Participants were volunteers by self-selection recruited through poster campaigns in the university, local retirement groups and Irish Heart Foundation community stroke support centers. Self-selection indicates that individuals chose to participate on their own initiative, without being directly invited or selected by the researchers. Participants were included if they were over 18 years of age, had no known neurological condition (other than stroke) and could stand independently (permitting the use of mobility aids) from a regular height surface. All stroke types and locations were considered eligible for inclusion, as were participants with any number of previous stroke occurrences. Participants were excluded if they presented with a physical impairment that limited data collection (modified Rankin Scale > 4) (Pożarowszczyk et al., 2023) or a cognitive impairment that limited their ability to provide informed consent (COGTEL score < 15, subject to further investigation) (Ihle et al., 2018). A COGTEL score below 15 was considered indicative of severe cognitive impairment and prompted further assessment to determine capacity to provide informed consent. For participants scoring below this threshold, additional capacity assessments were conducted in line with ethical guidelines. Individuals who demonstrated sufficient understanding, appreciation, reasoning, and ability to express a consistent choice, were deemed capable of providing informed consent. Informed and written consent was obtained from each subject prior to participation. The Human Research Ethics Committee at University College Dublin, the National University of Ireland approved the experimental protocols (LS-22-19-Lennon), conducted in accordance with the Declaration of Helsinki.

The comparator group consisted of 11 age and sex-matched healthy adults. Healthy participants had no known neurological conditions or musculoskeletal impairments that would impact data collection. These participants were selected from a larger dataset collected as part of a previous study in healthy participants (McDonald et al., 2025), ensuring sex matching and the closest possible age match available. An identical data collection protocol was employed in both studies.

On arrival to the motion analysis laboratory each participant’s height, weight Functional Ambulation Category (FAC) (Mehrholz et al., 2007) and modified Rankin Scale (mRS) for neurological disability (Pożarowszczyk et al., 2023) were recorded by a research physiotherapist. Participants were positioned on an armless, backless seat, adjusted in height to the knee joint height of each participant with the knee at 90 degrees flexion. Knee joint angle estimation was performed via visual observation by a physiotherapist. Initially a two-minute resting EEG trial was recorded. This was followed by a target of five sets of three sit-to-stand and stand-to-sit transitions at a self-selected pace, pausing for approximately 5 s between each transition. In cases where participants were unable to complete all trials due to fatigue, data collection concluded at that point. Participants were instructed to stand up and sit down in a natural way with the minimal use of upper limb assistance. Participants were permitted to use any mobility aids routinely employed in daily life to perform sit-to-stand transfers. To minimize movement artifact in EEG recordings, participants were asked to avoid excessive blinking, teeth grinding and jaw clenching. To ensure familiarity with the task and consistency across participants they performed a practice trial of 3 sit-to-stand repetitions with verbal correction or demonstrations provided where required.

Electroencephalography was recorded at a sampling frequency of 250 Hz using a wireless system of eight bipolar channels (FlexEEG, NeuroCONCISE, UK) over the sensorimotor cortex [C6, C4, C2, Cz, FCz, C1, C3, and C5 locations defined by the international 10/20 system of electrode placement (Klem et al., 1999)]. Electrode Afz was used as a reference. Conductive gel was used to maintain low electrode impedance. Impedance measurement was not conducted, as it was not supported by the recording system used. Data were recorded and saved using Matlab software (Mathworks, USA).

Electrical muscle activity was measured using eight wireless EMG electrodes with a 2000-Hz sampling rate (Delsys, Trigno, USA). Electrodes were placed bilaterally on vastus lateralis (VL), biceps femoris (BF), tibialis anterior (TA), and gastrocnemius (GAS) muscle groups according to SENIEM guidelines (Stegeman and Hermens, 2007).

A 3D motion capture camera system was used to record the kinematic joint angles (Codamotion 3D system, UK). Four cameras captured the movement by active infrared retroreflective wearable marker clusters, attached to both sides of the body, on the pelvis, upper leg, lower leg, calcaneus, and fifth metatarsal bases, and one marker cluster was placed in the center of the chest. In addition, reflective markers were attached bilaterally to anatomical landmarks (anterior superior iliac spine, lateral femoral epicondyle, and medial femoral epicondyle; lateral and medial malleoli) and the right acromioclavicular joint, to create a three-dimensional whole-body segment model. Kinematic data were recorded at a 100-Hz sampling rate. Hip, knee and ankle joint angles were estimated from spatial position data by the Codamotion ODIN software platform (ODIN, Codamotion, UK).

All EEG, EMG and ODIN data were synchronized. EMG data was integrated digitally with the kinematic data via Coda-Hub. synchronization of the FlexEEG system with Coda-Hub was achieved using a Delsys Trigger Module (Delsys, USA) and a LabJack U12 USB DAQ module (LabJack, USA). When kinematic recording began, a pulse from the trigger module was sent to Coda-Hub to start EMG recording, and a second pulse was sent from Coda-Hub to Matlab (Mathworks, USA) through the LabJack module via USB.

The sit-to-stand movement was divided into two broad dynamic phases based on kinematic joint angle data (collapsing the four known phases of sit-to-stand to two). Phase one, the flexion phase, began with the initial detection of hip flexion movement, which was manually identified by visual inspection of the hip joint angle time series as the first increase in flexion, and ended at the point of maximum hip flexion (combining upper body forward momentum and pelvic de-weighting phases). Phase two, the extension phase, began at maximum hip flexion and ended at maximum hip extension (combining the extension and stabilization phases). Pre and post movement activity were recorded in quiet sitting, 0.5 s prior to the beginning of the flexion phase, and in upright standing 0.5 s after termination of the extension phase.

EMG data were bandpass filtered between 20 to 250 Hz (zero-phase 2nd -order Chebyshev). A 50 Hz notch filter was used to remove power-line interference. The offset for each channel was removed by subtracting the trial signal average (Yin et al., 2020; Konrad, 2005).

Raw EEG were bandpass filtered, 0.1 to 60 Hz with a zero-phase 2nd -order Chebyshev. A notch filter centered at 50 Hz with a 2 Hz bandwidth was used to remove power line interference. EEG data were then submitted to an automated artifact rejection procedure, Artificial Space Removal (ASR) using the EEGLAB toolbox (clean_artifacts) in Matlab (Mathworks, USA) to remove artifacts related to low frequency drifts, noisy channels and short-time bursts. The input arguments used were: ‘BurstCriterion’ configured as ‘20’; ‘WindowCriterion’ and ‘ChannelCriterion’, configured as ‘off’ to avoid deleting data windows and channels; and ‘BurstCriterionRefMaxBadChns’ to pass the calibration data. For each participant, ASR was calibrated from the EEG data collected during the 2-min resting state condition and applied to all the movement trials performed, in line with previous studies (Tortora et al., 2020; Tortora et al., 2023; Artoni et al., 2017). The resting period was used instead of the more common task-specific period as the short trial duration and movement artifact limited calibration. Segments with absolute voltages above 100 μV were removed. Pre-processed EMG and EEG data underwent manual visual inspection. EMG and EEG channels were inspected in the time frequency domain and those with high level of artifact that were deemed not to contain meaningful information were excluded from analysis (Supplementary Table S1). Epochs that lacked characteristic EMG/EEG features such as well-defined activation/oscillatory patterns and consistent baseline activity were excluded. Signals with abrupt, irregular deflections, non-physiological flat-lining, high-frequency noise inconsistent with neural activity were flagged as likely artifacts and discarded. This manual inspection process ensured the retention of only physiologically plausible bio-signal data for further analysis.

Data sets were individually screened for quality of EMG and EEG signals. In the case where one of the two inputs was not viable or deemed not to be of sufficient quality, that entire sit-to-stand trial was discarded. Participants were excluded from analysis if they did not reach a minimum of 6 viable sit-to-stand transitions.

ERSP was calculated over the movement phases, to measure the modulation of the EEG activity in the frequency domain across the two sit-to-stand phases delineated. Time-frequency decomposition was performed on each channel and trial individually, followed by averaging across trials at each frequency. For normalization, the mean power across the entire task duration was computed for each frequency and used as a baseline. This method is well-suited for dynamic motor tasks, as it emphasizes time-varying spectral modulations relative to the overall task-related activity (Tortora et al., 2023; Gwin et al., 2010). This analysis was performed using the EEGLAB toolbox (newtimef), providing the epoch times at which the phases occur in each trial using the argument ‘timewarp’, and setting the ‘trialbase’ argument to ‘full’ to perform single trial normalization. These settings were used to align the timepoints of each phase for the different trials. Power spectral density (PSD) was computed from the ERSP for each phase of sit-to-stand in delta (0.5-4 Hz), theta (4–8 Hz), alpha (8–14 Hz), beta (14–30 Hz) and gamma (30–40 Hz) power bands, normalized to their average power. Topographic maps were generated using the EEGLAB toolbox (topoplot) applying a mask for the surrounding canals (T7, T8, Fz, Pz, P4, P3, F3, and F4).

The EMG envelope was obtained by estimating the root-mean-square (RMS) amplitude using 250 ms sliding windows, overlapping by 50 ms. The EMG signal corresponding to each phase was timewarped using linear interpolation (0.1% resolution), to align the timepoints for averaging the different trials. RMS envelopes were normalized on an individual basis to the maximum value in amplitude found within each set (of three or five trial repetitions) to reduce inter-participant variability.

To analyze corticomuscular coherence (CMC), we examined the frequency-domain coupling between the EEG electrode Cz, which is located over primary motor cortical area, and each of the eight recorded leg muscles. We chose this centrally located sensor because it is widely employed to assess CMC during gait (Jensen et al., 2019; Gennaro and de Bruin, 2020; da Silva Costa et al., 2024) and it was found to show maximal beta-band CMC amplitudes (Spedden et al., 2019). On account of individual trial durations of the sit-to-stand phases being too short to allow for reliable spectral analysis, EMG and EEG data from all trials were concatenated separately for each movement phase by stacking all repetitions of the corresponding phase across trials. This resulted in a continuous dataset per phase that aggregated neural and muscular activity from multiple instances of the same movement segment. This approach was used, as an alternative to averaging, to improve spectral resolution (Johnson and Shinohara, 2012; Gangwani et al., 2023). A low pass filter, 100 Hz with a zero-phase 2nd -order Chebyshev, was then applied on both EMG and EEG data. EMG data was full-wave rectified prior to coherence analysis. The Matlab coherence function, with a Hamming window of 512 and overlapping of 192 samples, was used for CMC calculation. The significance of the coherence was assessed through surrogate data analysis by generating 75 surrogate coherence spectra [consistent with the companion paper on healthy controls (McDonald et al., 2025)] for each participant. These surrogates were created from random signals of the same length as the original signals to preserve relevant properties (Faes et al., 2004). A cumulative density distribution with the surrogate coherence peaks were built to compare against the real coherence peaks for statistical significance (p < 0.05) (Piitulainen et al., 2015).

Data are reported separately for stroke participants and healthy controls. For stroke participants, we refer to the stroke side as the lesioned brain hemisphere, i.e., the left stroke side refers to right limb impairment, and the right stroke side refers to left limb impairment.

Stroke participants’ hemiplegic side, or affected lower limb, were compared with the same lower limb of their matched healthy control participant during comparative analyses performed, detailed below.

Comparative analyses were performed between the ERSP average of healthy controls and stroke survivors for each of the individual movement phase of sit-to-stand in the alpha and beta frequency bands and the Cz, C1 and C3 EEG channels. These electrodes were selected because a previous sit-to-stand investigation in healthy controls indicated these locations were likely to show the largest changes in activity (McDonald et al., 2025). The Shapiro–wilk test was conducted to assess normality and subsequently select appropriate statistical analysis methods. The independent T-test was performed for data that was found to be normally distributed, and the Wilcoxon rank sum test was performed in cases where data was found to be not normally distributed. The significance level was set at p < 0.05.

Statistical comparison of inter-hemisphere ERSP (C1 and C3 electrode compared to C2 and C4 electrode) of both groups was completed using the paired T-test for data that was found to be normally distributed, and the Wilcoxon signed-rank test in cases where data was found to be not normally distributed. In the case of right hemisphere strokes the C2 and C4 electrodes were compared to the C1 and C3 electrodes, while for left hemisphere strokes the C1 and C3 electrodes were compared to the C2 and C4 electrodes. Further subgroup analysis was performed using the same methods, dividing stroke survivors by their level of impairment as indicated by the mRS. Correction for multiple comparisons was not applied, as the statistical analysis focused on a small number of planned comparisons, remained exploratory in nature, and was intended to inform future hypothesis-driven research.

Post-stroke participants took significantly longer to complete a sit-to-stand transfer (2.45 s) compared to healthy controls (2.25 s) (p = 0.038). There were no significant differences however in the temporal parameters of the movement phases between stroke and healthy individuals, when considered as two dynamic movement phases, the sit-to-stand transfer comprised 33% as the flexion phase and 67% as the extension phase in both groups. In the early flexion phase the hips and ankles were observed to begin to flex to initiate the movement. Hip flexion then continued and the hip was unloaded from the chair with some continuing knee and ankle flexion evident. In the extension phase the hips and knees were observed to begin to extend, coinciding with ankle plantar flexion. Toward the end of the extension phase as the body sought stability in upright standing, small joint adjustments were seen largely at the ankle joint. This is illustrated in Figure 1.

Figure 2 graphically depicts the average lower-limb muscle forces (RMS) of the four lower-limb muscles across sit-to-stand movement phases. Tibilalis anterior was activated first in the flexion phase, closely followed by vastus lateralis and biceps femoris. All three muscles reached peak EMG activity at seat off, the moment when the hip is unloaded from the chair and the hip joint angle trajectory changes from flexion to extension, i.e., the transition point between the flexion and extension phases. Across these three muscle groups, activity then reduced until upright standing was achieved, when activity plateaued marginally above (~0.5 RMS) baseline sitting activity. The gastrocnemius was most active in the extension phase and remained active in upright standing. The muscle that reached the highest level of normalized RMS activity was the vastus lateralis. This pattern was largely consistent with that observed in healthy controls however, the stroke group exhibited peak EMG activity at later points during the transfer. Delays were observed in the bilateral vastus lateralis (left: 0.32 s and right: 0.16 s) and in the right biceps femoris (0.425 s), relative to the healthy controls.

Figure 2 shows muscle activation (as normalized RMS EMG) across phases of sit-to-stand averaged across all left hemisphere stroke participants compared to their matched healthy controls.

A statistically significant difference (p < 0.05) in ERSP was not found at group level (seven left hemisphere strokes verses seven matched healthy controls, Table 2) however some important differences were observed in grand average ERSP plots and between individual participants. As presented in Figure 3 using single participant ERSP, healthy controls exhibited consistent and well-defined desynchronization (ERD) in the beta (15–30 Hz) and alpha (8–12 Hz) bands during movement preparation and execution, followed by post-movement synchronization (ERS). Stroke Survivors however demonstrated inconsistent synchronization-desynchronization patterns across participants. Beta and alpha band ERD were delayed (by 0–2.2 s), reduced in magnitude (~0.2 dB) or absent in this group, particularly in the affected hemisphere (Figure 3). Healthy controls displayed bilateral but slightly dominant ERD/ERS in one hemisphere. In stroke survivors, asymmetric activity was observed with weaker ERD in the affected hemisphere. Healthy controls further exhibited a rapid return to baseline or increased beta ERS in the extension phase whereas stroke survivors showed prolonged and/or reduced ERS, in the affected hemisphere (Figure 3). Results were consistent in both right hemisphere and left hemisphere stroke. A similar pattern of synchronization/ de-synchronization is seen at population level for both groups in Figure 4 where a grand average of ERSP spectrogram across all seven left hemisphere strokes verses seven matched healthy control participants is presented.

Table 2 depicts the group level comparison of seven left hemisphere strokes verses seven matched healthy controls, across alpha and beta frequency bands, for selected left hemisphere EEG channels. A significant difference was observed between the left midline C1 electrode during the standing phase. A grand average for right-hemisphere and bilateral-hemisphere stroke participants was not computed, as the small sample size precluded reliable group-level analysis.

Topographical mapping of the ERSP spectrogram identified the cortical areas that are most active during the phases of sit-to-stand across the population groups. Results include data in pre-movement sitting and in standing on termination of the sit-to-stand transfer and, as typical for EEG, is described in terms of rhythmic activity divided into the five frequency bands, delta (0.5- 4 Hz), theta (4–8 Hz), alpha (8–14 Hz), beta (14–30 Hz) and gamma (>30 Hz). A grand average of all left hemisphere strokes (7 participants) are presented alongside their matched healthy control participants (Figure 5). Results show that healthy control participants display bilateral but slightly dominant left hemisphere activity during all phases of sit-to-stand. Stroke survivors however show asymmetric cortical activation with substantially increased activation in the right hemisphere (the contralesional hemisphere). This is pronounced during sitting pre-movement and the flexion phase, although statistical comparison of inter-hemisphere ERSP (affected hemisphere compered to unaffected hemisphere) of the stroke group revealed no statistically significant differences (p < 0.05). Although not statistically significant larger interhemispheric discrepancies were evident in the Topoplots of stroke survivors with a higher level of impairment as indicated by the mRS.

CMC between the central Cz electrode and individual lower limb muscles (Vastus Lateralis, Biceps Femoris, Tibialis Anterior, Gastrocnemius Lateralis) are depicted for all 10 participants averaged across all sit-to-stand trials (Figure 6). Color plots depict higher coherence in the latter half of the flexion phase and in the extension phase across all muscle groups and all participants consistent with ERSP and EMG findings. However, as detailed in the methods section and presented visually in Figure 6, when CMC is plotted for each phase of sit-to-stand across frequency bands, alongside 75 randomly generated surrogate coherence spectra, the findings do not consistently exceed the chance of identifying corticomuscular coherence from randomly generated data. CMC (p < 0.05) was observed during the extension phase between the Cz electrode and vastus lateralis in three participants, biceps femoris in two participants and gastrocnemius in two participants. For the participant presented below (P10) statistically significant coherence (p < 0.05) was observed in the extension phase, between the Cz electrode and the bilateral gastrocnemius.

As noted, kinematic and EMG data are consistent with those previously reported in the existing literature (Hsu et al., 2019; Hu et al., 2013; Hsu et al., 2023). Firstly, the overall movement time of the individuals with hemiparesis is longer (Hsu et al., 2019; Hu et al., 2013), noted in this study to be almost 10% longer than healthy controls (Figure 1). Secondly, while there are no significant differences in the temporal parameters of muscle activity between participants with stroke and healthy individuals, as observed in Figure 2, delayed peak activation of lower-limb muscles is evident in individuals with hemiparesis, particularly in the unaffected side (Hsu et al., 2019; Hsu et al., 2023), meaning post-stroke patients favor their unaffected limb (Hsu et al., 2019; Hsu et al., 2023), This explanation is supported by literature that observes larger medio-lateral center of pressure displacement and altered weight distribution in the lower-limbs post stroke (Cheng et al., 1998; Boukadida et al., 2015) In this current study (Figure 2), a delay in contraction of Biceps Femoris and to a lesser extent Vastus Lateralis is observed post-stroke when compared to healthy controls, suggesting impairment of important anticipatory postural adjustment that is central in origin (Hsu et al., 2019). EEG data can for the first time support this hypothesis showing delayed alpha and beta ERD, particularly in the ipsilesional hemisphere. This finding is novel as it has not been previously reported in the literature.

EMG findings, confirming what is already known regarding muscle activation patterns, are clinically meaningful as asymmetrical muscle activity during sit-to-stand post-stroke is associated with a higher risk of falls (Figure 2) (Cheng et al., 1998). Moreover, individuals who demonstrate substantial motor recovery following stroke exhibit increased peak muscle activation across all synergies, with the most pronounced changes observed in the first synergy, corresponding to the momentum transfer phase (Wang et al., 2022). Notably, the timing of peak activation shifts earlier in the movement sequence (Wang et al., 2022; Yang et al., 2019). Correct timing of peak activity in the quadriceps and hamstrings muscle groups during the flexion phase of sit-to-stand could be considered as a potential rehabilitation target post-stroke based on the findings reported here and in the wider literature (Wang et al., 2022; Yang et al., 2019).

ERSP analysis in the current study reveals differences in cortical activity during the execution of a sit-to-stand transfer between healthy controls and stroke survivors, particularly in the power modulation of alpha and beta frequency bands (Figures 3, 4). Healthy Controls for example were observed to have well-defined ERD in the alpha (8–12 Hz) and beta (15–30 Hz) bands during movement preparation and early execution, followed by post-movement ERS. This power modulation pattern is typical of voluntary movements (Toro et al., 1994; Fry et al., 2016), recognized as reflecting efficient motor control and sensorimotor processing (Delcamp et al., 2024). It has previously been observed in other whole body movements such as gait (Borhanazad et al., 2024). This research group recently reported the same finding during the complex task of sit-to-stand for the first time in healthy individuals. These advances now provide an important baseline from which abnormal cortical activity that may be driving the known kinematic and electromyographical changes post stroke can be critically examined (McDonald et al., 2025). ERSP modulation during sit-to-stand execution in stroke survivors was not consistent across participants. Data evidenced that stroke survivors exhibited weaker ERD during the movement, particularly in the affected hemisphere, indicating impaired cortical activation (Delcamp et al., 2024; Tangwiriyasakul et al., 2014). Comparison of these findings to the wider literature in stroke is limited as the majority of post-stroke ERSP studies focus on the upper limb (Delcamp et al., 2024; Tangwiriyasakul et al., 2014). However, a limited number of studies analyzing post-stroke gait have shown weaker alpha ERD compared to healthy controls (Mayor-Torres et al., 2022; Ko et al., 2021) and furthermore that increased impairment correlated with decreased alpha ERD during the stance phase of gait (Ko et al., 2021).

This study identified asymmetrical cortical activation in stroke survivors during pre-movement and sit-to-stand phases, not previously reported in the literature (Figures 3–5). Despite the stroke survivors included in this study having low levels of disability (mRS < 3) and functional impairment, nonetheless they demonstrated increased cortical activation in their unaffected hemisphere compared to their affected hemisphere, as displayed by increased power spectral density (Figure 5) and asymmetrical ERSP plots (Figure 3). While the degree of asymmetry varied across individuals, when considered on a case-by-case bases, individuals with higher levels of impairment (mRS 2/3) appear to have the largest asymmetries. In addition, all but one stroke participant were in the chronic stage of stroke recovery (>6 months post-stroke), while one participant was in the late-subacute phase (>12 weeks post-stroke) (Li et al., 2024). It has been shown in comparator gait studies that ERD and ERS modulation vary with the level of gait impairment and with stroke chronicity, with less impairment and more chronic stroke show more typical modulation patterns (Ko et al., 2021; Oliveira et al., 2018; Carino-Escobar et al., 2019). Furthermore, upper limb research has shown that higher Brain Symmetry Index values, reflecting more power asymmetry over the hemispheres, predict more motor impairment 6 months after stroke (Saes et al., 2021). As the first study to describe the presence of hemispheric asymmetries during sit-to-stand, these findings now warrant further examination. This study included participants only in the chronic phase of stroke, many of whom had low level of impairments and some of whom had bilateral stroke. Examination in the acute and subacute phases of stroke is now warranted alongside stratification of participants by impairment levels.

This research group recently reported CMC during sit-to-stand in healthy controls was inconsistent across participants and statistically significant coherence was observed in only a subset of healthy participants (McDonald et al., 2025). The lack of consistency in findings was attributed to several challenges associated with measuring CMC during dynamic whole-body movements. CMC has been shown to be lower in posturally focused, position-control tasks than in force-control tasks (Poortvliet et al., 2015). Studies examining CMC during isokinetic muscle contractions, such as those occurring in sit-to-stand transitions, are limited. The majority of published research focuses on isometric contractions which typically produce higher levels of coherence (Liu et al., 2019). Therefore, it was not surprising that inconsistent CMC patterns were observed in post-stroke participants. A notable finding warranting further investigation was that, in the subset of post-stroke participants exhibiting significant coherence, this coherence was predominantly observed in the quadriceps and hamstring muscle groups during the extension phase of the sit-to-stand movement. This contrasts with prior observations in healthy control participants, wherein significant CMC was more frequently detected during the flexion phase of the movement in the same muscle groups (McDonald et al., 2025). This may represent the delayed peak EMG activity of the quadriceps and hamstrings muscles during the flexion phase of sit-to-stand observed in the stroke participants. As a result of challenges assessing CMC during this complex movement no firm conclusions can be drawn from these findings given the intersubject variations in both healthy and stroke study participants.

Reduced ERD in stroke survivors evident in this study during sit-to-stand suggests impaired motor cortex activity which could prove a useful biomarker for recovery and/or a target for rehabilitation strategies such as neurofeedback or brain-computer interface technologies (BCIs). Neural (EEG) interfacing technology could offer new strategies for lower-limb robotic rehabilitation by promoting more active engagement in movement intent and/or neurophysiological feedback. In the upper limb, Human Machine Interface robotic therapy is proven to enhance motor outcomes in neurological populations such as stroke (Cervera et al., 2018). Furthermore, extensive literature detail approaches for decoding upper-limb kinematics or muscle activity using cortical recordings toward brain machine interface applications (Camargo-Vargas et al., 2021). Similar approaches to the lower-limb remain scarce where disappointingly, the same technology has failed to realize similar outcomes for lower-limb motor recovery. No intent-controlled Robotic Assisted Gait Training has been identified in clinical studies (Sarasola-Sanz et al., 2017) or in the stroke rehabilitation sector (Lennon et al., 2020). Limited research examining robotic sit-to-stand assistance exists and has, in the main, focussed on human intention (Li et al., 2021; Romano et al., 2017). This study offers important new information for the development of responsive BCI sit-to-stand rehabilitation devices for this population, highlighting potential difficulties in variable activity both between participants and within participants as they make functional improvements and suggests the need to train BCI systems at an individual level.

Results of this study must be interpreted with a degree of caution. The relatively small size of the stroke group and healthy counterparts limits statistical power and the ability to draw strong conclusions. Correction for multiple comparisons was not applied as the analysis was limited to 10 focused comparisons, aimed at guiding future hypothesis-driven research. However, this approach increases the risk of Type I errors due to the relatively large number of statistical tests conducted. Therefore, the results should be interpreted with appropriate caution. In effect, heterogeneity of findings across individuals and kinematic stages of sit-to-stand indicate a fundamental challenge for development of future brain-computer interface devices that can accurately address sit-to-stand movements.

The stroke group exhibited a significantly higher BMI compared to the control group (p = 0.037). Elevated BMI can attenuate both EMG and EEG signals due to increased adipose tissue, which may affect electrode impedance on the scalp and over muscles (Farina et al., 2002). However, we do not believe this difference substantially impacted the current findings. Only 2 out of 10 stroke participants were classified as having severe obesity (defined as BMI ≥ 35.0) (World Health Organization, 2000) indicating a level of adiposity that could meaningfully affect signal quality. As such, a subgroup analysis was not performed.

Future research should include a larger sample of stroke participants encompassing a broader range of impairment levels and varying times post-stroke to enhance generalisability. Additionally, while increasing the number of repetitions is challenging in this population due to fatigue, distributing repetitions across multiple testing sessions over several days could improve statistical power to detect meaningful effects.

A limitation of the peak-within-task normalization approach employed is that, while it reduces inter-participant variability and facilitates within-muscle temporal analyses and characterization of activation patterns, it does not permit valid inter-limb comparisons. The ideal approach, and current best practice, is to normalize to a maximum voluntary contraction (MVC) matched to the task (Besomi et al., 2019); however, this is challenging to implement for sit-to-stand movements due to their dynamic nature and the substantial changes in muscle length that occur throughout the movement. This approach should be considered in future studies.