros et al., 2020).Other studies did not report use of standard scales but conducted questionnaires or interviews exploring participants' levels of enjoyment and satisfaction, motivation and fatigue, workload (mental or physical demand) and comfort.No studies reported adverse events.Generally, participants reported enjoyment and satisfaction with the EEG-BCI training.
Eight studies reported increases in fatigue over the course of BCI training, two of which suggested that fatigue might have contributed to a larger variability or decline in BCI performance (Prasad et al., 2010;Resquin et al., 2016).Four studies delivered one or two training sessions (Hortal et al., 2015;Jadavji et al., 2023;Resquin et al., 2016;Sakamaki et al., 2022), whilst three delivered 10 -12 sessions (Frolov et al., 2017;Frolov et al., 2016;Prasad et al., 2010).The final study required participants to undertake at least 60 training sessions at home over 12 weeks (Bundy et al., 2017).This subset of studies reflects the broader variability across all identified studies with respect to the number of runs, trials, sessions and rest intervals employed in training interventions.
To improve engagement in the EEG-BCI systems, participants' suggestions included a ''pause" feature to reduce fatigue (Leeb et al., 2013), introducing variations in the neurofeedback game's animations and auditory stimuli (de Castro-Cros et al., 2020), and increasing the challenge level (Prasad et al., 2010).
Two of the paediatric studies reported usability.One study administered a questionnaire assessing BCI system acceptance, with the children reporting independent use of the system (Cincotti et al., 2008).The other paediatric study explored levels of fatigue, comfort and engagement (Jadavji et al., 2023).The most common complaints were headset discomfort (58%) and muscle fatigue (50%).The children ranked the BCI intervention as comparable to a long car ride.

Control and comparison groups
Overall, 63 studies incorporated a control condition.Most enrolled a distinct control group using a between-participant design (n = 51, 38%).Of these, 34 studies recruited only participants with neurological motor impairments who were randomly assigned to an experimental or control group.In the experimental group, the BCI system delivered neurofeedback based on patients' EEG signals (EEG-BCI), whereas the control group received sham feedback, or training that did not involve the BCI system.The remaining 17 between-participant study designs enrolled healthy volunteers as the control group, and both patients and controls received the same experimental intervention (EEG-BCI).
A smaller proportion of studies carried out a within-participant design with each patient serving as their own control (n = 12, 9%).In some, patients engaged in a cross-over control design where outcomes were measured during a control versus a BCI therapy phase (Mottaz et al., 2018;Remsik et al., 2018;Remsik et al., 2021).Alternatively, patients received EEG signal-driven feedback versus sham feedback (Alves-Pinto et al., 2017;Mukaino et al., 2014;Ono et al., 2013;Ono et al., 2018;Takahashi et al., 2012;Tsuchimoto et al., 2019;Wada et al., 2019), or engaged in trials with versus without neurofeedback (Kasahara et al., 2018;Pfurtscheller et al., 2000).Five studies compared the clinical improvement in patients' motor functioning between trial types or design phases (Mukaino et al., 2014;Ono et al., 2013;Ono et al., 2018;Takahashi et al., 2012;Wada et al., 2019).Three of these reported greater improvement after EEG-driven feedback compared to sham feedback (Mukaino et al., 2014;Ono et al., 2013;Takahashi et al., 2012), and two reported improvements only after the BCI therapy phase compared to the control phase (Ono et al., 2018;Wada et al., 2019).However, only one of these reported an associated trend between improvement in motor function and increase in ERD (neurofeedback performance) (Ono et al., 2018).
Eighteen studies compared two or more techniques of delivering neurofeedback.For example, three studies compared two neurofeedback modes: visual feedback on a display and haptic feedback via a robotic device (Frisoli et al., 2012;Ono et al., 2014;Sakamaki et al., 2022).Two studies reported that the BCI CA of the robotic feedback condition matched that of the visual feedback condition (Frisoli et al., 2012;Sakamaki et al., 2022).The third study reported improvement in finger function in the robotic feedback condition only (Ono et al., 2014).Control or comparison groups were not included in the paediatric studies.
3.10.Cognitive strategies and additional therapies 3.10.1.Motor imagery strategies Some studies indicated that BCI performance was influenced by the type of MI strategy performed.For example, a single case study reported that the participant's CA varied between 50 and almost 100%, with right-and left-hand MI strategy yielding relatively moderate classification rates, whilst foot MI increased CA considerably (Pfurtscheller et al., 2000).Additional studies suggested the importance of identifying participant-specific MI strategies that optimally support BCI control (Leeb et al., 2013;Neuper et al., 2003).In a further study in which participants explored different strategies, participants reported that employing more complex MI strategies (imagining hair combing and ironing vs opening/closing of hand) was more effective at controlling the BCI (Lioi et al., 2020).

Augmentative cognitive strategies
Cognitive strategies beyond MI were proposed to participants to assist them in controlling the BCI and to augment the neurofeedback in two studies: in one study, researchers suggested participants try mentally counting numbers (Spychala et al., 2020).In another study, to control the BCI, participants were asked to attempt and subsequently imagine 'finger individuation' (i.e., extending one finger while inhibiting the movement of another) (Norman et al., 2018).This task required complex cognitive effort to make cue-based decisions.
Cognitive strategies and additional therapies were not explored in the paediatric studies.

Discussion
Sensorimotor EEG-based neurofeedback has exciting therapeutic potential as an intervention for under-served clinical populations such as childhood-onset movement disorders.This scoping review maps the breadth of research exploring EEG-based sensorimotor neurofeedback in both children and adults with neurological motor impairment.The temporal profile of included articles (Fig. 2A) indicates the rapid expansion of the field, with growing interest from engineers and clinicians in the potential benefits for patients.However, there is a paucity of evidence on the application of these systems in children, with most studies focusing on adults with stroke (Fig. 2B).Furthermore, work is largely at an early stage on the spectrum from physiological proof-of-principle to full clinical translation, and this is reflected by the OCEBM classification of studies, with the majority being level 4.Even among articles described as RCTs, most were unpowered studies without sample size estimations, thus cannot be classified as level 2 evidence.Rather, these are exploratory pilot studies.This is to be expected given this is a relatively new and emerging field, and we emphasise the importance of these early studies for answering important methodological questions which will inform the design of full-scale RCTs in due course.

EEG-based sensorimotor neurofeedback
This review focused on studies using EEG-based sensorimotor neurofeedback.If real-time EEG data are to be used as the basis of a proposed clinical neurorehabilitation intervention, then both the recording parameters used, and the data quality are of paramount importance.However, there was considerable variability across the literature in how EEG parameters were reported.Nine studies were excluded from the review at the full-text screening stage as the reported EEG details were insufficient to determine the nature of the signal being used for neurofeedback.For example, some papers stated that the BCI system ''recognised the brain signals of patients and converted these into motor commands" but did not specify further methodological details.Across the included studies, there was also considerable variation in EEG frequency ranges, topography and the use of processing algorithms and classifiers (Table 1).To aid clarity, we separated the parameters reported in each paper into those relating to the online EEG used for neurofeedback and those relating to subsequent offline analysis.
Online signal processing involves extracting immediate, relevant features from EEG, as it is being recorded, to be used as neurofeedback.Almost half of the studies reported online EEG parameters covering multiple frequency bands, e.g., 0-45 Hz.Within this group, a small proportion used signal processing techniques to determine optimal participant-specific frequency ranges within the broader spectrum, but most did not.Therefore, there is ambiguity as to which frequency/rhythm may be influencing any neural effects observed.
Offline signal processing can be employed following calibration sessions of BCI training, or after the neurofeedback intervention itself, to assess participant performance and to perform a more comprehensive evaluation of changes in brain activity.This can include identifying and extracting complex patterns and nuanced EEG features related to specific cognitive states, tasks, or conditions (Mrachacz-Kersting & Aliakbaryhosseinabadi, 2018).Where offline EEG analysis was reported, frequency ranges were sometimes specified in more detail, but still not consistently.
Accuracy and timing of neurofeedback are further key considerations.Examples of raw EEG data were rarely provided, or such figures were often too small for readers to judge the data quality or exclude the possibility of significant contamination by EMG or other artefacts.When providing a participant with ''real-time" feedback of their EEG activity, there is necessarily a lag-time between the detection of EEG signal change and the delivery of feedback.This is because a real-time system needs to record and process ''packets" of data of a given length, which will vary between systems and studies, as will the interval at which feedback to the participant is updated (the feedback update interval (Darvishi et al., 2017)).Given the speed of physiological neural activity, shorter neurofeedback update intervals are likely to facilitate individuals in learning to modulate their EEG signals effectively.In the context of gaming and visual feedback, a ''lag" is felt with an update rate of more than around 200 ms.
In practice, online processing, with appropriate decisionmaking and adjustments, can be challenging to implement and the ideal of ''instantaneous" neurofeedback is difficult to achieve.Of the studies that reported the feedback update interval, these ranged from four milliseconds to one second, whereas others did not describe these temporal details, or reported that ''data were transmitted in real-time" or that feedback was ''sufficiently fast".

Performance and outcomes
There was also considerable variability across the literature in the performance measures and outcomes reported.These included BCI system performance, BCI participant performance, participant improvement in the selected aspect of brain activity (e.g., mu ERD) during the study, changes in other measures of brain function (e.g., neuronal connectivity), clinical outcome scores aiming to detect changes in motor function and, finally, correlation between the above measures.
We highlight the distinction between the performance of the BCI system itself and the performance of the participant: BCI system performance is a measure of the technical performance and effectiveness of the interface in correctly detecting and classifying the relevant changes in cortical activity, often expressed as classification accuracy (see section 3.8.1).It does not directly translate into, or guarantee, participant proficiency in controlling a BCI.Conversely, participant performance is a measure of the ability of an individual to learn effective control of a BCI and is influenced by many other factors such as session design, participant adaptation, feedback mechanisms, ability to perform motor imagery, fatigue and cognitive factors.It can be difficult to separate these aspects, due to the interaction between the BCI and participant, but it is important that they are considered.In addition, neurofeedback studies using BCIs for neurorehabilitation will often aim to enhance a particular feature of brain activity (e.g., mu ERD).Therefore, it is relevant to report not only whether the participants could use/control the BCI but also whether participants showed an enhancement of this brain activity feature between the start and end of the study (or pre-and post-training), indicating neuroplasticity.
The outcomes and performance measures reported often reflected the nature of the paper and whether the focus was on the development of an assistive/restorative BCI for those lacking movement or on developing/testing a BCI to provide neurofeedback for sensorimotor training (rehabilitative BCIs).Again, there is a degree of overlap: even where BCIs are used primarily with an assistive/restorative purpose, there will effectively be an element of positive feedback to the participant through the successful performance of an action (e.g., movement of a wheelchair or a remote-controlled toy car), which in turn may lead to neuroplastic change.Since the defined question for our review related specifically to how neurofeedback has been used in rehabilitation, papers reporting the development of assistive BCIs without a particular focus on neurofeedback were excluded.However, even some of the included papers, which focused on rehabilitation, measured performance purely based on the ability of the BCI system to detect the relevant change in cortical activity, rather than reporting participant performance separately (see section 3.8.1).
The variability in methodology and reporting of ERD enhancement makes it difficult to compare findings across studies.The two specific articles mentioned as examples in section 3.8.2both provide detailed descriptions of their methodology and analysis, and both demonstrate statistically significant enhancement in ERD following the intervention, but it remains difficult to compare the findings.A consideration for future groups reporting EEG-based neurofeedback studies would be to include a measure of the effect size, which may facilitate comparison across studies to some extent.Open source sharing of methodologies and raw data could also be beneficial.
While most studies reported participant performance in the neurofeedback task and reported improvement in motor function in terms of a clinical outcome score, fewer papers investigated whether the reported motor/clinical improvement correlated with the feedback-related neural changes.This highlights a very important knowledge gap: without this form of study design and analy-E.Cioffi, A. Hutber, R. Molloy et al. Clinical Neurophysiology 167 (2024) 143-166 sis, it is not possible to determine whether improved clinical scores arise specifically from the neurofeedback training, or whether they could just reflect a non-specific improvement relating to engagement in a study and its associated motor activities.Demonstration of ''brain-behaviour relationships" is therefore a crucial step in building an evidence-base for neurofeedback interventions (Khademi et al., 2022;Mottaz et al., 2018;Ros et al., 2020).Although many studies included in our review focus on ERD enhancement, others explored additional EEG outcome measures that may evolve our understanding of the impact of neurofeedback training.These features include measures of spectral power distribution such as the brain symmetry index or lateralisation index, measures of communication across brain regions, such as coherence and directed transfer function and network analysis parameters such as the global efficiency and clustering coefficient (see Section 3.7).Exploring these phenomena provides valuable insights into how neurofeedback training can modulate not only the original target feature (i.e., ERD), but also the complex and dynamic networks within the brain.Many motor disorders, including dystonia, are now considered network disorders (Lattore et al., 2020;McClelland et al., 2023), so exploring the relationship between these different neurophysiological phenomena is pertinent to understanding their pathophysiology.Although not a focus of the current review, studying the response to neurofeedback in healthy volunteers can also reveal important insights into plasticity within these sensorimotor networks and how this in turn relates to motor function/behaviour.Comparing how these networks function differently between participants with and without movement disorders could in turn inform the development of more effective interventions.

Clinical population
Despite the growing interest in this field, there is a significant gap in research investigating EEG-based neurofeedback for rehabilitation in patients with neurological motor impairments other than adult-onset stroke (Fig. 2B).In particular there are very few studies in children, with only three paediatric studies meeting the criteria for this review (Bobrov et al., 2020;Cincotti et al., 2008;Jadavji et al., 2023).This is concordant with findings from a recent scoping review, published since our original search, which focussed on improved motor outcomes in children and adults with non-progressive neurological disorders undergoing BCI-based neurofeedback training (Behboodi et al., 2022).Although aiming to explore the scope of the published literature in adults and children, all 23 of their included studies were in adults, with 22 in stroke and one involving adults with incomplete spinal cord injury (Behboodi et al., 2022).Their review excluded progressive neurological conditions such as Parkinson's Disease and Amyotrophic Lateral Sclerosis and only included studies in which participants attempted a voluntary motor task; paradigms exploring MI-induced mu modulation were excluded.Although for such reasons the Behboodi review is distinct from ours, it is notable that the authors aimed to map research in both children and adults but ultimately only included adult studies.This reinforces our finding that BCI research in children with neurological motor impairments is sparse.
Whilst we were keen to explore the literature on EEG-based neurofeedback across a broader range of neurological motor conditions, articles focussing on non-stroke diagnoses amounted to only 30 articles in total (23%).By including the literature on stroke our approach allowed us to capture the most extensive experience in the EEG-based neurofeedback literature.Potential reasons for the strong emphasis on adult stroke in the BCI-neurofeedback field are likely to be its high prevalence and a high level of motivation for research participation among individuals with stroke.The indi-viduals enrolled in these studies would be expected to have had normal neurological development prior to their adult-onset stroke and so are likely to have established typical patterns of movement and sensorimotor neuronal circuitry.There is still likely to be considerable heterogeneity, but tasks such as motor imagery may therefore be more straightforward to convey to participants, and data interpretation may be less confounded by variables relating to developmental cortical re-organisation, than in participants with perinatal or childhood-onset disorders.A clear understanding of the effects of neurofeedback in conditions such as stroke is therefore very informative and some of the principles are likely to be relevant when considering neurofeedback interventions in general.Nevertheless, it is important that comprehensive studies are also conducted in children and individuals with other adultonset neurological disorders, all of which are currently underrepresented in research (Fig. 2B).

Challenges of paediatric studies
Challenges of enrolling children in BCI/neurofeedback research include their diverse aetiologies, complex medical needs, cognitive and communication impairments, hyperkinesis, and a lack of paediatric-appropriate equipment (such as smaller commercial headsets) and engaging feedback systems (Jadavji et al., 2023), (see also section 3.8.4).Despite these challenges and the rigorous regulations applicable to research in children, it is important that dedicated paediatric neurophysiological studies are conducted and that the development of new therapies for children is not simply based on extrapolation from adult studies.Accepted models of ''normal" sensorimotor neurophysiology may not be applicable when a brain injury has occurred early in development, as is the case in cerebral palsy and in some genetic conditions (McClelland, 2017).Given the potential for cortical reorganisation following early brain injury, one cannot presume that cortical sensorimotor processing will occur in the expected brain regions, or what typical neurofeedback paradigms might produce in such complex brains (Basu et al., 2010;Eyre et al., 2001;Staudt et al., 2002).It should not be forgotten that this is also a consideration in adults with cerebral palsy.Additionally, neuroplasticity is generally greater during childhood than in adulthood and the underlying neurological substrates of plasticity vary throughout development (Ismail et al., 2017;Tien & Kerschensteiner, 2018).As a result, an insult to the brain in childhood may have different consequences from an insult during adulthood (McClelland & Lin, 2021).Likewise, the effects of an intervention may also have different consequences in childhood compared with adulthood.Indeed, there is evidence that children could be good candidates for EEG-BCI interventions due to their neuroplasticity (Jadavji et al., 2022;Jadavji et al., 2021;Zhang et al., 2019).For example, in one study, 12 children were able to perform mental strategies (MI and goal-oriented) to control a toy car and computer cursor (Jadavji et al., 2021) and in another study, eight children with quadriplegic CP controlled a powered wheelchair via EEG activity (Floreani et al., 2022).(These studies did not meet inclusion criteria for the formal review as they did not involve specific sensorimotor EEG-based neurofeedback, but the findings are pertinent when considering BCI research in children).
The three paediatric studies included in our scoping review (Bobrov et al., 2020;Cincotti et al., 2008;Jadavji et al., 2023) enrolled children with motor impairment diagnoses limited to CP (hemiplegic, spastic diplegic, tetraplegic and quadriplegic), Spinal Muscular Atrophy II and Duchenne Muscular Dystrophy.Dyskinetic/dystonic CP or other childhood-onset movement disorders were not represented, highlighting the need for research to assess the feasibility of EEG-BCI systems in this population.Although a large proportion of the included studies asked participants to perform MI to control the BCI, only two studies actively employed additional cognitive strategies to augment the neurofeedback (e.g., counting numbers), and the impact of these strategies was not explored in detail (Norman et al., 2018;Spychala et al., 2020).However, it is important to acknowledge that participants are likely to try various strategies of their own accord.For example, in two articles, participants were asked to retrospectively describe any strategies they used to control the BCI, and which appeared most effective, although this was not systematically tested (Lioi et al., 2020;Vourvopoulos et al., 2019b).One of these studies commented that their participants reported trying different strategies to control the BCI on different days, which could have influenced variability across sessions in BCI performance and in the resulting behavioural and neural changes (Ros et al., 2020;Vourvopoulos et al., 2019b).
It is acknowledged that sensorimotor paradigms may be more difficult to conduct in patients with childhood-onset motor impairments who have faced limitations in performing motor tasks throughout their lives.Indeed, physiological correlates of motor imagery are reduced in spastic CP (Jongsma et al., 2016), Parkinson's Disease (Tremblay et al., 2008) and in focal hand dystonia (Quartarone et al., 2005).Nevertheless, the observation that MI is used as a strategy by individuals with Parkinson's Disease (Nonnekes et al., 2019) and by children with dystonia (Butchereit et al., 2022), indicates its potential for improving motor performance.However, it is difficult to understand the exact nature of the MI in these individuals and whether it is used similarly across participants.The most effective strategies may vary between individuals and across different age-groups, necessitating a personalised approach (Floreani et al., 2022).Thus, more research is required to better understand the physiological correlates of this phenomenon in this population, along with a comprehensive evaluation of the role of cognitive strategies and their potential to augment EEG-BCI performance in both adults and children with neurological motor impairments.

Limitations and recommendations for future research
The methodology of a scoping review inherently limits researchers to mapping literature without providing analytical interpretation.However, even when researchers aim to conduct a systematic review, the limited breadth of adequately reported evidence hinders critical analysis, often making a scoping review more appropriate.To overcome this barrier to critical analysis for future work in this field, greater clarity of reporting is required regarding the raw neurophysiological data processed by BCI systems, along with greater transparency regarding the actual neurofeedback paradigms.Many commercial systems, such as BCI2000, employ various algorithms and classifiers for signal acquisition, signal processing, and neurofeedback, and the specific processes are unclear.Additionally, studies using Emotiv may lack clarity regarding the exact topography from which EEG signals are detected.
In the interests of transparency and the ability to reproduce published work, we advocate that all online neurofeedback parameters, including feedback update rate, should be reported, along with details of both the online and offline analyses.Where neurofeedback aims to enhance a particular neurophysiological feature (such as mu ERD), the resulting degree of ERD enhancement should be documented, and the effect size reported.Without such details it is difficult to understand which parameters are effective in pro-viding neurofeedback for different patient populations, limiting the depth of insight that can be derived from these studies.Additionally, we highlight the importance of understanding the brainbehaviour relationship and recommend that authors investigate whether reported improvements in motor performance or clinical outcome scores correlate with the enhancement of the neurophysiological parameters being studied (Table 2).
A key finding of this scoping review was the lack of detailed reporting across many studies.However, the exercise identified several studies that adhered to good reporting practice.We define good practice as providing sufficient detail to enable the complete replication of a studys methodology.Examples include, but are not limited to, studies by (Kumari et al., 2022;Pichiorri et al., 2015;Vourvopoulos et al., 2019a;b;Li et al., 2022, andWada et al., 2019).
The CRED-nf checklist recommends guidelines for the design and reporting of clinical and cognitive-behavioural neurofeedback studies (Ros et al., 2020), including details of the online and offline EEG analyses.The potential role of strategies and the consideration of neurofeedback-specific versus non-specific factors is also highlighted (Ros et al., 2020).Although not specifically developed for studies of sensorimotor feedback in neurological motor impairment, the underlying principles and the checklist are equally applicable to studies in this field.
We also recommend that future research into EEG-based neurofeedback training for patients with neurological motor impairments be designed and conducted by multidisciplinary teams.The combined expertise of both engineers and clinical neurophysiologists is invaluable in creating robust study designs.Such collaboration will enhance the reporting standards of EEG neurofeedback parameters and analyses, which in turn is fundamental to investigating the relationship between neurofeedback training-induced EEG changes and observed clinical motor improvements.

Conclusion
There has been a rapid growth of interest in innovative EEGbased neurofeedback techniques.However, this review highlights that much of the work is still exploratory and that greater transparency is required in reporting of EEG parameters.Although several neurofeedback studies have reported improved clinical outcomes in adult stroke patients, very few have provided evidence that these outcomes relate specifically to the EEG-based neurofeedback (Table 2), limiting the conclusions that can be drawn.Furthermore, the reporting of neurophysiological parameters is often insufficient to allow reproducibility of the methodology.Whilst evidence of improved clinical outcome is the ultimate goal for neurofeedback studies, we consider that there first needs to be robust documentation of the neurophysiological methods being applied, both for the online and offline data acquisition and analysis.This requires a comprehensive and systematic approach, evaluating parameters across individuals of different ages and in different patient groups.It is also critical to assess the relationship between changes in brain activity triggered by the neurofeedback and any observed clinical improvement.Finally, there is a huge gap regarding the role of EEG-based neurofeedback in children with movement disorders.Pioneering work on the use of BCIs by children has shown promising results (Floreani et al., 2022;Jadavji et al., 2022;Zhang et al., 2019) but the specific role of sensorimotor neurofeedback in children with dystonia and dystonic/dyskinetic CP is relatively unexplored.Understanding the potential benefits and challenges of implementing EEG-based BCIs in paediatric cohorts holds significant promise for advancing neurorehabilitation strategies tailored to children with dystonia and dystonic/dyskinetic CP.

Fig. 2 .
Fig. 2. (A) Number of included articles published by year.(B) Neurological motor impairment diagnoses of study populations.

Fig. 3 .
Fig. 3. (A) Sensorimotor task parameters.LL=Lower Limb, UL=Upper Limb.(B) Neurofeedback mode(s) employed in studies.As noted in the text, haptic feedback was delivered via multiple methods including robotic devices, vibrotactile and brush stimuli.

E
.Cioffi, A. Hutber, R. Molloy et al.Clinical Neurophysiology 167 (2024) 143-166 Our secondary question related to strategies applied to augment the effects of neurofeedback.
Fig. 1.PRISMA flow chart of screening process for scoping review.BCI=Brain-Computer Interface, CINAHL=Cumulative Index to Nursing and Allied Health Literature, DBS=Deep Brain Stimulation, EEG=Electroencephalography, fMRI=Functional Magnetic Resonance Imaging, SMC=Sensorimotor Cortex, SMR=Sensorimotor Rhythm.À Using a Brain-Machine Interface to Control a Hybrid Upper Limb Exoskeleton During Rehabilitation of Patients with Neurological Conditions À Low Latency Estimation of Motor Intentions to Assist Reaching Movements Along Multiple Sessions in Chronic Stroke Patients: A Feasibility Study À recoveriX: A New BCI-based Technology for Persons with Stroke À Tailoring Brain-Machine Interface Rehabilitation Training Based on Neural Reorganization: Towards Personalized Treatment for Stroke Patients À E. Cioffi, A. Hutber, R. Molloy et al.Clinical Neurophysiology 167 (2024) 143-166 Table 1 (continued) Restoration of Upper Limb Function after Chronic Severe Hemiplegia: A Case Report on the Feasibility of a Brain-Computer Interface-Triggered Functional Electrical Stimulation Therapy Sensorimotor Rhythm-Brain Computer Interface with Audio-Cue, Motor Observation and Multisensory Feedback for Upper-Limb Stroke Rehabilitation: A Controlled Study À Real-Time Control of a Video Game with a Direct Brain-Computer Interface À BCI-Based Rehabilitation on the Stroke in Sequela Stage À Applying a Brain-Computer Interface to Support Motor Imagery Practice in People with Stroke for Upper Limb Recovery: A Feasibility Study À Brain-Machine Interface in Chronic Stroke: Randomized Trial Long-Term Follow-Up (Follow-up study of the above)