The neural basis of human gait has long been a source of great debate with conflicting opinions on the relative importance of central pattern generators, spinal networks and the role of the brain and brainstem in its control ( Richer, Bradford and Ferris, 2024). More recently, the cortical control of gait has been better established. A review of the neural basis of gait in healthy individuals established gait related electrical activity in many different areas of the brain responsible for sensorimotor processing. Electrocortical power fluctuations were observed to occur in the anterior cingulate, posterior parietal, prefrontal, premotor, supplementary motor, occipital and/or sensorimotor cortices during gait ( Richer, Bradford and Ferris, 2024). Further research ( Zhao et al., 2022) has detailed additional gait related electrocortical fluctuations extending to the thalamus and cerebellum. With respect to phases of the gait cycle, increased desynchronization has been observed in the sensorimotor cortex during contralateral limb swing and increased synchronization during ipsilateral heel strike and double limb support period ( Richer, Bradford and Ferris, 2024). The impact of a central neurological pathology on cortical activation patterns during gait are not as well understood. A systematic review investigating brain activity during walking after stroke, for example, highlighted that generalised conclusions were difficult to reach given that pathogenesis of stroke is quite individual. However, some common trends did appear suggesting stroke survivors show greater cortical activation across all areas of the brain. Greater activation of the contra-lesional hemisphere also appears common. This asymmetry of brain activation appears to persist in individuals with severe gait impairments and/or those with larger cortical strokes. It is likely that these asymmetries are the human body’s natural attempt to compensate and that improved symmetry of brain activation during gait may prove a useful biomarker of walking recovery in those with acquired central pathology ( Lim et al., 2021). A similar phenomenon is also observed in spinal cord injury patients with research suggesting that increased beta-band cortical activity may be a neuroplastic compensatory mechanism and as such a biomarker of poor gait function and reduced likelihood recovery in SCI ( Simis et al., 2020). Furthermore, it was found that gait recovery is significantly associated with a relative decrease in high-beta power in the sensorimotor area.

Developments in mobile high density EEG technology have greatly aided our understanding of top-down control of gait and this has opened up the potential for movement-based gait classification systems using EEG to be developed. Seen as an important step in the development of BCI RAGTs, studies to date have demonstrated accuracy in classifying imagined movement and movement intention based on cortical activity. Studies have successfully classified kinaesthetic motor imagery of walking vs idling ( Do et al., 2013), as well as standing resting vs intention to move ( Sburlea, Montesano and Minguez, 2017), with the latter correctly classifying 66.5% of trials in healthy subjects, and 63.3% in stroke patients. Zhang et al. (2017) classified gait based on intention to stop, walk, turn right and turn left in one healthy subject and one spinal cord injury subject achieving a classification accuracy ranging from 68.4% to 74.5% in a four-class classification system. Movement classification during robotic assisted gait has also proven challenging. One study by Liu et al. (2017) used motor intention, indicated by imagery based sensorimotor rhythms (SMR) and movement related cortical potentials (MRCP) to develop a control system for a lower limb exoskeleton based. While an accuracy of 80% ± 5% was achieved with the SMR-based method, classification accuracy using the MRCP-based method peaked at 69% ± 9%. This reduced accuracy in classification using MRCP reflects the challenges of classifying movement intention during activity. It is worth noting that the aforementioned studies focused on classifying imagined movement and movement intention, rather than movement itself. This has important implications for data processing as pre-movement cortical activity has significantly less artifact. As such, decoding of the EEG activity is an easier endeavour. However, some progress has been made with research successfully classifying intention during actual movement i.e. the classification of actual movement vs intended movement. Ortiz et al. (2017) classified stopping gait based on intention in both offline and pseudo-online conditions while Hasan et al. (2021) successfully recognised intention to accelerate during gait.

The classification of movement execution (rather than the intention or imagination of movement) including gait, presents a challenge, largely limited by demands in data acquisition and high levels of movement artifact. Lower limb movement also presents a unique challenge due to the medial locations of the lower limb motor areas on the brain and with that, difficulties in differentiating between hemispheric signals ( Wieser et al., 2010). This is in contrast to the upper limbs where movement execution as part of a closed loop BCI system has been achieved ( Khan et al., 2023). However, the advent of wireless devices, improved artifact recognition and removal strategies have advanced the field further ( Zhao et al., 2021; Gorjan et al., 2022), allowing systems to classify overground walking ( Severens et al., 2014) and differences in slow gait speeds ( Lisi and Morimoto, 2015) and more recently, to classify walking from complex movements such as ascending and descending stairs and ramps ( Zafar et al., 2024). Steady-state visually evoked potentials ( Kwak, Müller and Lee, 2017) have been used to classify walking from standing with high accuracy rates but the subject dependent classifiers used required long training times. Preliminary data provide proof of concept that EEG-based neural networks can classify double stance as well as right and left single leg stance during the gait cycle ( Herbert and Munz, 2020; Tortora et al., 2023) during overground walking. A noteworthy finding by Tortora et al. (2023) was that overground classifications systems could not successfully classify the phases of gait during RAGT. Further research is required to advance the pool of research using EEG to classify free over ground walking (and its phases) and gait during RAGT. This is an important step in the development of a closed loop BCI controlled RAGT device.

Significant developments have been achieved in the field of BCI-based upper limb rehabilitation ( Cervera et al., 2018; Al-Quraishi et al., 2018) including the development of closed loop BCI controlled upper limb exoskeleton devices ( Tang et al., 2016; Bhagat et al., 2016). Despite these technological advancements, studies and research successfully closing the loop between the human and a RAGT device are limited. The majority of successfully created EEG driven closed loop brain-computer interface (BCI) systems for RAGT control have primarily relied on motor imagery ( Al-Quraishi et al., 2018; Padfield et al., 2019) or motor intention ( Jin et al., 2024; Lee et al., 2017). Despite providing real time feedback, these systems do not classify actual motor execution within the BCI system. To the best of the authors knowledge the development of a closed loop overground exoskeleton control system using an EEG based BCI which utilises online classification of actual movement execution remains elusive. This scoping review aims to collate and learn from the current state-of-the art in this field to highlight knowledge gaps and direct future research. Findings will have important implications for the field with respect to the development of online gait classification systems as well as closed loop BCI based control systems for robotic assisted gait trainers. As the authors will examine the broad body of literature addressing gait classification, a scoping review is a valid approach ( Munn et al., 2018). Results will provide a clear indication of the volume of studies as well as an overview of the methodologies employed in the field. Furthermore, findings from this scoping review will also indicate if a systematic review is indicated.

The primary aim of this scoping review is to collate existing studies addressing the classification of gait (and phases of gait) using EEG, summarise the research in the field and identify gaps and opportunities for further research studies. Specific objectives include: 1.

To systematically identify and report all studies classifying and/or its relevant phases using EEG.

The search strategy will be finalised by the primary research team (CR; OL) in conjunction with a liaison librarian. It will also be guided by the structured tool for peer review of electronic literature search strategies (PRESS) checklist ( McGowan et al., 2016). Controlled vocabulary terms (e.g., MeSH terms) and appropriate synonyms for controlled vocabulary terms, truncation and different spelling conventions (e.g., “categorisation” versus “categorization”) will be utilised during the search. Search strings related to the key concepts of “EEG”, “gait” and “classification” will be developed and combined using Boolean operators e.g. “EEG” AND “gait” AND “classification”. A copy of the indicative search strategy i.e. search strings and Boolean operators, is available at the online repository associated with this protocol (see extended data). Searches will be conducted across SCOPUS, EMBASE, PubMed and Web of Science scientific databases. Identified studies will be exported to COVIDENCE software for screening and data extraction stages. Citation management will be completed using EndNote software.

Identified studies will be screened independently by two researchers (CR; CW) against eligibility criteria (table 1; see extended data), firstly by title and abstract and then by full text review. Where disagreements arise, they will be discussed with a third reviewer (OL) to gain consensus. No publication date or study methodology restrictions will be applied. Only full text articles published in English will be considered. A log of excluded studies and reasons for exclusion will be kept. A PRISMA flow diagram will be included. No quality appraisal of evidence will be conducted as per the guidelines for scoping reviews ( Peters et al., 2020; Tricco et al., 2018; Munn et al., 2018).

Exclusion of all other brain imaging modalities other than EEG is justified as EEG is the most reliable, mobile, and cost-effective means of tracking neural activity in ambulatory states ( Richer, Bradford and Ferris, 2024) in comparison to other means (e.g. fMRI, PET, fNIRS and SPECT). In addition, these alternative technologies are not as well suited to study brain dynamics during walking ( Bakker et al., 2007). The review is focussed on data related to the healthy state as well as those with central neurological pathology to identify normal top-down control of gait and explore whether meaningful activity is retained in brain injury. The inclusion of those with central neurological pathology is particularly relevant as this is the cohort most often utilising robotic assisted gait training devices. For the purpose of this review, central neurological pathology will be considered as any condition affecting the central nervous system.

A data charting proforma will be developed to extract the pertinent data from each included study. Items for extraction will include but not limited to: •

EEG data capture methodology i.e. technology used, number of electrodes, electrode sites used.

Narrative and tabular synthesis will be employed to summarise the current state of play in the field.

The scoping review is pre-registered with the Open Science Framework (OSF) ( Ryan et al., 2025) and results will be submitted for peer reviewed publication and disseminated at relevant conferences and professional networks.

Ethical approval is not required for this scoping review as only published literature is being used.

At the time of protocol submission, the scoping review is ongoing. A search strategy has been developed, and initial title/abstract screening is ready to begin.