The Effects of Immersion and Visuo-Tactile Stimulation on Motor Imagery in Stroke Patients are Related to the Sense of Ownership

Visual guided motor imagery (MI) is commonly used in stroke rehabilitation, eliciting event-related desynchronization (ERD) in EEG. Previous studies found that immersion level and visuo-tactile stimulation could modulate ERD during visual guided MI, and both of two factors could also improve sense of ownership (SOO) over target limb (or body). Additionally, the relationship was also reported between the performance of MI and SOO. This study aims to investigate whether immersion and visuo-tactile stimulation affect visual guided MI through the SOO over virtual body in stroke patients. Nineteen stroke patients were recruited. The experiment included two phases (i.e., SOO induction and visual guided MI with SOO) that was manipulated across four conditions in a within-subject design: ${2}\times {2}$ , i.e., immersion (VR, 2D monitor display) $\times $ multisensory stimulation (visuo-tactile stimulation, observation without tactile stimulation). Results found peaks ERD amplitude during MI were significantly higher in stronger SOO conditions than weaker SOO conditions. Interestingly, the ERD during visual guided MI under the condition of vision only in VR and visuo-tactile stimulation in 2D monitor are similar, which indicates that SOO may be an important factor behind this phenomenon (due to the similar SOO between these two conditions). A moderate correlation was also found between SOO scores and peaks ERD amplitude during MI. This study discussed the possible factor underlying the effects of immersion and multisensory stimulation on visual guided MI in post-stroke patients, identifying the effect of SOO in this process, and could be referred in future studies for coming up with better MI paradigms for stroke rehabilitation.

be an important factor behind this phenomenon (due to the similar SOO between these two conditions).A moderate correlation was also found between SOO scores and peaks ERD amplitude during MI.This study discussed the possible factor underlying the effects of immersion and multisensory stimulation on visual guided MI in post-stroke patients, identifying the effect of SOO in this process, and could be referred in future studies for coming up with better MI paradigms for stroke rehabilitation.Index Terms-ERD, motor imagery, sense of ownership, stroke rehabilitation, virtual reality.

I. INTRODUCTION
M OTOR imagery (MI) is a mental practice of a motor task without execution in reality, and is commonly used in clinical stroke rehabilitation [1], [2], [3], [4].MI elicits event-related desynchronization (ERD), a decrease in oscillatory brain activity in α band (8∼13Hz) and β band (13∼30Hz) [5], which can be detected by electroencephalograph (EEG).Traditional MI-based rehabilitation is usually instructed by therapists to require patients with stroke to imagine simple repetitive movements.However, it is difficult to imagine the training movements vividly and accurately through verbal instructions [6].Action observation and visual guided training, such as traditional mirror therapy [7], mirror therapy in virtual reality (VR) [8], visual guided by robot [9], full body observation [10], etc., have been demonstrated to achieve a better MI training for stroke rehabilitation [11], and the enhanced MI training can be further improved by some methods: improved immersion and multisensory stimulations.
Some studies have reported better effects of VR device (e.g., head-mounted displays, HMD) than 2D monitor display on enhancing visual guided MI training, which proved by improved rhythmic patterns (e.g., higher ERD amplitude ratio in sensorimotor cortex) [12].This suggests that a higher immersion level could induce a stronger brain activity in visual guided MI training.Individual feeling presence in a specific immersive virtual environment (VE) designed for rehabilitation is the unique advantage of VR rehabilitation [13], and virtual body (i.e., the avatar which representing user self) is also presented in recent motor rehabilitation studies [10], [14], [15], [16].Sollfrank et al. [17] indicated that virtual avatar could be employed to represent the real body in immersive VR, providing end-users with a realistic 3D presentation of limb movements, which seems to help to get a concrete feeling of kinesthetic MI and exerts significant effects on motor cortex activation.These studies considered that action observation for MI training relies on presenting simulated virtual environments, the virtual body should be used naturally and should be anchored into immersive virtual world.The sense of ownership (SOO, i.e., the feeling that a body or a limb belongs to himself/herself, which also plays an important role in motor rehabilitation [18]) represents the extent to how individual believes the virtual body (which is interacting with and executing rehabilitation tasks in VE) is own body.Put simply, SOO is directly related to the rehabilitation target (affected limbs or body) in VR rehabilitation instead of indirect VE.
Furthermore, a recent review indicates that SOO is a modulator variable for limbs rehabilitation after stroke in VR [19].The relationship was also reported between the performance of MI and the SOO [20], [21], [22].However, the reason of immersion's advantage is still not completely clear.Although previous studies proved that immersion could improve body illusion [23], [24], [25], whether the SOO over virtual body is an important factor behind the immersive advantage need to be further confirmed.
Apart from high immersion of VR-based paradigms, the multisensory stimulation is also an effective approach to improve visual guided MI.Previous research found that visuo-tactile stimulation could significantly enhance visual guided MI in healthy participants [26], [27], [28], comparing to the condition without synchronous visuo-tactile stimulation [29].It is worth noting that the multisensory stimulation would form or improve SOO over target limb (or body), which is carried out by several illusion paradigms, including rubber hand illusion (RHI) [30], virtual hand illusion (VHI) [31], and full body illusion [32].Hence SOO over the target body is also probably a factor of the MI enhancement behind the multisensory stimulation.
Based on above studies (VR enhances MI and visuo-tactile stimulation enhances MI), Du et al. [33] have proposed the paradigm that combination of VR (HMD device) and visuo-tactile stimulation could enhance MI in healthy subjects.Simultaneously, previous studies found that SOO induced by visuo-tactile stimulus can be applied in stroke patients [34], [35], [36], which implies the feasibility of using SOO enhancing MI in stroke rehabilitation.According to the degree of MI enhancements among these paradigms, and the similar activation of common brain neural networks between SOO and MI in the sensorimotor cortex [37], we speculated that immersion and visuo-tactile stimulation may influence visual guided MI through SOO over virtual body.To sum up, this study aims to explicitly confirm the role of SOO in visual guided MI training enhanced by immersion and multisensory stimulation, and the hypotheses of this study are twofold: 1) SOO can be modulated by combination of immersion and multisensory stimulation, and influences the visual guided MI, i.e., SOO is an important factor of modulating visual guided MI, and the SOO level will be positively correlated with the ERD amplitude during MI, similar SOO would lead to similar ERD whether the level of immersion and multisensory stimulation changes during visual guided MI; 2) The combination of visuo-tactile stimulation and virtual reality could be applied to enhance visual guided MI training in stroke patients.This study could be referred in future studies for coming up with better MI paradigms for stroke patients.

A. Participants
Nineteen individuals who had suffered from stroke (6 females, 13 males, mean age 61.05±11.03years, 9 patients had left hemisphere damage, and 10 patients had right hemisphere damage, Brumstrom of affected upper limb ranges from I to V) were recruited from the Rehabilitation Department of Binzhou Medical University Affiliated Hospital for this study.The inclusion criteria were: (1) under the age of 80 years; (2) diagnosed with stroke through computerized tomography (CT) or magnetic resonance imaging (MRI); (3) first stroke (happened more than one month ago, i.e., chronic stage) and with unilateral hemiplegia; (4) absence of severe cognitive impairment or a history of psychiatric illness; and (5) capable of understanding the questionnaire questions and rating standards accurately, and performing the proprioceptive drift test conducted in this study.All patients understood the experimental procedures, operations, and possible situations.This work has been carried out in accordance with The Code of Ethics of the World Medical Association (Declaration of Helsinki) for experiments involving humans, and was approved by the local ethics committee, Beihang University (BM20200064).

B. Equipment and Scene
The EEG signal acquisition equipment used in this study is the NeuSen W wireless EEG acquisition system (Neuracle, Changzhou, Jiangsu province, China) and the NeuSen W 32-channel electrode cap.The electrode positions of the NeuSen W series are arranged in accordance with the international 10-20 system, and the signals of 28 channels (FP1, FP2, F7, F3, FZ, F4, F8, FC3, FCZ, FC4, T7, C3, Cz, C4, T8, M1, CP3, CPZ, CP4, M2, P7, P3, PZ, P4, P8, O1, OZ, O2) were recorded (Figure 1d).Among them, C3 and C4 channels are closed to sensorimotor area, which are related with MI.All electrodes were referenced with a full-head average.The EEG signal acquisition software (provided by Neuracle) was used to record the raw data at a 1000Hz sampling rate, the magnification is set to 1000, the bandwidth is 0.5-100Hz, and the impedance of each channel is less than 5K during the experiment.
The experiment was conducted within a room of approximately 20m 2 , in which two VIVE infrared locators were arranged at opposing diagonal corners, forming an VR activity area of approximately 15m 2 .To display the VR scene, a headmounted display (HMD) was employed (Valve Index, Valve Corporation, Seattle, State of Washington, USA), which boasts a monocular resolution of 1440 * 1600 pixels, a horizontal visual range of 130 degrees, and a refresh rate of 144Hz.
Tactile stimulation was administrated by means of a vibrotactile devices fastened on back of the hand of the participant and controlled by an Arduino uno R3 (Arduino srl., Ivrea, Italy, https://www.arduino.cc/)board connected to the computer with a USB cable (Figure 1b).The vibrotactile device vibrated for 100ms at a frequency of 150Hz [33].Tactile feedback was activated once when Arduino board received a signal sent by the C# script in Unity3D through USB cable.
The scenario in HMD is a virtual room that simulates the laboratory environment, with a table 2 * 1 * 1(m) both in virtual and real world.The virtual body (matched with subject gender) sits on the sofa in front of the table, and the hands are naturally placed on the virtual table.The subject enters the scene and sees the virtual body in first-person perspective (Figure 1a).There is also a horizontally placed 2D monitor display, and a pair of hands are shown in that (Figure 1b).Whether it is an HMD scene or a 2D monitor display scene, the participants cannot see their real hands, but only the virtual hands.The VR scene was created using Unity 3D (developed by Unity Technologies of San Francisco, CA) and the SteamVR SDK.

C. Experiment Design and Measurement
The experiment was manipulated through two factors (i.e., the immersion and multisensory stimulation) and proceeded in two steps: 1) SOO induction and 2) MI with SOO.Each participant underwent all four conditions with a random order.Four conditions were visuo-tactile stimulation in HMD (HMD-VT), observations without tactile stimulation in HMD (HMD-V), visuo-tactile stimulation in 2D monitor display (HD-VT), observations without tactile stimulation in 2D monitor display (HD-V).
We manipulated these two factors by combining virtual reality and visuo-tactile stimulation, which was also used in previous study for MI enhancing in healthy participant [33].The visuo-tactile stimulation method referred to the previous illusion studies (e.g., RHI [30], VHI [31], and full body illusion [32], etc.).The first step was the SOO induction stage.In HMD-VT and HD-VT conditions, the red virtual ball rolls from the side of the virtual table to the virtual left/right hand and touches virtual hand, at this time, the subject's task hand (affected) can also feel the vibration device on the back of the hand synchronously vibrating [38].In HMD-V and HD-V conditions, the subjects just observed virtual hand without ball touching.This step lasts for 2 minutes, and the subject was asked to score the SOO ("I felt as if the virtual hand is my own hand") through the Likert scale (seven levels ranging from −3 to 3, −3 indicating strongly disagree, 3 indicating strongly agree, and 0 indicating uncertain).The second step was the MI stage, EEG signals were recorded.This step included 30 trials, each trial (last for 10 seconds) consisted of a task phase (3 seconds) and a rest phase (7 seconds).In task phase, the virtual hand (affected side) begins to perform wrist flexion and extension movements, and the subject was required to simultaneously imagine that their real hand performs the same movement, but cannot produce the actual execution movement (Figure 1c).In rest phase, the movements stopped and virtual hand remained stationary.
The virtual scene was presented by the HMD device (HMD-VT and HMD-V) and 2D monitor display (HD-VT and HD-V) respectively.The flowchart of whole experiment is shown in Figure 1e.

D. Data Processing and Analysis
The Matlab 2021b and EEGLAB toolbox was used to process the EEG signals, and the questionnaire scores were investigated by asking participants verbally.
1) Pre-Processing: EEG data were first filtered from 0.1Hz to 40Hz, in order to remove power interference and other noise while retaining signals in α (8∼13Hz) and β band (13∼30Hz).Wrist flexion and extension of virtual hand was defined as the event of aligning the EEG signals across trials.Around 28-32 epochs (6-s epochs, ranging from −2s to 4s around the 28-32 events of MI task, which were marked in the EEG signal acquisition software through trigger from COM in computer.The marking signals were from virtual hand movement event in Unity3D, sent by C# scripts which controlling the virtual hand) were extracted from entire piece of EEG data of each participant in each condition.In order to identify and remove artifacts in the EEG signals, trial rejection was conducted [39]: For a given trial, a channel was marked if data points were obviously abnormal in comparison to mean channel data.If a trial contained more than 3 marked channels (10% of all channels), the trial was excluded from the analysis.Independent Component Analysis (ICA) was also applied to each epoch for removing noises from other origins, mainly including muscles nearby and electrooculography during MI.
2) ERSP and Topographical Maps: The ERD during MI training is mainly in α band (8∼13Hz) and β band (13∼30Hz).Therefore, the analysis was focused on these two rhythms.Due to the positions of C3 (left hemisphere) and C4 (right hemisphere) electrodes were closed to the SMC, PMC, the data of these channels was short-term Fourier transformed in a time window of 0.3s, extracting data of 40 frequency bands ranging from 1Hz to 40Hz.In order to visualize the effects of SOO on ERD during visual guided MI, the baseline normalized event-related spectral perturbation (ERSP, visualization of power in all time and frequency simultaneously) was conducted.According to the ERSP maps, specific frequency ranges were found and topographic distribution (visualization of power in spatial distribution) was also conducted.
3) Data Analysis of Questionnaire Scores and Peaks of ERD: The Shapiro-Wilk normality test was performed on the peak ERD amplitudes.If the data is normally distributed, a 2 × 2 repeated measures ANOVA and paired sample t-tests would be performed with two factors: multisensory stimulation (visuo-tactile or vision only) and immersion (VR-HMD or 2D monitor).If the data does not meet the normal distribution, Scheirer-Ray-Hare test and Wilcoxon matched-pairs signed-rank (two-sample) test would be performed on the two independent variables separately.Due to questionnaire scores were discrete data, Scheirer-Ray-Hare test and Wilcoxon matched-pairs signed-rank (two-sample) test (non-parametric test) were performed.Additionally, Spearman correlation analysis was conducted between SOO scores and peaks ERD amplitude.To reduce the risk of false positives, a Bonferroni correction was used to adjust the significance level α, with p < 0.05/2 (i.e., p < 0.025) used as the critical value for the level of statistical significance.All statistical tests were performed by using SPSS 26.0 (IBM Corporation, Armonk city, State of New York, USA).

A. Questionnaire Scores
The questionnaire scores are shown in table I. Scheirer-Ray-Hare test found that no significant interactive effects on SOO scores between two factors (Immersion and Multisensory stimulation), H=0.0005, p=0.982>0.05;The main effect of immersion was significant (H=6.70,* * p=0.01), the SOO scores under HMD conditions were higher than that under HD conditions.In addition, the main effect of multisensory stimulation was significant (H=16.02,* * * p<0.001), the SOO scores were higher under visuo-tactile conditions than that under visual only conditions.
Due to the non-interactive effect between two factors in Scheirer-Ray-Hare test (H=0.0005,p=0.982>0.05),Wilcoxon matched-pairs signed-rank (two-sample) test could be performed to analyze the differences between each condition.Results suggested SOO score was significantly higher under HMD-VT than that under HMD-V, Z= −3.023, * * p = 0.003; Similarly, SOO score under HD-VT was also higher than that under HD-V, media of difference = 1, Z= −3.001, Asterisks indicate significant levels ( * * p<0.01).* * p = 0.003; And SOO score of HD-VT had no significant differences comparing to that of HMD-V (Z= −1.394, p = 0.163>0.05),and was significantly lower than that of HMD-VT (Z= −2.625, * * p = 0.009), as illustrated in Figure 2.

B. Group-Level Event-Related Spectral Perturbation (ERSP) and Topographical Maps
The positions of C3/C4 electrodes are closed to the sensorimotor area, and the ERSP time-frequency map of C3/C4 is illustrated in Figure 3. Blue areas indicated stronger ERD.At ipsilesional side (left column), the energy reduction of α band (8∼13Hz) in HMD-VT was obvious, that is, the ERD during the visual guided MI was significantly stronger than the other three conditions, and the β band (13∼30Hz) also showed ERD significantly (mainly in range of 13∼20Hz).The ERD in the α band (8∼13Hz) in HMD-V and HD-VT conditions were slightly weaker than the HMD-VT, and stronger than HD-V.The ERD of the β band was also consistent with the conclusion of the α band.The difference between the β band and the α band was most significant in HMD-VT, which gradually decreased in HMD-V and HD-VT, and basically disappeared in HD-V.Additionally, this regularity kept similar at contralesional side (right column).
In order to further explore the spatial distribution of ERD during MI, according to obvious ERD frequency interval in ERSP maps (α band: 8∼13Hz; low β band: 13∼20Hz), topographical maps of patients with stroke were illustrated under four conditions: HMD-VT, HMD-V, HD-VT and HD-V (Figure 4).Blue areas indicated stronger ERD.The results of the topographic map were similar to ERSP, confirming the differences among four conditions.The amplitude of ERD under visuo-tactile synchronous stimulation was more significant than under vision alone.ERDs under HMD-V and HD-VT conditions were similar in both the contralateral and ipsilateral hemispheres.In terms of spatial distribution, both left and right hemisphere injured patients showed bilateral ERD activation during visual guided MI training.
Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.   of ERD amplitude were normally distributed, therefore, the 2× 2 repeated measures ANOVA and paired t test were performed.
For ERD at ipsilateral hemisphere (Intact side), no significant interactive effect was found in α band between two factors (Immersion and Multisensory stimulation), F=0.023, η2=0.001,p=0.880, neither in β band (F=0.658,η2=0.035,p=0.428).The main effect of immersion in α band was significant (F=17.995,η2=0.500,* * * p=0.000), and the main effect of multisensory stimulation in α band was also significant (F=19.449,η2=0.519,* * * p=0.000).But in β band, there were no significant main effects of immersion (F=2.474,η2=0.121,All of these results are shown in figure 5. The results for the SOO scores were same as those for the peak amplitudes of the ERD at ipsilesional side in α band.Due to this similarity, the correlation between the SOO over the virtual hand and the peak amplitudes of ERD during MI were analyzed.The correlation between SOO scores and the peak amplitudes of ERD at ipsilesional side in α band was significant (ρ =−0.60, * * * p<0.001).The correlation between SOO scores and the peak amplitudes of ERD at contralesional side in α band was significant (ρ =−0.47, * * * p<0.001).The correlation between SOO scores and the peak amplitudes of ERD at ipsilesional side in β band was significant (ρ =−0.44, * * * p<0.001).The correlation between SOO scores and the peak amplitudes of ERD at contralesional side in β band was significant (ρ =−0.27, * p<0.02).As shown in the figure 6.
Additionally, an extra phenomenon was found between visuo-tactile conditions and visual only conditions.As illustrated in figure 7, the power of α band was higher than that of β band in visuo-tactile conditions (the left column), particularly significant in HMD-VT.But this kind of difference did not exist in visual only conditions (the right column).

IV. DISCUSSION
This study investigates the visual guided MI in patients with stroke through manipulating two factors (immersion and multisensory stimulation) related with SOO, providing evidence for possible between SOO and MI.We that two factors modulated of SOO and ERD MI similarly in patients with stroke.Multisensory stimulation factor would influence the power ratio of α band to β band.Bilateral activation was found in four conditions.And a moderate correlation was found between SOO scores and peaks ERD amplitude during MI.Additionally, the similarity of SOO level and MI under the condition of HMD-V and HD-VT may improve the MI effect of 2D monitor display which is widely used in clinic.
The results of SOO scores suggested that patients with stroke could have SOO over virtual hand, and the factors of immersion and multisensory stimulation could affect the degree of SOO: the SOO over virtual hand under HMD was significantly stronger than that of 2D monitor display, and the SOO under visuo-tactile stimulation was significantly stronger than that of visual stimulation alone.This is consistent with the conclusion of previous studies that both immersion and multisensory stimulation can enhance the SOO over the virtual hand [31].It is worth noting that the degree of SOO over virtual hands under the condition of HMD-V and HD-VT was similar, which indicated that the decreased SOO in 2D monitor display rebounded due to visuo-tactile stimulation.On one hand, previous studies found that level of immersion could modulate the degree of SOO, whether in healthy sub-Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.jects [40], [41], [42], individuals with stroke [35], or some of other neurological diseases [43].On the other hand, visuotactile stimulation was the most typical method of inducing (or enhancing) SOO in many illusion researches, such as RHI [30], VHI [31], full body illusion [32], etc. Combining this information with results of this paper (the degree of SOO over virtual hands under the condition of HMD-V and HD-VT is similar), it is obvious that the reduced SOO caused by low immersion could be improved by visuo-tactile stimulation.
During the MI of the affected hand, the peak ERD amplitude of stroke patients showed similar results to scores: under HMD conditions, the peak amplitude of α band (8∼13Hz) at C3/C4 channel was significantly higher than that of 2D monitor display, and the peak ERD amplitude under visuo-tactile stimulation was significantly higher than that of visual stimulation alone.The peak ERD amplitude under the condition of HMD-V and HD-VT was similar, which also indicated that the reduced ERD caused by low immersion could be improved by visuo-tactile stimulation.The similarity of SOO scores and ERD peaks under influences of two factors, means that there may be a close relationship between SOO and MI: the higher level of SOO over the virtual hand, the stronger ERD would occur during visual guided MI.Spearman correlation analysis provided evidence for this conjecture: there was a moderate positive correlation between the SOO scores and the peaks amplitude of ERD, which indicated that the degree of brain activation during MI of the affected limb was related to the level of SOO over the virtual hand.Immersion and multisensory stimulation may modulate the activation of damaged brain areas during MI in stroke patients by manipulating the SOO over virtual hands.Previous studies have shown that both MI [12], [27] and SOO [44] depend on multisensory mechanism and relate to activation of sensorimotor cortex.This study further verified that SOO enhanced visual guided MI from two aspects (subjective perception and neural mechanism).
Previous study thought that low-frequency β ERD reflected a release of cortical inhibition associated arousal of motor systems during movement execution and observation [45].Some literatures reported that within the β band several cooccurring β rhythms might be distinguished in some subjects (high β: >20Hz; low β: <20Hz) [46].It is worth noting that there was also a certain degree of ERD at low-frequency β band (13∼20Hz) in our results.Although the overall actiwas weaker than that in the α band, the distribution among the four conditions was similar to that in the α band.Unexpectedly, an extra phenomenon was found between visuo-tactile conditions and visual only conditions.The power of α band was higher than that of β band in visuo-tactile conditions, particularly significant in HMD-VT, but this kind of difference did not exist in visual only conditions.Previous studies have shown that sensorimotor oscillatory rhythms were usually recorded in α band and β band, in which α rhythm mainly reflected the sensorimotor process in more extensive brain regions such as post parietal cortex, premotor cortex and somatosensory cortex, while β rhythm had a stronger correlation with process of motor cortex [47].This finding may be important evidence supporting the effect of multisensory stimulation on visual guided MI.Interestingly, there was no significant ERD in the β band during MI in healthy participants [12], [27].In this study, the ERD of β band in stroke patients may be related to the neural compensation caused by impaired motor function in stroke patients, or the results of MI of gross movement caused by the disability of fine motor.Some studies have pointed out that the ERD phenomenon in the β band may be related to a wide range of movement (such as movement of full body).Therefore, the activation of α band and β band should both be important for visual guided MI in stroke patients.
In addition, previous studies have reported conflicting results about the laterality of ERD during MI.Some studies have found that the activation of ERD during visual guided MI was mainly in the contralateral hemisphere of the task hand [12], while other studies suggested that the ERD caused by visual guided MI was the activation of bilateral brain regions [27], [48], [49].Generally speaking, the significant laterality (contralateral) of ERD would occur in fine movement, and the bilateral activation of ERD would occur in gross movement (e.g., shoulder movement [48]).In this study, the laterality of ERD was also demonstrated.The results show that the ERD occurred in bilateral brain area.We speculated that the bilateral activation in stroke patients may be related to the MI of gross movement compensation caused by impaired fine motor function in stroke patients.This may be the source of the controversy about laterality in the previous literature.
For ERD at both affected side and intact side, there was no significant interactive effect between immersion and multisensory stimulation.This phenomenon is worth further discussing.According to the assessing method of immersion proposed by Slater and Wilbur [50], i.e., "the degree of immersion can be objectively assessed as the characteristics of a technology, and has dimensions such as the extent to which a display system can deliver an inclusive, extensive, surrounding and vivid illusion of virtual environment to a participant", we set two levels of immersion: HMD (high) and 2D monitor display (low).In other word, 'immersion' in this paper referred to technological immersion of devices, including the level of user-tracking, the use of stereoscopic visuals, and wider fields of view of visual displays, etc. [51].Thus, the immersion factor in this study was mainly manipulated by different vision input.But the multisensory stimulation factor was manipulated by tactile stimuli.Considering that these two factors originate from different channels, it is plausible that there was no significant interactive effect between immersion and multisensory stimulation.Simultaneously, both the vision [12], [17] and tactile [27], [28], [49] are related to bilateral activation during MI training, it is responsible that this non-interactive effect of intact side was similar to that of affected side in ERD.
Beyond SOO, there are other characteristics related to the extent of immersion exist, notably sense of presence [52], and some technological characteristics mentioned in above discussion [51].These characteristics can be investigated by presence questionnaire (PQ) and immersive tendencies questionnaire (ITQ) [53].All of these factors are directly related to the virtual environment (VE) that contains the virtual body, instead of the virtual body itself.But for stroke motor rehabilitation, body (or limbs) itself is the direct interesting target rather than VE.Thus, SOO may be a more important characteristic than other characteristics of immersion in VR stroke motor rehabilitation.
The task in this study was mixed with both motor imagery and sensation brought by tactile stimulation.The tactile stimulation could induce ERD, as demonstrated in a previous study [54], the experimental design cannot completely rule out the impact of tactile stimulus.The reason that we designed this experiment (participant imagined hand movement with simultaneously receiving visuo-tactile stimulus) is that SOO may gradually fade out without continuous multisensory stimulations.This phenomenon was predicted by updating model [55], [56] related to SOO forming.The updating model states that multisensory bindings decay over time, the embodied entity is assumed to fade from the body representation unless its integrated status is refreshed continually.Furthermore, SOO over the virtual hand has been found gradually fading out after multisensory stimulation disappearing in some researches [57], [58].Thus, the cessation of visuo-tactile stimulation during the MI phase would lead to a rapid fading of SOO: the MI phase lasted more than 300s (10s per trial * 30 trials), whereas SOO disappeared about 1min after the stimulation was stopped.Considering this possibility, the experiment design applied the visuo-tactile stimulation during MI, maintaining the illusion existing.
Additionally, it is worth noting that ERD induced by tactile stimulus occurs immediately (500ms) after the stimulus and lasts for about 6 seconds, then returns to the resting state [54].In other words, ERD induced by tactile stimulus lasts for at least 6 seconds.However, in this study, tactile stimulation was applied every 2-3 seconds throughout the entire process, meaning that the decreased voltage induced by tactile stimulus remained in a stable task state throughout the experiment.Therefore, this stable voltage can be considered as the 'resting state' of the motor task.Our results were computed by subtracting the average voltage in the baseline (before 0s in every trial, resting state) from the voltages at each time point.In this computation, the decrease in voltage (ERD) induced by tactile stimulus was offset.Hence, we believe that tactile stimulation could not significantly influence the results of motor imagery in this work, or at most, has only limited effects.
This study showed that the immersion and visuo-tactile stimulation were effective for enhancing visual guided MI in stroke rehabilitation, and the curative effect of these methods may due to the cortical recoding of body image and action observation network mediated by virtual limb.As mentioned previously, action observation and visual guided training [7], [8], [9], [10] have been demonstrated to achieve a better MI training outcome for clinical stroke rehabilitation [11].For instance, the Fugl-Meyer assessment (FMA) and pinch strength test (PST) scores of stroke patients with upper limb motor dysfunction were significantly improved after receiving visual guided MI training for four consecutive weeks [11].Kaneko et al. [59] also found that clinical outcomes of motor function (FMA) and spasticity (Modified Ashworth Scale, MAS) significantly improved after visual guided training for ten days in stroke patients with severe paralysis of upper limb motor function.A systematic review pointed out that MI is an effective intervention when added to physical practice for upper limb recovery after stroke [3].According to these evidence, visual guided MI has been more widely used in stroke rehabilitation with better clinical rehabilitation outcomes.This study suggested that SOO may be an important factor underlying the effects of immersion and multisensory stimulation on visual guided MI in stroke patients, and was probably associated with clinical stroke rehabilitation.Additionally, the similarity of SOO level and MI effect under the condition of HMD-V and HD-VT is of great significance to improve the MI effect of 2D monitor display, and 2D monitor display is widely used in clinic temporarily.
However, there are also some limitations in this study.Firstly, this study had a wide age distribution (ranges from 36 to 76 years), and future studies should control the age distribution as much as possible.Secondly, in order to maintain SOO over virtual hand, this study applied visuo-tactile stimulation during MI training.Future studies should look for new ways to both exclude interference of tactile and maintain SOO during visual guided MI.Thirdly, the other VR devices (e.g., CAVE) with different immersion level should also be considered in future studies.Finally, the multisensory stimulation also includes several sensory channels (e.g., visuomotor stimulus) not only visuo-tactile stimulus in this study.
V. CONCLUSION This study discussed the possible factor underlying the effects of immersion and multisensory stimulation on visual guided MI, through comparing the differences of ERD among four SOO levels in post-stroke patients during MI, identifying the effect of SOO in this process, and proved the feasibility of the combination of visuo-tactile stimulation and virtual reality enhancing visual guided MI in post-stroke patients.The results of questionnaire and EEG analysis showed that the higher level of SOO, the stronger ERD activation of related brain regions occurred during MI.There was a moderate positive correlation between SOO and peak ERD amplitude.Patients could experience SOO similar to HMD under visuo-tactile stimulation of 2D monitor display, and this similarity also keeps in ERD during visual guided MI, which indicates that multisensory stimulation (visuo-tactile) can provide a better MI treatment for 2D monitor display which is widely used in clinic.The conclusion of this work could be referred in future studies for coming up with better MI paradigms for stroke rehabilitation.

Fig. 1 .
Fig. 1.The experiment settings (Illusion induction and motor imagery): (a) The experiment settings in VR (HMD) conditions.Virtual scene in HMD shows a pair of virtual hands placing on a virtual disk similarly to real world, and a red virtual ball will touch virtual hand in illusion stage.(b) The experiment settings in 2D monitor display conditions.All settings keep same with VR conditions except for 2d monitor display replacing HMD.(c) One trial in motor imagery stage (virtual hand performs wrist flexion and extension movements).(d) The channel location on EEG cap.(e) The flowchart of whole experiment.

Fig. 2 .
Fig. 2. Comparison of SOO scores among four conditions.The points in figure represent the distribution of SOO scores among all participants.Asterisks indicate significant levels ( * * p<0.01).

Fig. 3 .
Fig.3.The group-level ERSP map at C3/C4 channels (maps in left column show ipsilesional side, and maps in right column show contralesional side) for (a) HMD-VT, (b) HMD-V, (c) HD-VT, (d) HD-V conditions.These maps illustrate the relative amplitudes (%) for each paradigm.The horizontal dashed white lines represent 8Hz, 13Hz and 30Hz, respectively.The vertical red and yellow lines represent the start and end of the MI task stage, respectively.Dark blue corresponds to negative relative amplitude and indicates stronger ERD.For each participant, the STFT for each trial was first averaged and baseline normalized.The resulting subject-level ERSPs were then averaged to obtain the group-level ERSP maps for each condition.

Fig. 4 .
Fig. 4. The group-level topographical maps in (a) α band and (b) low β band of four conditions during MI task of affected hand (maps in top row show right hemisphere injured, maps in bottom row show left injured).

Fig. 6 .
Fig. 6.The correlation analysis between SOO scores and the peak ERD amplitudes at (a) ipsilesional side in α band, (b) contralesional side in α band, (c) ipsilesional side in β band, (d) contralesional side in β band.The data was illustrated with scatter plots, and there were 76 points (4 conditions * 19 subjects) in every subfigure.The green histograms show the amount of data points in specific SOO score, and the red histograms show the amount of data points in specific ERD value.The ρ spearman and p values are present in the top of each subfigure.

Fig. 7 .
Fig. 7.The difference between α and β band at (a) ipsilesional side and (b) contralesional side in four conditions.The vertical red lines indicate the start (500ms) and end (3500ms) of the data pieces tested by paired sample t-tests.Asterisks indicate significant levels ( * * p<0.01, * * * p<0.001).

TABLE I THE
QUESTIONNAIRE SCORES OF SOO IN STROKE PATIENTS

TABLE II THE
PEAK ERD AMPLITUDES IN α AND β BANDS OF C3/C4 CHANNELS ON CONTRALATERAL HEMISPHERE AND IPSILATERAL HEMISPHERE DURING MI TASKS OF AFFECTED HAND, GIVEN AS MEAN STANDARD ERROR (%)