Brain Connectivity During Walking and Obstacle Avoidance in Persons With Multiple Sclerosis and Healthy Controls: A Pilot EEG Study

Abstract

This study investigated effective connectivity and hemispheric asymmetry in persons with multiple sclerosis (pwMS) compared to healthy controls (HC) during two walking conditions: walking alone and walking while avoiding unpredictable obstacles. Cognitive-motor interference (CMI) was analyzed using electroencephalography (EEG) across beta, alpha, and theta frequency bands. Directed functional connectivity was estimated using partial directed coherence (PDC) to assess differences in connectivity patterns between conditions and groups. In healthy controls, obstacle avoidance increased connectivity in motor and cognitive regions including left central (LC), left temporal (LT), and right frontal (RF) regions, p<0.0014. In contrast, pwMS demonstrated weaker and more localized connectivity, primarily in the left central regions (sensorimotor cortices) p<0.0013, suggesting reduced efficiency in brain networks and compensatory mechanisms to maintain task performance. Further, pwMS showed left laterality toward the central region during both walking conditions compared to HC, p<0.05. Correlational analysis revealed that connectivity during obstacle avoidance in HC positively correlated with comfortable walking speed (r = 0.57), indicating efficient neural pathways. In pwMS, connectivity showed a negative correlation with walking speed (r = −0.65), indicating compensatory but inefficient neural engagement. These findings highlight disruptions in brain connectivity during motor-cognitive tasks in pwMS, with potential implications for designing targeted rehabilitation strategies to improve gait and neural efficiency.

I. Introduction

Multiple sclerosis (MS) is a progressive neurodegenerative disease affecting nearly 1 million adults in the United States, with an annual economic burden exceeding $85 billion [1], [2]. Persons with multiple sclerosis (pwMS) often experience significant motor and cognitive impairments that reduce their quality of life and independence [3]. One of the most common challenges faced by pwMS is difficulty walking, particularly in complex environments that require simultaneous cognitive and motor coordination, such as avoiding obstacles. This difficulty arises from MS-induced damage to the central nervous system (CNS), which disrupts neural connectivity in motor-control regions [4].

Walking while performing a cognitive task, known as dual-tasking, creates cognitive-motor interference (CMI), which is particularly pronounced in pwMS. CMI arises when cognitive and motor processes compete for limited neural resources, resulting in impaired gait performance [5], [6].

Previous studies have shown pwMS exhibits higher cortical activation in motor and cognitive regions, such as the prefrontal cortex, to compensate for gait impairments [7], [8], [9] and increased cortical activation is observed during obstacle avoidance tasks relative to walking condition [10], [11]. Altered cortical connectivity due to CMI [12], [13], [14], [15] may result from competition for common neuronal resources during simultaneous dual tasks [16]. This heightened neural activity may indicate compensatory mechanisms or inefficient neural processing in pwMS when confronted with complex motor challenges. However, the exact neural connectivity disruptions in pwMS during real-time walking and obstacle avoidance remain unclear. While earlier studies have explored dual-task walking, few have examined how unpredictable obstacle avoidance impacts connectivity patterns in pwMS compared to healthy individuals. Understanding brain connectivity during obstacle avoidance is critical because such tasks simulate real-world challenges that pwMS face daily. Disrupted connectivity may lead to compensatory but inefficient neural strategies, which could impact gait and increase fall risk. Assessing walking performance during both unobstructed walking and virtual unpredictable obstacle avoidance provides critical insights into the neural mechanisms underlying these challenges. Virtual obstacle avoidance tasks simulate dynamic environments requiring rapid integration of visuospatial processing, motor planning, and cognitive control functions often disrupted in pwMS due to MS-related central nervous system damage. TABLE I summarizes the abbreviations used in this paper.

TABLE I.

Listed of Abbreviations Used in The Paper

Abbreviation	Description	Abbreviation	Description
HC	Healthy controls	LF	Left frontal
pwMS	Persons with multiple sclerosis	RF	Right frontal
MS	Multiple sclerosis	LC	Left central
HCW	Healthy control during walking alone	RC	Right central
HCO	Healthy control during obstacle	LT	Left temporal
pwMSW	Persons with multiple sclerosis during walking	RT	Right temporal
pwMSO	Persons with multiple sclerosis during obstacle	LPO	Left parietal-occipital
PFC	Prefrontal cortex	RPO	Right parietal-occipital
PDC	Partial directed coherence	LI	Laterality index

Electroencephalography (EEG), with its high temporal resolution, provides a powerful tool for examining brain connectivity during dynamic tasks such as obstacle avoidance while walking. In healthy adults, EEG studies have shown reduced alpha band power during complex movements suggesting a decrease in thalamic inhibition to enhance sensory input for motor coordination, and increased beta synchrony within the basal ganglia-cortical loop to regulate voluntary movement [17]. Additionally, EEG research has shown theta band desynchronization when walking patterns are disrupted by perturbations [18], [19], with recent findings indicating that theta band activity reflects sensory prediction errors during gait adaptation [20]. Mobile EEG studies have further advanced our understanding of neural dynamics during complex motor tasks in clinical populations, providing a foundation for investigating CMI in real-world settings. For example, studies by Mustile et al. [21], [22] and Stuart et al. [23], [24] demonstrated increased prefrontal and sensorimotor activation during obstacle avoidance in Parkinson’s disease and older adults, emphasizing the roles of theta and alpha band dynamics in visuospatial and motor planning. Similarly, Silva-Batista et al. [25] reported heightened cortical engagement during gait tasks in neurological cohorts, while Nordin et al. [26] identified theta band contributions to obstacle stepping in healthy individuals. However, these studies primarily focused on regional EEG activity or spectral features, and functional connectivity approaches, such as directed connectivity analysis have not been applied to map whole-brain network interactions in pwMS during unpredictable obstacle avoidance. Such an approach could reveal MS-specific disruptions in neural integration and hemispheric asymmetry underlying CMI-induced gait deficits. This study enhances previous research by examining directed connectivity, which quantifies interactions and information flow between neural networks [27], [28]. This approach provides new insights into network-level compensatory mechanisms and lays the groundwork for targeted neuromodulation strategies in MS rehabilitation. Analyzing functional connectivity is essential for understanding disease-specific patterns of network disruption and potential compensatory mechanisms in various neurological disorders, including MS.

Gait impairments are highly prevalent in individuals with multiple sclerosis (pwMS), creating an urgent need to understand the neural mechanisms underlying real-world tasks, such as unpredictable obstacle avoidance. This pilot study investigates how brain connectivity networks and hemispheric asymmetry differ between pwMS and healthy controls (HC) during two tasks: walking alone and walking with obstacle avoidance, using EEG. Specifically, we analyze directed connectivity networks and hemispheric laterality across the beta, alpha, and theta frequency bands to uncover brain network dynamics. We hypothesized that (H1) HC would exhibit high within-brain connectivity in obstacle avoidance versus single walking, and (H2) pwMS would exhibit weaker and more sparse connectivity networks compared to HC during both walking tasks. We also hypothesize that (H3) pwMS will exhibit lateralization in their connectivity networks due to a compensatory mechanism and functional dominance in cognitive-motor tasks and based on previous findings in our fNIRS-based functional connectivity data [11]. The findings from this study could inform future research on neuromodulation-based therapies and help predict the effects of motor training and rehabilitation strategies in pwMS.

II. Methodology

A. Participants

A total of twenty-one participants were recruited for this study, ten pwMS (51.5 ± 8 years old, 7 females, 3 males) and eleven HC’s (51.2 ± 6.4 years old, 8 females, 3 male). The inclusion criteria for MS participants are as follows: (a) a definitive diagnosis of MS; (b) relapse-free for the past 30 days; and (c) the ability to walk with or without a cane, but not with a walker or rollator. Exclusion criteria include major depression (if medication dosage has changed in the last year), schizophrenia, bipolar disorder, or substance abuse disorders. Participants should not be taking medications that can affect cognition or movement. For healthy controls (HC), we included individuals matched by age, gender, and education level, with no history of neurological conditions or motor impairments. The study received approval Kessler Foundation Institutional Review Board, and all participants provided written informed consent prior to the start of the study. Two HC and two pwMS participants dropped out of the study before data collection was completed. Therefore, the EEG data included in this manuscript are for 8 MS and 9 HC.

B. Experimental Procedure Condition

First, all participants practiced walking and avoiding virtual unpredictable obstacles on a treadmill at varying speeds to determine their comfortable walking speed, which was fixed and consistent across both conditions. Virtual obstacles, designed as red traffic cones (pylons), were projected onto the treadmill belt using a side-positioned projector, as shown in Fig. 1(a). Obstacles appeared at randomized intervals and locations along the continuously moving belt, mimicking real-world unpredictability. No two obstacles were presented simultaneously, ensuring participants avoided one obstacle at a time. The obstacle presentation was programmed to align with each participant’s walking speed, personalizing the task and maintaining ecological validity. Second, an EEG cap was placed over the head of the participants, and they had to walk while actively avoiding the virtual obstacles. The experimental protocol consisted of a 30-second walk followed by a 15-second walk with obstacle avoidance. This sequence was repeated at least 20 times throughout the experiment. The total duration varied depending on individual performance and fatigue levels, ranging from 15 to 30 minutes. The first 15 minutes of data were analyzed for all subjects. This design enabled evaluation of cognitive-motor interference and adaptive strategies under dynamic walking conditions.

Fig. 1.

EEG experimental setup and data acquisition, a) gait adaptation during obstacle avoidance, b) EEG brain regions of interest.

C. Data Acquisition

EEG data was collected while participants walked on an instrumented treadmill (C-MILL, Motekforce, The Netherlands). The EEG data was recorded using a 64-channel wireless ActiCap EEG system (Brain Products, Munich, Germany) equipped with active electrodes for dual-stage amplification, significantly reducing movement artifact noise. The electrode montage follows the 10–20 system locations. Participants were asked not to clench their jaws during the movement to avoid EEG artifacts. The EEG data was collected at sampling rate of 500Hz and FCz was chosen as the reference during data collection.

D. EEG Data Analysis

EEG data preprocessing was performed using the EEGLAB MATLAB toolbox [29], incorporating both Artifact Subspace Reconstruction (ASR) [30] and Independent Component Analysis (ICA) to ensure high data quality during dynamic walking tasks. ASR was applied with a threshold of k = 20 to identify and remove high-amplitude artifacts, such as those arising from motion and muscle activity, resulting in the removal of approximately 5–10% of data variance across participants. Channels identified as excessively noisy were interpolated using spherical splines to preserve spatial information. Following ASR, ICA was conducted to identify and remove components corresponding to eye blinks, muscle activity and walking-related artifacts. This ICA-based artifact rejection accounted for an additional 15–20% of total variance, in line with established protocols for EEG studies involving gait and complex motor activities [31]. This rigorous preprocessing pipeline ensured that the resulting EEG data accurately reflected neural activity relevant to the cognitive-motor tasks under investigation. The EEG data were then re-referenced to the average of all 64 channels. Epochs containing artifacts were manually removed. After preprocessing, the data were epoched to split walking condition data and obstacle avoidance data. The total number of epochs in each subject and condition was set to 52 epochs. Thus, we have a 3D matrix of size 64^∗1500^∗52 corresponding to the number of electrodes, segmented data points and number of epochs/trails. From the cleaned epochs, we estimated the directed functional connectivity networks (FCN) using partial directed coherence (PDC) between regions [32], [33]. The full mathematical equations for estimating the PDC are provided in the supplementary material. By calculating the impacts of all other signals, PDC avoids volume conduction and quantifies the relationship between pairs of N-electrode signals. This approach is based on multivariate autoregressive model estimation (MVAR). The PDC value ranges from 0 to 1. The PDC connectivity maps were calculated for every pair of EEG electrodes highlighted in Fig. 1(b) (58 electrodes) at three frequency bands: beta, alpha and theta. We then performed automated matrix binarization using the orthogonal minimum spanning trees (OMSTs) method [34] to select the true connections between the EEG electrodes. The resulting binarized PDC maps subsequently served as the features that were used for correlation analysis, laterality index, and statistical analysis. The final EEG connectivity networks were averaged with 4 subregions in each hemisphere, forming 8 regions of interest as shown in Fig. 1(b).

E. Correlation and Laterality Index

Bivariate correlation analysis assessed the relationship between brain connectivity networks and comfortable walking speed in HC and pwMS. This analysis was performed individually for each of the eight brain subregions of interest, as indicated by the highlighted electrodes in Fig 1(b). The lateralization of functional connectivity in these brain regions was evaluated using the laterality index (LI), calculated as LI = (L - R) / (L + R), where L and R represent the estimated connectivity values for the left and right hemispheres, respectively. The LI was estimated based on outflow (outward connectivity) values, such that a value of 1 for central electrodes means complete lateralization toward outflow of connectivity from left central electrodes and a value of −1 means complete lateralization toward outflow from right hemisphere electrodes. Hence, a positive LI value indicates left hemisphere dominance (left lateralization), while a negative value denotes right hemisphere dominance (right lateralization). Only LI distributions with mean values significantly different from zero, at a threshold of p < 0.05, were considered to demonstrate statistically significant lateralization toward either hemisphere. This approach provides a more detailed comprehension of hemispheric dominance in brain connectivity across various regions in both HC and pwMS groups.

F. Statistical Analysis

The features estimated by partial directed coherence (PDC) were tested for normality using the Kolmogorov-Smirnov test to ensure the appropriateness of parametric statistical methods. Following this, t-test analyses were performed to compare: 1) connectivity differences within and between pwMS and HC during walking and obstacle avoidance, 2) hemispheric laterality (using the Laterality Index, LI) and 3) correlations between brain connectivity network during obstacle avoidance and walking speed. A paired two-sample t-test was used to assess the significant differences between walking and obstacle avoidance within each group (HC and MS) (within-group comparisons). An unpaired two-sample t-test was conducted to evaluate the significant differences between HC and MS groups in each condition (between-group comparisons). The feature sizes for this analysis, in each group and condition, were averaged to a matrix size of 58^∗58^∗52. Lastly, a one-sample t-test was specifically employed to assess whether the LI significantly deviated from zero, indicating lateralization. Pearson correlation analysis was used to examine connectivity relationships with walking speed. A p-value of less than 0.01 (p < 0.01) was considered statistically significant following Keppel modified Bonferroni criteria [35], thus reducing the risk of Type I error and increasing the rigor of the statistical inference. This comprehensive approach allowed for a robust analysis of the connectivity differences between conditions and groups, as well as the evaluation of hemispheric lateralization.

III. Results and Analysis

A. Connectivity Results

Figures 2–4 present the maps of directed connectivity pairs that differ significantly between conditions and groups for the beta, alpha and theta frequency bands, respectively. The EEG artifacts free signals and the full maps of the functional connectivity networks covering the 58-electrodes for all frequency bands and conditions are summarized in the supplementary material. In these statistical maps, within a&b panels, red lines represent higher connectivity during walking alone, and blue lines indicate greater connectivity during obstacle avoidance while walking. Similarly, within c&d panels, red lines represent higher connectivity in HC group, and blue lines represent higher connectivity in pwMS group for each walking condition.

Fig. 2.

Effective connectivity network in beta frequency bands; (a)- the differences in connectivity networks in HC between walking alone and walking with obstacle avoidance (b)- the differences in connectivity networks in MS between walking alone and walking with obstacle avoidance; (c)- the differences in connectivity networks between HC and pwMS during walking alone and (d)- the differences in connectivity networks between HC and pwMS during walking with obstacle avoidance.

Fig. 4.

Effective connectivity network in theta frequency bands; (a)- the differences in connectivity networks in HC between walking alone and walking with obstacle avoidance (b)- the differences in connectivity networks in MS between walking alone and walking with obstacle avoidance; (c)- the differences in connectivity networks between HC and pwMS during walking alone and (d)- the differences in connectivity networks between HC and pwMS during walking with obstacle avoidance.

1). Beta Band (13–30 Hz):

In the HC group (Figure 2a), during walking alone, results show a higher connectivity network with information flows involving the right motor cortices and left frontal (LF), with the right central (RC) serving as a major hub (p < 0.0012; 2.86 < t < 3.37). Meanwhile, during walking with obstacle avoidance, the left motor cortices and right frontal regions exhibit stronger connectivity patterns with directed information flows, reflecting the recruitment of cognitive-motor networks to handle increased task complexity (p < 0.0013; −3.65 < t < −2.85). In contrast, pwMS (Figure 2b) demonstrated a more localized connectivity pattern during walking alone, with information flows primarily from the left frontal (LF) and left parietal-occipital (LPO) regions (p < 0.0014; 2.82 < t < 3.67). However, during walking and obstacle avoidance, the left central (LC) and right parietal-occipital (RPO) regions exhibit higher connectivity with altered information flow dynamics (p < 0.0013; −3.06 < t < −2.83).

Figures 2(c) and 2(d) illustrate the differences in connectivity between HC and pwMS across walking conditions. During walking alone (Figure 2c), HC exhibited a more distributed and efficient connectivity network, with robust information flows across all cortical regions. The right motor cortex and left parietal-occipital (LPO) region emerged as major hubs for information integration, facilitating efficient motor-cognitive processing (p < 0.0001; 6.06 < t < 10.92). In contrast, pwMS demonstrated a less efficient network, with weaker and more localized connectivity patterns, particularly within the bilateral frontal cortices, where information flow was primarily directed toward the motor cortices, suggesting a compensatory but inefficient neural strategy (p < 0.0001; −13.5 < t < −6.05).

During obstacle avoidance (Figure 2d), HC maintained a well-balanced inter-hemispheric connectivity network, with strong connectivity and bidirectional information flow between central, temporal, and parieto-occipital regions, (p < 0.0001; 6.02 < t < 11.31). Conversely, pwMS exhibited a further reduction in connectivity, with most connections restricted to the right hemisphere. Information flows were predominantly directed from the right frontal (RF) region toward the right motor cortex, reflecting a more constrained and compensatory activation pattern (p < 0.0001; −13.34 < t < −6.80). The node densities further quantified these changes in all frequency bands (see supplementary material).

2). Alpha Band (8–13 Hz):

In the HC group (Figure 3a), connectivity patterns in the alpha band during walking alone closely resembled those observed in the beta band, with an increase in major hubs extending to the right frontal (RF) region, facilitating enhanced motor-cognitive integration (p < 0.0011; 2.87 < t < 3.57). Meanwhile, during walking with obstacle avoidance, connectivity patterns remained consistent with those in the beta band, exhibiting strong inter-hemispheric information flows that reflect efficient neural coordination for visuomotor integration (p < 0.001; −3.55 < t < −2.93). pwMS (Figure 3b) also demonstrated connectivity patterns similar to those in the beta band during walking alone, but with an extended network covering the RF region and inter-central connections (p < 0.0011; 2.96 < t < 3.60). However, during walking with obstacle avoidance, connectivity became more restricted, with information flow primarily limited to the left central (LC) region (p < 0.0014; −3.49 < t < −2.86).

Fig. 3.

Effective connectivity network in alpha frequency bands; (a)- the differences in connectivity networks in HC between walking alone and walking with obstacle avoidance (b)- the differences in connectivity networks in MS between walking alone and walking with obstacle avoidance; (c)- the differences in connectivity networks between HC and pwMS during walking alone and (d)- the differences in connectivity networks between HC and pwMS during walking with obstacle avoidance.

Figures 3(c) and 3(d) illustrate the differences in connectivity between HC and pwMS across walking conditions. In HC, connectivity in the alpha band during both walking conditions was less distributed than in the beta band, though still exhibiting robust information flows from the right to left hemisphere, highlighting task-related inter-hemispheric coordination (p < 0.0001; 6.03 < t < 9.05 for walking alone, p < 0.0001; 6.14 < t < 9.73 for obstacle avoidance). Meanwhile, pwMS displayed nearly identical connectivity patterns to those observed in the beta band, with similar statistical values indicating a persistent compensatory but inefficient network (p < 0.0001; −12.78 < t < −6.0 for walking alone, p < 0.0001; −11.71 < t < −6.56 for obstacle avoidance).

3). Theta Band (4–7 Hz):

In the HC group (Figure 4a), walking alone showed localized connectivity in the left frontal (LF) and right temporal (RT) regions (p < 0.0014; 2.82 < t < 3.28), while obstacle avoidance engaged left central (LC) and right frontal (RF) as major hubs, supporting adaptive motor control (p < 0.0014; −3.56 < t < −2.81). In contrast, pwMS (Figure 4b) exhibited theta-band connectivity patterns similar to the alpha band during walking alone (p < 0.001; 2.92 < t < 3.96) and resembled beta-band connectivity during obstacle avoidance (p < 0.0014; −2.93 < t < −2.81). Figures 4(c) and 4(d) highlight group differences. In HC, theta-band connectivity was less distributed than in the alpha band, but still showed strong inter-hemispheric information flows, supporting task-related coordination (p < 0.0001; −9.98 < t < −6.72 for walking alone, p < 0.0001; 6.05 < t < 10.18 for obstacle avoidance). pwMS exhibited reduced connectivity, with bilateral frontal-to-central connectivity during walking alone (p < 0.0001; −10.41 < t < −6.06) and right frontal-to-central connectivity during obstacle avoidance (p < 0.0001; −9.98 < t < −6.72). The restricted connectivity in pwMS suggests deficits in adaptive control mechanisms, potentially impairing gait performance and increasing fall risk.

B. Laterality Results

The laterality index analysis consistently revealed distinct patterns between HC and pwMS during walking and obstacle avoidance across four brain regions and three frequency bands, as shown in Fig. 5. In HC, there was a consistent trend of left hemisphere dominance (laterality) in the frontal, central, and parietal-occipital regions, along with right hemisphere dominance in the temporal cortex across all frequency bands. However, these patterns did not achieve statistical significance for either walking condition. Conversely, pwMS exhibited a different laterality profile. The pwMS group showed left hemisphere dominance in the frontal, temporal, and central regions, with right hemisphere dominance in the parietal-occipital region across all frequency bands and walking conditions. Notably, pwMS demonstrated significant left lateralization in the central cortex, with p-values of 0.0316, 0.0330, and 0.0325 during walking alone and p-values of 0.0254, 0.0269 and 0.0241 during obstacle avoidance for the three frequency bands (beta, alpha and theta), respectively.

Fig. 5.

Laterality Index for HC and pwMS during walking alone and walking with obstacle for beta, alpha and theta frequency bands at four brain regions.

C. Correlation Between Connectivity Networks and Comfortable Walking Speed

TABLE II presents the results of the correlation between brain connectivity networks during obstacle avoidance and comfortable walking speed in HC and pwMS, with correlation values (r) and p-values reported across three frequency bands at different brain regions of interest (ROI). In the beta band, the HC group demonstrated moderate positive correlations, particularly in the RC, LC, and RPO regions, with r-values of 0.573, 0.444, and 0.385, respectively. Conversely, the pwMS group exhibited moderate negative correlations in regions such as the LC, RT, and RC, with r-values of −0.657, −0.597, and −0.472, respectively, while showing a positive correlation in the RF region (r = 0.523). Effect sizes (Hedges’ g) comparing HC and pwMS ranged from moderate to large, with the most significant effect sizes observed in the RC (Hedges’ g =0.998) and LC (Hedges’ g =0.918), indicating substantial differences in connectivity related to walking speed between the two groups (HC speed: 1.5±0.30; MS speed: 1±0.50). Similar patterns were observed in the alpha and theta frequency bands, with higher effect sizes at the central cortex as summarized in TABLE II.

TABLE II.

Correlation Analysis and Effect Size Between Brain Connectivity Network and Comfortable Walking Speed at Eight-Brain Regions

		HC		pwMS		Effect size (HC vs MS)
Beta	ROI	r	P-value	R	P-value	(Hedges’ g)
	LF	0.001	0.997	−0.291	0.484	−0.550
	RF	−0.087	0.824	0.523	0.184	−0.332
	LT	−0.192	0.622	−0.379	0.354	0.161
	RT	0.008	0.983	−0.597	0.118	0.572
	LC	0.444	0.232	−0.657	0.077	0.918
	RC	0.573	0.107	−0.472	0.237	0.998
	LPO	−0.235	0.543	0.378	0.356	0.330
	RPO	0.385	0.306	−0.161	0.703	−0.379
Alpha	LF	−0.018	0.964	−0.312	0.452	−0.610
	RF	−0.086	0.825	0.523	0.183	−0.397
	LT	−0.196	0.613	−0.415	0.306	0.149
	RT	−0.010	0.981	−0.634	0.091	0.534
	LC	0.464	0.208	−0.657	0.076	0.891
	RC	0.554	0.122	−0.479	0.230	1.005
	LPO	−0.242	0.530	0.336	0.416	0.333
	RPO	0.349	0.358	−0.222	0.598	−0.407
Theta	LF	−0.058	0.881	−0.229	0.472	−0.630
	RF	−0.100	0.798	0.519	0.187	−0.360
	LT	−0.164	0.673	−0.342	0.407	0.106
	RT	−0.008	0.983	−0.601	0.115	0.519
	LC	0.435	0.242	−0.546	0.162	0.756
	RC	0.594	0.092	−0.448	0.265	0.911
	LPO	−0.304	0.426	0.360	0.381	0.245
	RPO	0.341	0.370	−0.181	0.668	−0.462

Overall, the correlation analysis suggests that the relationship between connectivity networks and comfortable walking speed is more variable and potentially disrupted in pwMS than HC. This potential disruption in the relationship between connectivity networks and walking speed in pwMS is a significant finding of our research.

IV. Discussion

This study provides novel insights into the neural mechanisms underlying the cortical control of obstacle avoidance during walking in pwMS versus healthy controls. By examining directed connectivity patterns and hemispheric asymmetry in pwMS and HC during walking alone and obstacle avoidance, this study 1) demonstrates the neural correlates of obstacle avoidance and walking in healthy adults, and 2) highlights critical disruptions in cortical control of walking and obstacle avoidance in pwMS.

To the best of our knowledge, this is the first study to investigate effective brain connectivity during real-time obstacle avoidance in pwMS using EEG, providing a direct neural assessment of motor-cognitive integration deficits. These findings support hypotheses H1 and H2, which suggested more connectivity in obstacle avoidance versus single walking in HC and more sparse connectivity in pwMS compared to HC in both conditions, and hypothesis H3 which suggested a compensatory mechanism that will be reflected in lateralization in connectivity networks. In addition to these hypotheses, our findings offer a foundation for a better understanding of altered brain connectivity in pwMS.

A. Connectivity Patterns and Network Efficiency

Between conditions:

In HC, directed connectivity analysis revealed robust and widespread connectivity networks during obstacle avoidance, particularly in the left central (sensorimotor), temporal, and right frontal regions across beta and alpha frequency bands compared to walking alone. The theta band exhibited increase connectivity from the left central (sensorimotor) and right frontal to the parietal-occipital region, see Fig.4(a). These findings suggest that HC efficiently recruits motor and cognitive networks to meet the increased task demands of navigating obstacles, reflecting enhanced cognitive-motor integration.

In contrast, pwMS showed weaker and more sparse brain connectivity networks, particularly in motor-related areas, during both walking conditions. The lower or more diffuse connectivity in pwMS suggests a limit in the cognitive resources available to respond to cognitive load in dual-tasks [36]. Notably, the persistently sparse connectivity observed across all frequency bands in pwMS suggests that compensatory neural recruitment mechanisms are activated to support motor control.

Our results support our hypothesis and are consistent with earlier studies showing changes in neural activity in individuals with multiple sclerosis when performing motor tasks [11]. These findings suggest that MS-related damage may limit access to cognitive resources that are needed for dual-tasking, which hinders the smooth integration of motor and cognitive functions necessary for walking and avoiding obstacles, leading to inefficient compensatory strategies.

Between groups:

During walking alone, pwMS showed increased connectivity from the frontal regions (left frontal and right frontal) toward central and temporal regions, especially in the beta and alpha bands. However, these connections were less distributed than those in HC, which showed stronger and more widespread connectivity across central, temporal, and parieto-occipital regions. During obstacle avoidance, pwMS relied heavily on intra-hemispheric, right hemisphere pathways, specifically from the right frontal to right central and right temporal regions. This contrasts with HC, who engaged in more balanced and widespread networks across both hemispheres. Overall, these findings suggest that pwMS experience reduced network efficiency and rely on compensatory neural pathways, particularly in frontal-to-central regions, to manage increased task complexity. Previous studies on pwMS have reported higher cortical activation across various regions during dual-tasks paradiagms [8], [10], [37], [38], [39], [40], [41]. However, these studies did not employ EEG-based connectivity analysis to examine interactions among these regions during dual-tasking with obstacles avoidance or unobstructed walking, limiting the direct comparisons with our findings. Our study builds on mobile EEG research in Parkinson’s disease and aging populations, which showed heightened prefrontal and sensorimotor activation during complex gait tasks [21], [22], [23], [24]. Unlike those studies, which primarily analyzed prefrontal dynamics or single-region activity, our work is the first to use directed connectivity analysis to map whole-brain network interactions in pwMS during real-time obstacle avoidance. Nordin et al. [26] observed theta band dynamics in HC during obstacle stepping, corroborating our findings of theta band involvement in visuospatial processing. Notably, theta band dynamics appear generalizable across populations, reflecting shared neural mechanisms for motor-cognitive integration in complex tasks. However, our focus on MS-specific connectivity disruptions and hemispheric asymmetry reveals unique network-level compensatory mechanisms in pwMS, distinguishing our findings from those in other cohorts. These insights highlight the novel contribution of our approach in identifying MS-specific neural deficits, paving the way for targeted neuromodulation strategies in MS rehabilitation and informing broader understanding of motor-cognitive interference across neurological conditions.

The preliminary findings of our correlational analysis revealed differences between groups in the relationship between connectivity and walking speed. In HC group, stronger connectivity in key regions (e.g., right central and parietal-occipital) was positively correlated with walking speed (e.g., r=0.573 for RC in beta band), indicating efficient motor performance. In contrast, pwMS showed negative correlations, especially in the left central and right temporal regions (e.g., r=−0.657 for LC in beta band), suggesting that increased connectivity may denote inefficient compensatory mechanisms rather than improved performance. This indicates disrupted neural integration and the need for interventions that enhance network efficiency. Nevertheless, given the small sample size in this study (8 pwMS, 9 HC), these results require validation in larger cohorts to confirm their generalizability and clinical relevance. Future studies with increased statistical power will clarify the predictive value of these connectivity patterns.

B. Frequency Band Specificity

Our analysis across beta, alpha and theta frequency bands revealed task- and group- specific patterns. In HC, both the beta and alpha bands showed widespread connectivity increases during obstacle avoidance, particularly involving motor (central) and cognitive (frontal, temporal) regions. This suggests an efficiently integrated network supporting motor-cognitive coordination. In contrast, the theta band displayed a more localized connectivity pattern, although it still maintained strong inter-hemispheric information flow, particularly from the left central (LC) and right frontal (RF) toward the parieto-occipital regions, indicating its role in visuospatial processing. In pwMS, connectivity patterns were less distinct, and more localized with consistent increases in the left central region across all three bands. The beta band, which is typically associated with motor planning and execution, showed compensatory increases that likely reflect heightened cognitive resource allocation to support motor deficits. However, theta-band connectivity remained limited, suggesting deficits in visuospatial integration and adaptive control during dual-tasking which requires gait adaptation [20]. These findings underscore the frequency-specific roles of brain rhythms and the impact of MS-related damage on neural efficiency, particularly in tasks requiring simultaneous motor and cognitive engagement.

C. Hemispheric Asymmetry and Compensatory Strategies

The laterality analysis revealed distinct hemispheric patterns between groups. In HC group, no significant lateralization was observed while performing motor-cognitive tasks, indicating a well-integrated and balanced network. In contrast, pwMS demonstrated leftward lateralization in outflow of connectivity from central cortex across the beta, alpha and theta bands, particularly during unpredictable obstacle avoidance. This leftward dominance likely reflects a compensatory mechanism to address motor control deficits by over-recruiting sensorimotor regions. In addition, obstacle avoidance in pwMS may involve verbal mediation as a compensation strategy. Verbal mediation refers to the internal use of language-based processes to guide behavior, particularly in complex motor and visuospatial tasks [42]. In pwMS, deficits in visuospatial processing and motor planning can lead to increased reliance on self-directed verbalization to navigate obstacles. This aligns with previous research showing that individuals with impaired visuospatial perception can enhance their task performance by engaging verbal working memory networks [43]. The consistent leftward dominance may also indicate cognitive overload and less efficient neural processing, increasing the risk of gait impairments and falls during complex tasks. These findings are consistent with previous studies reporting the leftward lateralization of attention networks in pwMS [44]. Lateralization toward the left hemisphere may also be associated with the resilience or adaptability of the sensorimotor cortex by shifting to lateralized activity to account for complex visuomotor tasks [44].

D. Clinical Implications and Rehabilitation Strategies

The findings of this study suggest potential targets for rehabilitation interventions in pwMS. Interventions focusing on dual-task training that combine motor and cognitive challenges could help strengthen weakened networks and improve gait performance. The distinct connectivity patterns observed during obstacle avoidance tasks highlight the need to incorporate task-specific rehabilitation strategies that simulate real-world challenges. Additionally, interventions such as non-invasive brain stimulation (e.g., transcranial electrical stimulation) [45], [46], [47] and neurofeedback-based training [48] could help normalize altered connectivity and address leftward lateralization in the sensorimotor cortex. By targeting these compensatory patterns, rehabilitation approaches may improve motor-cognitive integration and reduce gait impairment in pwMS.

E. Limitations and Future Directions

While our study provides valuable insights into the neural mechanisms involved in gait adaptation for healthy controls and persons with multiple sclerosis (pwMS), it is important to acknowledge its limitations. First, the small sample size limits the generalizability of our findings, highlighting the need for future studies with larger cohorts to validate the observed connectivity and laterality patterns. Second, although EEG provides excellent temporal resolution, it has low spatial resolution, particularly at the channel level. We did not perform source localization in this study due to the quality of the EEG data collected during walking tasks. Future research should involve more experimental trials, larger samples, and improved EEG quality regarding movement artifacts. These improvements will facilitate better source localization and enhance the spatial resolution of network analyses. Additionally, conducting longitudinal studies will help clarify how connectivity patterns evolve over the progression of multiple sclerosis and in response to various interventions.

V. Conclusion

This study shows altered brain connectivity in healthy controls (HC) and pwMS during complex walking tasks. pwMS exhibited weaker connectivity and a significant decrease in node densities compared to HC. While HC showed a positive correlation between connectivity and walking speed, pwMS had a negative correlation, indicating less efficient neural activity. These findings suggest that EEG connectivity networks could help understand multiple sclerosis and inform personalized interventions. Future research should integrate EEG with other neuroimaging techniques to validate these results and improve MS treatments.