Noisy galvanic vestibular stimulation improves postural stability under virtual reality perturbation by enhancing vestibular processing and multisensory integration

Abstract

Background

Noisy galvanic vestibular stimulation (nGVS) is a non-invasive neuromodulation technique that enhances vestibular inputs via stochastic resonance, thereby improving postural stability. However, it remains unknown whether nGVS can ameliorate virtual reality (VR)-induced visual-vestibular conflict, which arises from an incongruence between moving visual cues and the absence of corresponding vestibular motion signals during quiet standing. The underlying neural-muscular mechanisms through which nGVS may modulate such conflict warrant clarification. This study aimed to investigate the effects of nGVS on postural stability, muscle activation, and cortical activation in healthy individuals during upright stance under different visual conditions.

Methods

Thirty healthy young adults participated in this study. Six standing trials involving 3 visual conditions (eyes-open, eyes-closed, VR) and 2 vestibular stimulation conditions (nGVS and sham stimulation) were randomly allocated to each participant. nGVS was delivered via electrodes placed over bilateral mastoid processes. A synchronized multimodal assessment was conducted, using force plates to measure the center of pressure (COP), surface electromyography to record muscle activity, and functional near-infrared spectroscopy to monitor hemodynamic responses in the frontoparietal cortex.

Results

Compared to sham stimulation, nGVS significantly improved postural stability and hip muscle co-activation under VR perturbation, as evidenced by reduced COP-related parameters (ps < 0.05), decreased muscle activity in the bilateral rectus femoris (left: p = 0.003; right: p = 0.018), left biceps femoris (p = 0.020), and reduced co-contraction indices (left: p = 0.026; right: p = 0.027). In the VR condition, nGVS concurrently leaded to significantly lower dorsolateral prefrontal cortex activation (left: p = 0.038; right: p = 0.047) and higher supramarginal gyrus activation (left: p = 0.048; right: p = 0.029).

Conclusions

nGVS effectively mitigated VR-induced sensory conflict and enhanced postural stability. The underlying mechanism may involve the induction of stochastic resonance to optimize vestibular processing and multisensory integration, thereby reducing reliance on the hip strategy.

Trial registration

Chinese Clinical Trial Registry (ChiCTR# 2300078910); Date of registration: December 20th, 2023.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12984-026-01944-5.

Introduction

Postural stability, defined as the ability to sustain equilibrium by controlling the center of mass within the base of support, is commonly quantified by measuring the displacement of center of pressure (COP) in laboratory settings [1]. Maintaining postural stability during upright stance relies on sensory inputs from multiple modalities, including visual, vestibular, and proprioceptive systems [2]. The central nervous system (CNS) processes and integrates these multisensory inputs to generate appropriate motor commands and initiate corrective postural adjustments [3]. Sensory conflict arises when multiple sensory modalities concurrently convey incongruent information to the CNS, prompting it to dynamically redistribute sensory weights to resolve the sensory mismatch, a process known as sensory reweighting [4].

Vision provides critical spatial orientation information regarding body position relative to the environment, facilitating postural adaptation during locomotor activities [5]. To investigate the role of visual feedback in maintaining postural stability, previous researchers commonly employ paradigms of eye closure or blindfolding, which effectively block visual feedback. However, an inherent limitation of this approach lies in its non-adaptive, all-or-nothing characteristic, which leads to sensory deprivation rather than genuine sensory conflict [4]. A typical example is reading in a moving vehicle, where vestibular afferents signal self-motion while visual inputs indicate a stationary environment relative to the book, thereby inducing motion sickness. If visual feedback is completely eliminated under this circumstance (e.g., eye closure during vehicle motion), only vestibular motion signals persist without inter-sensory incongruence, and motion sickness is typically attenuated. This distinction underscores that sensory deprivation (absence of one modality) fundamentally differs from sensory conflict (simultaneous presence of incongruent modalities). Consequently, the binary elimination of visual inputs not only artificially disrupts human–environment interactions, but fails to elicit sensory conflict, restricting the investigation of postural control mechanisms under sensory-conflicted conditions [6].

Over the past decade, virtual reality (VR) has been widely used to simulate virtual environments for inducing visual perturbations and postural instability [7]. By presenting visual motion cues without corresponding vestibular acceleration signals, VR induces sensory conflict and spatial disorientation [8]. Unlike visual deprivation, VR perturbation preserves visual inputs while distorting their reliability, thereby replicating the defining feature of sensory conflict. For instance, the visual-vestibular incongruence elicited by VR simulations shares the same conflict signature as real-world scenarios such as the illusory self-motion experienced when standing in a stationary train while observing adjacent train movement, where visual inputs suggest movement but vestibular inputs indicate stability. This capacity enables VR to reproduce sensory conflict under controlled laboratory conditions, thereby offering superior validity compared to eyes-closed paradigms [5]. Consistent with this rationale, Chander et al. [9] demonstrated that VR perturbation causes a greater reduction in postural stability relative to blindfolding, as evidenced by larger increases in COP displacement and sway velocity. Another study measuring COP displacement and surface electromyography (sEMG) revealed that VR perturbations elicited a shift from the ankle to hip strategy, two fundamental postural control strategies for maintaining standing balance [10]. This adaptive change, associated with greater hip moment and shorter muscle recruitment latency, serves to maintain postural stability during significant visual perturbations [11, 12]. Functional near-infrared spectroscopy (fNIRS) studies further identified enhanced cortical activation in the prefrontal cortex, primary motor (M1) and sensory (S1) cortices, and parietal cortex during standing tasks with VR, highlighting an increased demand for cognitive resources to resolve sensory conflict and maintain postural stability [12–14].

Given that VR-induced postural instability stems from visual-vestibular incongruence, restoring postural stability requires the CNS to dynamically reweight sensory contributions according to relative reliability of sensory inputs [4, 15]. During VR perturbation, visual inputs become spatially incongruent with vestibular signals, reducing their reliability and eliciting maladaptive reliance on distorted visual feedback [9, 13]. To resolve this incongruence, the CNS adaptively up-weights more reliable sensory modalities while down-weighting unreliable visual inputs to generate accurate motor commands [15]. Therefore, enhancing the relative reliability of vestibular inputs may facilitate adaptive reweighting process, thereby reducing dependence on distorted visual cues and alleviating the adverse effects of VR-induced sensory conflict on postural stability [16, 17]. Noisy galvanic vestibular stimulation (nGVS) represents a promising neuromodulation approach to achieve this goal [18]. By delivering sub-threshold electrical noise to vestibular afferents, nGVS induces stochastic resonance to lower the vestibular perceptual threshold (VPT), thereby optimizing the detection and temporal coherence of weak vestibular signals [19–21]. The enhancement of vestibular signal fidelity has been shown to reduce cybersickness and improve motion perception under VR perturbation, supporting the plausibility that nGVS facilitates adaptive sensory reweighting to mitigate visual-vestibular conflict [8, 22, 23]. These improvements may be reflected in altered activation patterns within frontoparietal networks governing vestibular processing and multisensory integration [24].

Published fNIRS studies show that nGVS significantly activates the supramarginal gyrus (SMG), a dominant region of the vestibular cortex involved in vestibular signal processing and spatial orientation [24, 25]. Furthermore, McCarthy et al. [26, 27] reported altered activation in the dorsolateral prefrontal cortex (DLPFC) during nGVS, revealing its crucial role in multisensory integration and the top-down control of vestibular inputs. Collectively, these neurophysiological findings provide preliminary insights that nGVS may not only enhance peripheral vestibular inputs but also engage higher-order cortical networks to counteract sensory conflict [28]. However, it remains unclear how nGVS modulates cortical activation and the neuromuscular control strategies by which it maintains postural stability. More importantly, the potential roles of the SMG and DLPFC in processing nGVS-enhanced vestibular inputs require systematic exploration.

Therefore, this study aims to investigate the effects of nGVS on postural stability, muscle activation, and cortical hemodynamics during upright stance with VR perturbation using a synchronized multimodal assessment. We hypothesized that: (1) nGVS would enhance postural stability under VR perturbation; (2) nGVS synchronized with VR perturbation would modulate neuromuscular control strategies; and (3) nGVS-induced improvements in postural stability and neuromuscular control would be associated with altered activation in the SMG and DLPFC, suggesting their involvement in vestibular processing and multisensory integration.

Methods

Participants

Healthy young adults aged 18 to 35 years were recruited through social media. The exclusion criteria were as follows: (1) a history of neurological, psychiatric, musculoskeletal, cardiovascular, ophthalmological, or vestibular disorders; (2) current experience of any pain; (3) pregnancy; (4) use of medications known to affect the CNS or sensorimotor system; and (5) a history of severe cybersickness susceptibility (e.g., nausea, dizziness, or disorientation) during prior immersive VR/3D experiences. This pre-screening strategy filtered out highly susceptible individuals while retaining those with normal physiological variability in VR tolerance [29]. Following recruitment, a total of 30 participants (12 males and 18 females) were enrolled in this study (Table 1). All participants were right-handed and had normal or corrected-to-normal vision. This study was approved by the Independent Ethics Committee for Clinical Research and Animal Trials of the First Affiliated Hospital of Sun Yat-sen University (IRB# [2023]793), and registered at the Chinese Clinical Trial Registry (ChiCTR# 2300078910). Written informed consent was obtained from each participant prior to data collection. Participants were permitted to withdraw from the study at any time without providing a justification.

Table 1.

Demographic characteristics, vestibular perceptual threshold, noisy galvanic vestibular stimulation (nGVS) intensity, and virtual reality (VR)-induced cybersickness of all participants

	Mean ± SD
Age (years)	21.00 ± 1.80
Sex (male/female)	12/18
Height (cm)	167.22 ± 9.30
Weight (kg)	59.23 ± 10.61
Body Mass Index (kg/m2)	21.07 ± 2.49
Vestibular perceptual threshold (μA)	472.60 ± 53.75
nGVS intensity (μA)	378.08 ± 43.02
VR-induced cybersickness with nGVS	4.32 ± 1.12
• Video 1: Spaceflight	4.19 ± 1.05
• Video 2: Moving corridor	4.35 ± 1.18
• Video 3: Spiral tunnel	4.25 ± 1.02
VR-induced cybersickness without nGVS	5.80 ± 1.45
• Video 1: Spaceflight	5.73 ± 1.34
• Video 2: Moving corridor	5.80 ± 1.58
• Video 3: Spiral tunnel	5.78 ± 1.49

nGVS

nGVS was delivered using a battery-driven current stimulator (BC102-IV; BrainCLOS Co., Ltd., Shenzhen, Guangdong, China) via a pair of circular silicone gel electrodes (surface area: 1.75cm²) placed bilaterally over the mastoid processes (Fig. 1A). Electrodes were securely affixed to the skin using hypoallergenic tape to ensure stable contact. The electrical signal comprised zero-mean Gaussian white noise within a frequency range of 4-20 Hz [19]. The nGVS intensity was set to 80% of each participant’s VPT, which has been established as the optimal nGVS intensity for healthy individuals [30]. The VPT was determined by the stepwise method, in accordance with the protocol established by Wuehr et al. [21]. Participants were instructed to maintain a natural upright stance, facing forward with eyes open and arms relaxed at their sides. After the nGVS device was properly positioned, electrical stimuli were delivered in 10-s intervals starting at an initial intensity of 0.1 mA. The current intensity was then incrementally increased in steps of 0.05 mA until the participant perceived a mild tingling sensation at the electrode sites. This procedure was repeated three times, and the average intensity recorded across these trials was defined as the final VPT. Group means for VPT and nGVS intensity are presented in Table 1. For the sham condition, the nGVS intensity was set to 0 mA.

Fig. 1.

A Experimental setup in this study. Participants maintained an upright stance on a force plate while equipped with (1) a functional near-infrared spectroscopy (fNIRS) cap, (2) sixteen surface electromyography (sEMG) sensors, and (3) noisy galvanic vestibular stimulation (nGVS) device. nGVS was delivered with electrodes placed over bilateral mastoid processes by a portable current stimulator. Three visual conditions included eyes-open, eyes-closed, and virtual reality (VR). B The experimental protocol for each trial. AP: anterior–posterior; COP: center of pressure; ML: medial–lateral

As nGVS at 80% of VPT constitutes a subthreshold stimulation that is imperceptible to conscious awareness, participants could not subjectively detect the presence of stimulation and were therefore unable to distinguish active nGVS from sham stimulation. Furthermore, the stimulator operated silently in both conditions without generating audible clicks, beeps, or tactile cues that might serve as discrimination signals. This blinding approach has been previously validated in our study employing identical nGVS protocols [19].

COP measurements

A six-axis force plate (464 mm × 508 mm × 83 mm, AMTI, Watertown, MA, USA) was used to record COP data at a sampling rate of 1000 Hz via AMTI NetForce software (version 3.6.04). The force plate recorded ground reaction forces (F) and moments (M) along the medial–lateral (ML, x), anterior–posterior (AP, y), and vertical (z) directions, denoted as F_x, F_y, F_z, M_x, M_y, and M_z, respectively. The analog signals were amplified via a signal amplifier and subsequently processed using AMTI NetForce software (AMTI, Watertown, MA, USA). Raw data were filtered with a fourth-order low-pass Butterworth filter at a cut-off frequency of 10 Hz using a custom MATLAB script (The MathWorks, Inc., Natick, MA, USA). Then, COP coordinates were computed using the following formulae [31],

where d_z represented the vertical distance from the force plate surface to its origin (0.04125 m). In the present study, two COP-derived metrics were used to evaluate postural stability during upright stance, as shown in Fig. 1A. These parameters have proven effective in assessing both linear and non-linear aspects of balance control, as validated in healthy populations.

1. The 95% confidence ellipse area

   A 95% confidence ellipse was constructed to encompass approximately 95% of the COP data points per task, along with the AP and ML axis lengths, was calculated as follows,

   

   where donated the F-statistic at confidence level 1—α for n data points, S_AP and S_ML represented the standard deviations of AP and ML axes, and S_AP,ML was their covariance. Based on prior research, was used for large datasets (N ≥ 120) [32]. In this study, each task yielded 25,000 data points (25 s × 1000 Hz) for analysis, justifying the use of . It should be noted that larger values of the 95% confidence ellipse area, AP axis, and ML axis indicated poorer postural stability in standing [33].

2. Sample entropy (SaEn)

   The measurement of SaEn was used to assess the regularity of COP displacement along AP (SaEn_AP) and ML (SaEn_ML) axes in time series during upright standing. As a non-linear COP-related parameter, SaEn was defined as the negative natural logarithm of the conditional probability that a sequence of data points spaced a certain distance apart, for a given embedding dimension m, will repeat itself at m + 1. The calculation of SaEn involved four critical parameters: the embedding dimension (m), the tolerance (γ), the time delay (τ), and the time series length (N). Given a time series with constant time intervals, a template vector of length m was defined as and a distance function was applied as . SaEn was then computed as:

   

   where both and were less than γ. In accordance with our recent publications, the values γ = 0.35 and m = 4 were selected, as these parameters were verified to detect subtle differences in COP trajectories between healthy young and older adults [34, 35]. Additionally, a time delay (τ) of 5 was applied to all COP data to reduce the sampling rate from 1000 to 200 Hz, resulting in a final dataset size of 5000 points (N = 25 s × 200 Hz). The rational for introducing this time delay was that a unity delay (τ = 1) may predominantly reflect the linear autocorrelation properties of COP data, thereby limiting the capacity of SaEn to quantify non-linear features of COP displacement within the time series.

sEMG acquisition and analysis

The Trigno^™ wireless sEMG system (Delsys Inc., Natick, MA, USA) was employed to record sEMG signals from bilateral rectus femoris (RF), biceps femoris (BF), tibialis anterior (TA), gastrocnemius lateralis (GL), transversus abdominis/internal oblique complex (TrA/IO), and paraspinal muscles at the seventh (T₇), twelfth (T₁₂) thoracic, and third lumbar (L₃) vertebral levels, with a sampling rate of 2000 Hz. The above-mentioned muscle targets were selected to comprehensively assess neuromuscular control strategies across the lower limb and trunk, specifically enabling the detection of nGVS-induced modulations in postural control strategies and trunk stabilization under visual-vestibular conflict [36, 37]. Following standard skin preparation procedures (shaving and cleansing with exfoliating cream and 75% alcohol), sEMG sensors were positioned over the muscle bellies in alignment with the orientation of muscle fibers, in accordance with the SENIAM project (http://www.seniam.org/). The localization of target muscles for sensor placement strictly adhered to established guidelines [38]. Notably, composite activity of TrA/IO was captured with sensors placed 2 cm medial and inferior to the anterior superior iliac spine, which has been validated to reflect deep abdominal muscle activation during quiet standing tasks where rectus abdominis contribution is minimal [39]. TrA and IO activate synergistically during upright stance to maintain postural stability, rendering their composite signal biomechanically valid for balance control assessment [40]. Sensor placement details are provided in Supplementary Table S1. A customized MATLAB script was used to process sEMG signals (The MathWorks, Inc., Natick, MA, USA). Raw data were first processed to remove the DC offset by subtracting the mean value, then bandpass-filtered (20–450 Hz) using a fourth-order zero-lag Butterworth filter to eliminate low-frequency baseline drift and high-frequency noise while preserving the physiological frequency content of muscle activity [39]. The filtered signals were subsequently subjected to full-wave rectification for calculation of following sEMG parameters [40].

Normalized muscle activity was quantified using the root-mean-square (RMS) value was employed to quantify muscle activation levels during upright stance. To enable inter-subject comparisons, the RMS values of all target muscles were normalized to their respective maximum voluntary isometric contraction (MVIC) values. Specifically, prior to experimental trials, participants were instructed to perform three 5-s MVIC tasks for each target muscle. Detailed protocols for MVIC assessment are provided in Supplementary Table S1. The mean RMS value from these tasks served as the normalization reference. Normalized muscle activity was expressed as a percentage of MVIC (%MVIC), computed as follows [39]:

where RMS_trial and RMS_MVIC represented the RMS values obtained from standing trials and MVIC tasks, respectively. This approach offered a physiologically meaningful interpretation of muscle activation by representing effort relative to each target muscle’s maximum capacity.

The co-contraction index (CCI) was computed to assess neuromuscular coordination patterns for selected muscle pairs. After RMS conversion and MVIC normalization, the CCI was calculated using the following formula [41]:

where sEMG₁(t) and sEMG₂(t) denoted the %MVIC values of two target muscles in each pair at sample point t. In the present study, the CCI was computed for agonist–antagonist pairs (RF-BF, TA-GL, and TrA/IO-L₃) and bilateral synergist pairs (left–right RF, BF, TA, GL, TrA/IO, T₇, T₁₂, and L₃). The RF-BF, TA-GL, and TrA/IO-L₃ pairs were included to reflect recruitment and modulation strategies within the lower limb and trunk musculature relevant to postural control in the sagittal plane [36]. Bilateral synergist pairs were analyzed due to their role in evaluating segmental stabilization and postural regulation in the frontal plane [37]. The mean CCI value across trials was used for statistical comparisons among standing conditions.

fNIRS recording and processing

Hemodynamic activity in the frontal, parietal, and occipital cortices was measured using a multi-channel fNIRS system (NirSmart-6000A; Huichuang Medical Equipment Co., Ltd., Danyang, Jiangsu, China). This system operated at wavelengths of 730 nm and 850 nm, with a sampling rate of 11 Hz. A custom fNIRS cap, comprising 28 sources and 28 detectors, was designed to yield 65 effective channels. Optode placement was guided by the international 10–20 system for standardized positioning, with a source-detector distance of 3.0 cm. One optode along the midline (labeled D3) was positioned at the frontal pole zero (Fpz) location. The regions of interest (ROIs) included bilateral (left: L; right: R) DLPFC, premotor cortex (PMC), supplementary motor area (SMA), M1, S1, and SMG. These specific cortical regions were selected to target neural networks critical for vestibular processing (SMG), multisensory integration and cognitive control (DLPFC), and motor execution (M1, SMA, PMC), thereby allowing for a comprehensive assessment of the central mechanisms underlying nGVS-induced modulations in postural control, according to published studies [24–27]. Figure 2 and Supplementary Table S2 display the spatial arrangement of optodes and the Montreal Neurological Institute coordinates of representative channels within each ROI.

Fig. 2.

Functional near-infrared spectroscopy (fNIRS) montage showing probe distribution and channel locations. DLPFC: dorsolateral prefrontal cortex; Fpz: frontal pole zero; M1: primary motor cortex; PMC: premotor cortex; SMA: supplementary motor area; SMG: supramarginal gyrus; S1: primary somatosensory cortex

fNIRS data were preprocessed using MATLAB-based NirSpark software (version 1.8.1). Raw data were first converted to optical density. Then, a bandpass filter (0.01–0.2 Hz) was applied to attenuate motion artifacts and physiological noise [42]. Subsequently, filtered optical density data were converted to changes in oxygenated (HbO) and deoxygenated hemoglobin via the modified Beer-Lambert law [43]. Given its heightened sensitivity to cortical hemodynamic changes, HbO served as the primary outcome measure in this study. Hemodynamic responses within ROIs were quantified using a general linear model, with beta values (β, μmol/L) serving as indicators of cortical activation [44]. Specifically, positive β values reflected increased task-evoked cortical activation, whereas negative β values indicated task-related deactivation [44]. For each ROI, the mean β value across multiple channels was used for subsequent analysis.

Experimental protocol

This study employed a repeated-measures design including six standing trials, combining three visual conditions (eyes-open (EO), eyes-closed (EC), and VR) with two stimulation conditions (nGVS and sham stimulation). Six trials were administered to each participant in a randomized order, following a standardized protocol comprising: (1) a 10-s baseline period, (2) three 30-s task blocks alternating with (3) three 40-s rest blocks. The experimental protocol for a single trial is illustrated in Fig. 1B.

During each trial, participants quietly stood on a force plate while wearing a fNIRS cap and sEMG sensors, with their hands resting at both sides. At the onset of each task, a trained experimenter: (1) instructed participants to either open/close their eyes or initiated VR perturbation, and (2) administered either nGVS or sham stimulation. For VR trials only, participants wore a pair of VR goggles (G05A; Shinecon Industrial Co., Ltd., Dongguan, Guangdong, China) weighing 329 g (plus the embedded smartphone). The VR goggles featured soft silicone padding around the eyes and nasal bridge, with three adjustable straps to ensure a secure and comfortable fit. Three distinct VR videos (spaceflight, moving corridor, and spiral tunnel) were displayed at a constant optic flow velocity of 0.5 m/s to provide visual perturbation [45]. Technical specifications of the VR system and videos are summarized in Supplementary Table S3. The VR device and paradigm were identical to our previously validated protocol, which confirmed equivalent stereoscopic perception and sensory conflict induction across the three VR videos [46, 47]. To minimize familiarization and anticipation effects, each participant experienced all three VR videos exactly once per stimulation condition, with the presentation order randomized across participants [47]. After each VR trial, subjective cybersickness was quantified using a 10-cm visual analog scale ranging from 0 (no cybersickness) to 10 (extreme cybersickness) to quantify inter-individual differences in VR susceptibility. This repeated-measures approach enabled trial-by-trial quantification of cybersickness, thereby confirming whether visual perturbation across VR scenarios was comparable. During baseline and rest periods, participants stood upright and looked forward with eyes open. Adverse effects (e.g., nausea or dizziness) were systematically queried after each trial, with provision for early termination if significant discomfort occurred. A mandatory 2-min interval was implemented between trials to minimize carry-over effects, with additional rest available upon request [35]. Figure 1A illustrates a schematic overview of experimental conditions.

COP, sEMG, and fNIRS signals were continuously recorded throughout each trial in a time-synchronized manner. Considering potential postural adjustments at the beginning of standing tasks, which could cause transient signal artifacts, the first 5-s interval was discarded. Thus, a standardized 25-s interval (from 5 to 30 s) was extracted from each task block for analysis. Consequently, each experimental condition yielded a total of 90 data segments (30 participants × 3 task blocks) for statistical analysis.

Statistical analysis

Statistical analysis was conducted by SPSS 20.0 (IBM Corp., Armonk, NY, USA). Normality of each dependent variable was assessed using the Shapiro–Wilk test. For normally distributed data, a two-way repeated-measures ANOVA with two within-subjects factors (3 visual conditions × 2 stimulation conditions) was conducted to examine interactions and main effects, followed by pairwise comparisons for post-hoc tests. For non-normally distributed data, the Friedman test was applied, and post-hoc comparisons were conducted using the Wilcoxon signed-rank test. All post-hoc tests were corrected for multiple comparisons using the Bonferroni method to control Type I error inflation at the contrast level within each statistical model. However, no adjustment across the separate families of analyses for distinct outcome domains (i.e., COP-related parameters, %MVIC, CCI, and β values) was applied. This approach, correcting within families but not across them, aligned with established methodological recommendations for hypothesis-driven research with pre-specified primary outcomes and a priori selection of ROIs and muscle targets [48]. Moreover, a two-way repeated-measures analysis of variance (ANOVA) was used to compare cybersickness levels across three VR videos under nGVS and sham stimulation. Partial eta squared (η²_p) was calculated to assess effect sizes, based on Cohen’s guidelines (large: ≥ 0.14; medium: ≥ 0.06; small: ≥ 0.01) [49]. The significant level was set at 0.05.

Results

All participants completed the experiment without early termination due to intolerable discomfort. The assumption of normality was met for all dependent variables, as assessed by the Shapiro–Wilk test (ps > 0.05). Analysis of the VAS ratings revealed a moderate inter-individual variability in VR-induced cybersickness (coefficient of variation: 25.9% with nGVS; 25.0% without nGVS). A significant main effect of nGVS was observed on cybersickness level (F_(1,29) = 7.635, p = 0.010, η²_p = 0.208), with nGVS reducing cybersickness compared to sham stimulation. No significant main effect of VR videos was found (F_(2,58) = 1.201, p = 0.308, η²_p = 0.040), confirming equivalent levels of visual perturbation across VR videos (Table 1).

COP-related parameters

Figure 3 and Supplementary Table S4 present the COP-related parameters and corresponding statistical results. Significant two-way interactions were observed for the 95% confidence ellipse area (F_(2,58) = 8.013, p = 0.001, η²_p = 0.217), AP axis (F_(2,58) = 3.796, p = 0.028, η²_p = 0.116), and ML axis (F_(2,58) = 4.332, p = 0.018, η²_p = 0.130). Post-hoc analyses revealed that (1) compared to sham stimulation, nGVS significantly reduced the 95% confidence ellipse area (EC: t = -4.339, p < 0.001; VR: t = − 2.703, p = 0.009), AP axis (EC: t = − 2.796, p = 0.007; VR: t = − 2.285, p = 0.026), and ML axis (EC: t = − 2.998, p = 0.004; VR: t = − 2.239, p = 0.029) in the EC and VR conditions; (2) compared to the EO condition, the EC and VR conditions significantly increased the 95% confidence ellipse area (EC: t = 5.261, p < 0.001; VR: t = 3.922, p < 0.001), AP axis (EC: t = 4.748, p < 0.001; VR: t = 5.048, p < 0.001), and ML axis (EC: t = 3.446, p < 0.001; VR: t = 3.679, p < 0.001) under both nGVS and sham stimulation; and (3) there was no significant difference between the EC and VR conditions irrespective of stimulation conditions. Additionally, significant main effects of stimulation conditions (F_(1,29) = 15.748, p < 0.001, η²_p = 0.352) and visual conditions (F_(2,58) = 24.402, p < 0.001, η²_p = 0.457) were observed for SaEn_ML. Specifically, nGVS significantly decreased SaEn_ML compared to sham stimulation (t = -2.887, p = 0.005). Moreover, SaEn_ML was significantly higher in the VR condition than the EO (t = 5.316, p < 0.001) and EC (t = -3.935, p < 0.001) conditions.

Fig. 3.

The center of pressure (COP)-related parameters in different standing trials. ### represented the significant main effect of visual conditions (p < 0.001). * represented the significant differences between noisy galvanic vestibular stimulation (nGVS) and sham stimulation (*: p < 0.05; **: p < 0.01; ***: p < 0.001). & represented the main effect of noisy nGVS (p < 0.05). ###/& denoted significant main effects of both visual conditions and nGVS for the parameter. AP: anterior–posterior; EC: eyes-open; EO: eyes-closed; ML: medial–lateral; SaEn: sample entropy; VR: virtual reality

sEMG-%MVIC

The %MVIC values of all target muscles and corresponding statistical results are presented in Fig. 4 and Supplementary Table S5. Significant two-way interactions were found for L_RF (F_(2,58) = 3.691, p = 0.031, η²_p = 0.113), R_RF (F_(2,58) = 3.277, p = 0.045, η²_p = 0.102), and L_BF (F_(2,58) = 3.331, p = 0.043, η²_p = 0.103). Post-hoc analyses indicated that (1) compared to sham stimulation, nGVS significantly decreased %MVIC of L_RF (t = -3.086, p = 0.003), R_RF (t = -2.440, p = 0.018), and L_BF (t = -2.396, p = 0.020) during standing with VR perturbation; and (2) the VR condition significantly increased %MVIC of L_RF compared to the EO (t = 2.618, p = 0.011) and EC (t = 2.041, p = 0.046) conditions, and in R_RF compared to EO condition (t = 2.501, p = 0.015) under sham stimulation, but not nGVS. Furthermore, a significant main effect of visual conditions was observed for R_BF (F_(2,58) = 3.890, p = 0.026, η²_p = 0.118). Post-hoc comparisons showed that the VR condition significantly increased %MVIC relative to both EO (t = 3.314, p = 0.002) and EC (t = 2.471, p = 0.016) conditions.

Fig. 4.

The %MVIC values of target muscles in different standing trials. # represented the significant main effect of visual conditions (p < 0.05). * represented the significant differences between noisy galvanic vestibular stimulation (nGVS) and sham stimulation (p < 0.05). BF: biceps femoris; EC: eyes-open; EO: eyes-closed; GL: gastrocnemius lateralis; L: left; L₃: paraspinal muscle at the third lumbar vertebral level; R: right; RF: rectus femoris; TA: tibialis anterior; TrA/IO: transversus abdominis/internal oblique complex; T₇: paraspinal muscle at the seventh thoracic vertebral level; T₁₂: paraspinal muscle at the twelfth thoracic vertebral level; VR: virtual reality.

sEMG-CCI

The CCI values of all muscle pairs and corresponding statistical results are presented in Fig. 5 and Supplementary Table S6. Although no significant two-way interaction was observed, significant main effects of stimulation and visual conditions were identified for the CCI values of left (stimulation: F_(1,29) = 7.373, p = 0.011, η²_p = 0.203; visual: F_(2,58) = 3.428, p = 0.039, η²_p = 0.106) and right (stimulation: F_(1,29) = 6.855, p = 0.014, η²_p = 0.191; visual: F_(2,58) = 4.047, p = 0.023, η²_p = 0.122) RF-BF pairs. Pairwise comparisons revealed that (1) nGVS significantly reduced CCI of left (t = -2.292, p = 0.026) and right (t = -2.272, p = 0.027) RF-BF pairs compared to sham stimulation; (2) CCI of left (t = 2.555, p = 0.013) and right (t = 3.084, p = 0.003) RF-BF pairs significantly increased in the VR condition than the EO condition.

Fig. 5.

The CCI values of selected muscle pairs in different standing trials. # represented the significant main effect of visual conditions (p < 0.05). & represented the significant main effect of noisy galvanic vestibular stimulation (nGVS) (p < 0.05). #/& denoted significant main effects of both visual conditions and nGVS for the parameter. BF: biceps femoris; EC: eyes-open; EO: eyes-closed; GL: gastrocnemius lateralis; L₃: paraspinal muscle at the third lumbar vertebral level; RF: rectus femoris; TA: tibialis anterior; TrA/IO: transversus abdominis/internal oblique complex; T₇: paraspinal muscle at the seventh thoracic vertebral level; T₁₂: paraspinal muscle at the twelfth thoracic vertebral level; VR: virtual reality.

fNIRS-β values

The β values of all ROIs are presented in Fig. 6 and Table 2. Significant two-way interactions were identified for L_DLPFC (F_(2,58) = 6.128, p = 0.004, η²_p = 0.174), R_DLPFC (F_(2,58) = 4.228, p = 0.019, η²_p = 0.127), L_SMG (F_(2,58) = 7.663, p = 0.001, η²_p = 0.209), and R_SMG (F_(2,58) = 9.446, p < 0.001, η²_p = 0.246). Post-hoc analyses demonstrated that (1) in the VR condition, there were significantly lower β values of L_DLPFC (t = -2.124, p = 0.038) and R_DLPFC (t = -2.029, p = 0.047) but higher β values of L_SMG (t = 2.024, p = 0.048) and R_ SMG (t = 2.236, p = 0.029) observed under nGVS; (2) under nGVS, both EC and VR conditions significantly increased β values of L_DLPFC (EC: t = 2.328, p = 0.023; VR: t = 3.859, p < 0.001) and R_DLPFC (EC: t = 2.079, p = 0.042; VR: t = 2.244, p = 0.029) than the EO condition; (3) under sham stimulation, there was a stepwise increase in β values of the bilateral DLPFC with increasing visual perturbation (EO < EC < VR) (ps < 0.05); and (4) regardless of stimulation condition, β values of L_SMG (EO: t = 2.404, p = 0.019; EC: t = 3.843, p < 0.001) and R_ SMG (EO: t = 2.457, p = 0.017;EC: t = 2.333, p = 0.023) were significantly higher in the VR condition. Moreover, significant main effects of visual conditions were observed for L_PMC/SMA (F_(2,58) = 6.482, p = 0.003, η²_p = 0.183), R_PMC/SMA (F_(2,58) = 7.589, p = 0.001, η²_p = 0.207), L_M1 (F_(2,58) = 44.190, p < 0.001, η²_p = 0.604), R_M1 (F_(2,58) = 38.609, p < 0.001, η²_p = 0.571), L_S1 (F_(2,58) = 33.274, p < 0.001, η²_p = 0.534), R_S1 (F_(2,58) = 44.413, p < 0.001, η²_p = 0.605). Specifically, post-hoc comparisons demonstrated that there were significantly higher β values of the above-mentioned ROIs in the VR condition compared to both EO and EC conditions (ps < 0.05).

Fig. 6.

Differences in the beta values (β, μmol/L) of oxygenated hemoglobin (HbO) concentration for regions of interest (ROIs) between noisy galvanic vestibular stimulation (nGVS) and sham stimulation in (A) eyes-open (EO), (B) eyes-closed (EC), and (C) virtual reality (VR) conditions. * represented significant differences between nGVS and sham stimulation (*: p < 0.05; **: p < 0.01). DLPFC: dorsolateral prefrontal cortex; L: left; M1: primary motor cortex; PMC: premotor cortex; R: right; SMA: supplementary motor area; SMG: supramarginal gyrus; S1: primary somatosensory cortex.

Table 2.

The beta values (β, μmol/L) of oxygenated hemoglobin concentration in regions of interest (ROIs) and statistical results from a two-way repeated-measures ANOVA

	Visual conditions	nGVS	Sham	Interaction			Vision effect			nGVS effect
				F(2,58)	p	η2p	F(2,58)	p	η2p	F(1,29)	p	η2p
L_DLPFC	EO	− 8.739 ± 7.644	2.546 ± 6.712	6.128	0.004*	0.174	37.727	< 0.001*	0.565	6.289	0.018*	0.178
	EC	19.918 ± 5.696	25.475 ± 4.883
	VR	22.284 ± 7.711	43.797 ± 6.568
R_DLPFC	EO	− 6.776 ± 10.355	3.271 ± 7.864	4.228	0.019*	0.127	10.181	< 0.001*	0.260	7.225	0.012*	0.199
	EC	15.591 ± 7.737	34.466 ± 7.917
	VR	15.066 ± 10.810	51.904 ± 14.585
L_PMC/SMA	EO	− 1.507 ± 8.038	4.749 ± 7.948	0.191	0.827	0.007	6.482	0.003*	0.183	0.132	0.719	0.005
	EC	9.812 ± 8.539	11.451 ± 5.312
	VR	23.676 ± 9.863	29.189 ± 10.489
R_PMC/SMA	EO	− 1.225 ± 8.355	3.579 ± 8.465	0.262	0.770	0.009	7.589	0.001*	0.207	0.852	0.364	0.029
	EC	18.227 ± 10.920	25.847 ± 4.038
	VR	27.233 ± 13.269	39.371 ± 8.927
L_M1	EO	10.017 ± 11.255	17.075 ± 9.148	0.238	0.789	0.008	44.190	< 0.001*	0.604	0.020	0.888	0.001
	EC	30.985 ± 4.788	18.700 ± 8.490
	VR	59.022 ± 9.172	62.460 ± 10.622
R_M1	EO	3.591 ± 8.415	7.240 ± 10.009	0.058	0.943	0.002	38.609	< 0.001*	0.571	0.512	0.480	0.017
	EC	12.809 ± 8.834	20.697 ± 6.190
	VR	41.333 ± 10.794	47.422 ± 4.889
L_S1	EO	10.805 ± 7.856	9.472 ± 7.140	0.146	0.865	0.005	33.274	< 0.001*	0.534	0.284	0.598	0.010
	EC	12.392 ± 10.207	17.974 ± 13.031
	VR	61.173 ± 17.053	70.453 ± 17.283
R_S1	EO	9.084 ± 7.110	14.217 ± 5.672	0.133	0.875	0.005	44.413	< 0.001*	0.605	0.241	0.627	0.008
	EC	25.544 ± 9.967	35.682 ± 8.014
	VR	53.919 ± 16.367	70.152 ± 17.813
L_SMG	EO	6.531 ± 8.496	0.951 ± 9.406	7.663	0.001*	0.209	51.403	< 0.001*	0.639	5.141	0.031*	0.151
	EC	37.566 ± 5.469	22.211 ± 8.250
	VR	64.078 ± 10.787	33.568 ± 13.947
R_SMG	EO	14.033 ± 9.011	6.767 ± 9.824	9.446	< 0.001*	0.246	27.066	< 0.001*	0.483	8.622	0.006*	0.229
	EC	20.783 ± 6.047	14.451 ± 7.952
	VR	56.290 ± 10.127	25.017 ± 9.642

Data were shown as the Mean ± SE. * indicated statistical significance (p < 0.05). DLPFC: dorsolateral prefrontal cortex; EC: eyes-closed; EO: eyes-open; L: left; M1: primary motor cortex; nGVS: noisy galvanic vestibular stimulation; PMC: premotor cortex; R: right; SMA: supplementary motor area; SMG: supramarginal gyrus; S1: primary somatosensory cortex; VR: virtual reality perturbation; η²_p: the partial eta squared value

Discussion

The present study systematically investigated the effects of nGVS on postural stability, muscle activation, and cortical hemodynamics in healthy young adults during upright stance with VR perturbation. The results supported our hypotheses, demonstrating that nGVS modulated cortical activity in regions involved in vestibular processing and multisensory integration, thereby mitigating VR-induced visual-vestibular incongruence. This central effect subsequently led to attenuated hip muscle co-activation, reduced reliance on the hip strategy, and improved postural control. These findings reveal that nGVS may serve as a promising neuromodulation approach to ameliorate the VR-induced visual-vestibular incongruence and enhance postural stability, with potential applications in sensorimotor adaptation training for clinical rehabilitation and aerospace medicine.

nGVS mitigated VR-induced sensory conflict and improved postural stability by reducing reliance on hip strategy

In the present study, significant main effects of visual conditions were observed across COP-related parameters, muscle activation, and cortical hemodynamics, reflecting the CNS's adaptive sensory reweighting in response to varying visual reliability. Specifically, both eyes-closed and VR conditions increased COP displacement, hip muscle co-activation, and cortical engagement relative to the eyes-open conditions, indicating increased reliance on proximal control and cortical processing when visual inputs become unreliable [4, 6]. Notably, VR perturbation, which preserves but distorts visual feedback, elicited the most pronounced effects across central and peripheral domains, suggesting that visual-vestibular incongruence poses a greater challenge to postural control than simple visual deprivation [9]. Consistent with this interpretation, VR perturbation significantly compromises postural stability during upright stance, as evidenced by significantly increased 95% confidence ellipse area, AP and ML axis lengths. These findings align with previous studies demonstrating that VR effectively induces sensory conflict and increases postural sway [5, 46, 50]. Under such conditions, healthy individuals typically shift toward greater reliance on the hip strategy, a compensatory postural response characterized by coordinated hip muscle contraction, particularly the quadriceps femoris and hamstrings, to generate rapid torque and stabilize the upper body [12, 51]. Moreover, the increased CCI of RF-BF under VR perturbation may reflect heightened neuromuscular co-activation, potentially intensifying the hip joint stiffness. Elevated CCI is a recognized EMG marker of a stiffening strategy aimed at increasing postural rigidity under sensory conflict [52]. Notably, several studies have reported similar shifts toward hip-dominant control in response to sensory conflict. For instance, Ritzmann et al. [53] observed that while standing with perturbation, healthy young adults exhibited greater co-activation of antagonistic muscle pairs around the hip, accompanied by reduced hip joint excursions, thereby limiting COP displacement. Although this adjustment of postural strategy promotes stability, it comes at the cost of lower movement efficiency and higher energy expenditure [54]. Likewise, Peterka [15] suggested that unpredictable visual inputs from VR perturbation prompted a transition from ankle- to hip-dominant strategy, particularly when vision became the dominant sensory cues. Collectively, these observations reinforce the insight that visual perturbation elicits proximal (hip-dominant) control mechanisms for maintaining postural stabilit, highlighting its potential use for sensorimotor training in balance rehabilitation [9].

To mitigate VR-induced sensory conflict, nGVS has emerged as a feasible approach for reducing cybersickness and improving spatial memory [22]. According to available evidence, nGVS also benefited postural control and locomotion during standing and walking across various populations, manifested as reduced COP displacement and enhanced gait performance [55–58]. To our knowledge, this study is the first to evaluate the effects of nGVS on VR-induced postural instability using multimodal assessment. Aligned with previous studies, we found that nGVS synchronized with VR perturbation significantly decreased COP displacement in both AP and ML directions [56, 58]. These results suggest that nGVS induces a sustained improvement in postural balance during standing with visual perturbation [59]. Application of nGVS also significantly reduced SaEn_ML, indicating increased regularity and predictability of COP trajectories. In the context of VR-induced sensory conflict, where visual-vestibular incongruence typically causes disorganized, high-complexity postural sway, this reduction signifies a shift toward more stable and controlled COP dynamics [35]. Critically, decreased SaEn_ML does not reflect maladaptive rigidity; rather, it aligns with our prior studies demonstrating that vestibular modulation reduces COP complexity during challenging sensory conditions, thereby enhancing postural control without compromising flexibility [33–35]. Furthermore, we observed concurrent reductions in hip muscle co-activation, as evidenced by lower %MVIC and CCI values for RF-BF pairs. These findings align with previous H-reflex studies reporting that nGVS reduces muscle tone and reflexive contraction in antigravity muscles, as reflected by decreased H-reflex amplitudes [60, 61]. Therefore, we propose that nGVS may mitigate visual-vestibular incongruence, which in turn downregulates motor neuron excitability, attenuates hip muscle co-activation, and ultimately diminishes reliance on the compensatory hip strategy [62]. This neuromuscular adjustment likely underlies the observed improvement in postural stability during VR-induced sensory conflict.

nGVS enhanced vestibular signal reliability to facilitate sensory reweighting and multisensory integration

According to our results, nGVS significantly improved postural stability and neuromuscular control during upright standing under VR perturbation. This improvement may be attributed to stochastic resonance induced by nGVS-delivered sub-threshold electrical noise, which optimizes the detection of weak vestibular signals by lowering VPT [63]. This mechanism enhances signal-to-noise ratio and temporal coherence of vestibular afferent firing without increasing baseline firing rates, thereby improving the reliability of vestibular inputs [21]. This nGVS-mediated improvement enables the CNS to dynamically reweight sensory contributions (up-weighting vestibular signals and down-weighting unreliable visual cues), which resolves the adverse effects of sensory conflict on postural stability, as evidenced by significantly reduced COP displacement and hip muscle co-activation under nGVS [16, 17]. The absence of nGVS effects in the eyes-closed condition further underscores that nGVS does not enhance the magnitude of vestibular inputs; rather, it optimizes multisensory integration specifically when visual-vestibular incongruence creates a demand for adaptive sensory reweighting [22, 23]. Consequently, the nGVS-mediated biomechanical enhancement likely reflects a shift away from maladaptive reliance on distorted visual feedback toward adaptive sensory reweighting that prioritizes reliable vestibular inputs for postural stability.

Previous neurophysiological studies suggest that vestibular afferents project to cortical regions involved in body orientation and vestibular processing, including SMG, parieto-insular vestibular cortex (PIVC), and cerebellar vermis [64]. Due to inherent limitations of the fNIRS, including limited penetration depth and probe placement constraints, it is infeasible to record hemodynamic responses from the insula and cerebellum [65]. We observed that nGVS significantly increased HbO concentration in the SMG, a key region for vestibular processing and self-motion perception [66, 67]. This nGVS-mediated heightened SMG activation appears to reflect a general enhancement of vestibular signal processing during sensory conflict rather than a postural control-specific mechanism [68]. Valdés et al. [69] reported increased SMG activation with nGVS in seated, healthy young adults, whereas Hernández-Román et al. [24] observed similar parietal-temporal activation in healthy participants receiving nGVS during a standing anterior–posterior sway task. This cross-paradigm consistency indicates that nGVS-induced higher SMG activation represents a fundamental cortical response to vestibular enhancement. Furthermore, two functional magnetic resonance imaging (fMRI) studies also demonstrated increased BOLD signals in the SMG in response to nGVS [70, 71]. Collectively, these findings support that SMG plays a critical role in vestibular signal processing, potentially refining the internal representation of self-motion and body orientation [64].

This refined vestibular representation from the SMG is subsequently integrated within frontoparietal networks, notably the DLPFC, which is essential in sensory reweighting and resolution of sensory conflict [26]. McCarthy et al. [27] demonstrated that transcranial alternating current stimulation on DLPFC modulates vestibular processing and attenuates visual-vestibular conflict. Consistent with DLPFC’s functional role, we observed a progressive increase in bilateral DLPFC activation across visual conditions (eyes-open to eyes-closed to VR) under sham stimulation, reflecting escalating cognitive demand as visual feedback transitioned from reliable to absent to incongruent with vestibular signals [72]. This gradient indicates that visual deprivation or distortion reduces available information for postural control, thereby increasing reliance on DLPFC-mediated sensory reweighting and multisensory integration [73]. Critically, nGVS significantly attenuated DLPFC hyperactivation under VR perturbation to a level comparable to the eyes-closed condition (VR + nGVS vs. EC + nGVS), demonstrating that nGVS primarily alleviates the excessive cognitive load imposed by VR-induced visual-vestibular incongruence. The absence of nGVS effects in the eyes-closed condition, aligning with Matsugi et al. [58], further supports that nGVS optimizes DLPFC-mediated sensory reweighting and multisensory integration primarily in sensory-conflicted environments. In contrast, cortical regions governing motor command generation and execution for postural stability, including M1, S1, and SMA/PMC, exhibit significantly higher activation under VR perturbation, paralleling progressive elevations in COP displacement and hip muscle co-activation [74]. However, there was no significant nGVS effect on these sensorimotor cortices, despite concurrent improvements in COP and sEMG metrics. This dissociation between biomechanical improvements and absent sensorimotor cortical modulation demonstrates that nGVS does not directly influence motor command generation for postural control [75].

Taken together, we propose that the beneficial effects of nGVS emerge upstream by enhancing vestibular signal reliability via stochastic resonance and facilitating sensory reweighting and multisensory integration within frontoparietal networks to resolve VR-induced visual-vestibular incongruence, with improved postural stability and muscle co-activation representing a downstream consequence.

Clinical and translational implications of context-dependent nGVS efficacy

The context-dependent nature of nGVS effects observed in this study, significant improvement specifically under VR-induced visual-vestibular conflict, reflects that nGVS-mediated enhancement of postural stability manifests most prominently when baseline signal reliability is compromised [3]. Among healthy young adults with intact sensorimotor function, low-challenge postural tasks (e.g., eyes-open/closed quiet standing) exhibit ceiling effects that mask subtle nGVS benefits; only when sensory conflict impairs postural control does vestibular enhancement by nGVS produce measurable improvements [21, 57].

Notably, VR-induced visual-vestibular incongruence replicates core conflict signatures encountered in multiple real-world scenarios: (1) microgravity adaptation during spaceflight where visual cues conflict with the absence of gravitational reference signals from the otolith organs [34, 76]; (2) motion sickness during vehicular travel while focusing on stationary objects (e.g., books or screens); and (3) cybersickness in immersive 3D environments [32]. These contexts share the common feature of visual-vestibular incongruence, precisely the condition modeled by our VR paradigm. Furthermore, patients with vestibulopathy or Parkinson's disease frequently present with vertigo, dizziness, and postural instability due to visual-vestibular conflict [18, 55]. nGVS has demonstrated therapeutic efficacy in the above-mentioned populations by enhancing vestibular inputs and postural stability, validating the translational relevance of our findings [30]. Consequently, while nGVS efficacy is context-dependent, this specificity precisely defines its clinical indication: mitigating postural instability under visual-vestibular incongruence.

Limitations

The present study has several limitations that should be acknowledged. First, the inclusion of only healthy young adults may restrict the generalizability of our findings. Future studies would benefit from a more diverse demographic range to investigate potential age-related or disease-related differences in response to nGVS [77]. Second, as the current analysis focused on fundamental sEMG and fNIRS parameters, the complexity of nGVS-induced neuromodulation may not be fully captured. More advanced sEMG and fNIRS metrics should be considered to gain deeper insights into the central-peripheral coupling. Additionally, sEMG cannot isolate TrA activity due to its deep anatomical location beneath IO. Although sensor placement minimized cross-talk from superficial rectus abdominis during quiet standing and has been validated in our prior work [37, 38], the recorded signal inevitably represents TrA/IO co-activation. However, it is unlikely to compromise the validity of our within-subject comparisons, as TrA and IO activate synergistically during upright stance. Future studies should employ fine-wire electrodes to achieve precise signal separation. Third, due to the inherent constraints of fNIRS, which primarily accessed the activation of cerebral cortex, neural responses in deeper brain regions associated with vestibular processing remained unexplored [78]. Future research should integrate complementary neuroimaging modalities, such as fMRI and electroencephalogram, to establish a more comprehensive understanding of the neural mechanisms underlying nGVS. Fourth, although Bonferroni correction was applied to all post-hoc pairwise comparisons, no secondary correction was performed across multiple independent ANOVAs, which may theoretically increase the family-wise error rate. Future studies should address this limitation by applying stricter correction methods across ANOVAs. Fifth, despite pre-screening exclusion of individuals with severe cybersickness susceptibility and within-subject crossover design, a moderate inter-individual variability in VR-induced cybersickness persisted; however, this residual variation, representing physiologically inherent diversity, is unlikely to confound our conclusions given the robust nGVS effects observed under VR perturbation. Finally, the VR goggles were worn only during VR trials, potentially introducing confounding effects across visual conditions. Although the goggles featured soft padding and adjustable straps to ensure comfortable fit, future studies should employ a sham condition, in which participants wear the goggles with the display turned off, to isolate visual perturbation effects from mechanical loading effects.

Conclusions

The present study revealed that nGVS effectively mitigated VR-induced sensory conflict and enhanced postural stability during upright standing. By promoting stochastic resonance, nGVS facilitated vestibular processing in SMG and reduced the demand for cognitive resources on DLPFC, leading to more efficient multisensory integration and reduced reliance on the hip strategy. These findings suggested the clinical potential of nGVS as a promising, non-invasive neuromodulation approach for improving postural stability in rehabilitation and aerospace medicine.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

ANOVA

Analysis of variance

AP

Anterior–posterior

BF

Biceps femoris

CCI

Co-contraction index

CNS

Central nervous system

COP

Center of pressure

DLPFC

Dorsolateral prefrontal cortex

EC

Eyes-closed

EO

Eyes-open

fNIRS

Functional near-infrared spectroscopy

GL

Gastrocnemius lateralis

HbO

Oxygenated hemoglobin

L₃

Paraspinal muscles at the third lumbar vertebral level

ML

Medial–lateral

MVIC

Maximum voluntary isometric contraction

M1

Primary motor cortex

nGVS

Noisy galvanic vestibular stimulation

PIVC

Parieto-insular vestibular cortex

PMC

Premotor cortex

RF

Rectus femoris

RMS

Root-mean-square

ROIs

Regions of interest

SaEn

Sample entropy

sEMG

Surface electromyography

SMA

Supplementary motor area

SMG

Supramarginal gyrus

S1

Primary sensory cortex

TA

Tibialis anterior

TrA/IO

Transversus abdominis/internal oblique complex

T₇

Paraspinal muscles at the seventh thoracic vertebral level

T₁₂

Paraspinal muscles at the twelfth thoracic vertebral level

VPT

Vestibular perceptual threshold

VR

Virtual reality

Author contributions

H.X.: Data Curation, Formal analysis, Funding acquisition, Writing—Original Draft, Writing—Review and Editing; Y.L.: Data Curation, Investigation, Validation, Writing—Review and Editing; Z.H.: Investigation, Methodology; L.Z.: Formal analysis, Methodology; J.H.C.: Resources, Software, Supervision, Writing—Review and Editing; C.W.: Conceptualization, Funding acquisition, Project administration, Writing—Review and Editing. All authors read and approved the final manuscript. H.X. and Y.L. contributed equally to this work.

Funding

This study was funded by the National Key Research and Development Program of China (2022YFC2009700), Guangdong-Hong Kong Technology and Innovation Cooperation Funding (2023A0505010014), National Natural Science Foundation of China (82172532), Guangdong Basic and Applied Basic Research Foundation (2023A1515110185), and 2026 Suzhou Municipal Hospital Ke Jiao Xing Wei Special Fund (Szslyyrc202601040).

Data availability

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study was approved by the Independent Ethics Committee for Clinical Research and Animal Trials of the First Affiliated Hospital of Sun Yat-sen University (IRB# [2023]793), and registered at the Chinese Clinical Trial Registry (ChiCTR# 2300078910). All participants provided informed consent before participating in the study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.