Event-Related Desynchronization Induced by Tactile Imagery: an EEG Study

Abstract It is well known that both hand movements and mental representations of movement lead to event-related desynchronization (ERD) of the electroencephalogram (EEG) recorded over the corresponding cortical motor areas. However, the relationship between ERD in somatosensory cortical areas and mental representations of tactile sensations is not well understood. In this study, we employed EEG recordings in healthy humans to compare the effects of real and imagined vibrotactile stimulation of the right hand. Both real and imagined sensations produced contralateral ERD patterns, particularly in the μ-band and most significantly in the C3 region. Building on these results and the previous literature, we discuss the role of tactile imagery as part of the complex body image and the potential for using EEG patterns induced by tactile imagery as control signals in brain-computer interfaces (BCIs). Combining this approach with motor imagery (MI) could improve the performance of BCIs intended for rehabilitation of sensorimotor function after stroke and neural trauma.


Introduction
The capacity to form internal mental representations of physical activities, such as overt actions and their associated sensations, which is also referred to as mental imagery, is a fundamental aspect of human cognition. Mental imagery could be associated with cognitive functions, including memory, creativity, motor control, navigation, arithmetic, moral decisions, and mind wandering (Pearson, 2019). One of the most studied forms of mental imagery is motor imagery (MI), which induces neuroplasticity changes at the cortical level and facilitates motor learning (Ruffino et al., 2017;Ladda et al., 2021). MI practice is used as supplementary training in sports (Mizuguchi et al., 2012a) and neurorehabilitation (Machado et al., 2019). Depending on the sensory modality, kinesthetic and visual types of motor imagery have been distinguished (Stevens, 2005). These two distinct strategies have specific effects on corticospinal excitability (Stinear et al., 2006), electroencephalogram (EEG) activation patterns (Neuper et al., 2005), and metabolic brain activation (Guillot et al., 2009). Studies suggest that kinesthetic motor imagery is processed by the motor circuits that overlap with the circuits activated during the preparation and execution of voluntary movements. Since imagining sensations arising from muscle contraction (muscle and skin tension, changes in joint angle) is typically a part of kinesthetic MI strategy, it is reasonable to suggest that such imagery engages somatosensory areas, as well. The somatosensory component of MI was studied experimentally (Naito et al., 2002) and analyzed theoretically (Grush, 2004). Recent studies have also shown that somatosensory inputs can facilitate cortical excitability during MI (Mizuguchi et al., 2009(Mizuguchi et al., , 2011(Mizuguchi et al., , 2012bKaneko et al., 2014;Yakovlev et al., 2019).
A transient decrease in EEG/magnetoencephalography (MEG) power in a specific frequency band (Pfurtscheller and Da Silva, 1999;Neuper et al., 2006) called event-related desynchronization (ERD) is a conventional marker of sensorimotor processing. ERD and the opposite process called event-related synchronization (ERS) represent changes in correlated activity of the underlying neuronal populations, and these changes correspond to increased and decreased cortical processing, respectively (Pfurtscheller and Da Silva, 1999). As such, ERD/ERS can be used for functional cortical mapping (Pfurtscheller, 2001;Zhao et al., 2019) or as classification features for brain computer interfaces (BCIs; Pfurtscheller and Neuper, 2006;McFarland and Wolpaw, 2017;Vasilyev et al., 2017). Sensorimotor rhythms is a type of oscillatory activity in the a (8-12 Hz) and b (13-30 Hz) ranges that occurs in cortical sensorimotor regions during voluntary movements, motor imagery, movement observation, and tactile stimulation . It is assumed that the sensorimotor a (also referred as to m-rhythm) is generated in the primary somatosensory cortex, whereas sensorimotor b is generated in the primary motor cortex . Frolov et al. (2012) and Frolov et al. (2014) confirmed the involvement of primary somatosensory areas in m-rhythm generation. They found that motor imagery-associated EEG activity and blood-oxygen level-dependent (BOLD) signals had the same source localized in the hand representation of the sensorimotor cortex.
While studies on MI are plentiful, fewer research has been conducted on tactile imagery (TI). Neuroimaging studies of TI are limited to the fMRI method (Yoo et al., 2003;Schmidt et al., 2014;Schmidt and Blankenburg, 2019), and there is a lack of systematic work with the EEG approach. Yoo et al. (2003) reported activation of the sensorimotor cortex caused by mental imagery of brushing stimuli. Schmidt et al. provided evidence of changes in coupling between cortical areas during TI (Schmidt et al., 2014;Schmidt and Blankenburg, 2019). Additionally, Bashford et al. (2021) observed cortical activity in human subjects imagining the sensations that were previously evoked by intracortical microstimulation (ICMS) of their somatosensory cortex using an implanted microelectrode array. Somatosensory imagery (also referred to as somatosensory attention orientation, SAO) has been proposed as a complement (Yao et al., 2017a) or alternative (Yao et al., 2018(Yao et al., , 2022a to MI for use in BCIs. Yao and colleagues demonstrated that SAO can be reliably decoded using a BCI paradigm and used as an independent modality. They also improved SAO classification by using congruent tactile stimulation (Yao et al., 2017b(Yao et al., , 2022b and showed that the sources of sensorimotor ERD are mainly located in the somatosensory cortex (Yao et al., 2022a). Overall, these studies demonstrated similarity between neural patterns exhibited during real perceptions and TI tasks, which is consistent with the results of MI studies showing that overlapping brain regions are engaged in physical movements and motor imagery.
Building on these previous results, here we further clarified the EEG correlates of TI. By analogy with the MI, we expected that TI would modulate the contralateral m-rhythm. In agreement with this expectation, we observed ERD of the m-rhythm during both tactile stimulation (TS) and TI, the finding that adds to the literature on mental imagery and contributes to the development of a new generation of BCI technologies.

Participants
Twenty healthy right-handed volunteers (age 22.5 6 2.8 years; nine females) without a history of neurologic disorders took part in this study. All of them had no prior experience in mental imagery of the motor and tactile types. The study protocol was approved by the Lomonosov Moscow State University ethical committee and followed the Declaration of Helsinki. All participants were informed about the experimental procedures and signed an informed consent.

Experimental procedure
The experiment duration did not exceed 90 min. During an experimental session, subjects sat in a comfortable armchair in a room with uniform lighting. Visual cues were presented on a 24-inch LCD monitor positioned at the distance of ;1 m from the subjects' eyes.
An experimental session consisted of four consecutive conditions signaled by screen cues: TS, control, learning of TI, and TI. This experimental sequence is explained in Figure 1 and Table 1. Each condition comprised randomly mixed trials. The duration of each trial was 6s. The trials came in four varieties: TS, TI, control and the reference state (rst). TS and TI trials were cued by the corresponding pictograms. During TS trials, a vibrotactile stimulus was presented. TI trials required imagining being stimulated in the absence of the stimulus. During control trials, the TS pictogram was shown but no tactile stimuli were presented and TI was not required. During rst trials, participants mentally counted objects shown on the screen (circles, dots, lines, squares, etc.; see Fig. 1) while avoiding explicit eye movements. We chose to use rst-trials for consistency with the previous studies of MI where similar complex visual scenes were used as a reference state (Tangwiriyasakul et al., 2013;Vasilyev et al., 2017;Vasilyev et al., 2021). A gray screen was shown during the intertrial intervals, which lasted 300-400 ms.
As shown in Table 1, the TS condition included 20 TS trials randomly intermixed with 20 rst trials. The control condition was a mixture of 20 control trials and 20 rst trials. The learning condition included 20 TS trials intermixed with 20 rst trials. Stimulation intensity gradually decreased, and participants were required to train their TI. Finally, the TI condition was a mixture of 20 TI and 20 rst trials. In summary, participants started with experiencing vibrotactile stimulation or counting visual objects (TS condition). Next, a control was run with the same visual cues but no stimulation. In the learning condition, vibrotactile stimulation resumed, but its amplitude decreased, and participants had to perform TI. During the TI condition, participants imagined being stimulated in the absence of actual stimulation or counted visual objects, depending on the trial type.

Selection of the site for tactile stimulation and imagery
The skin area for the application of tactile stimulation was selected based on the tests in ten subjects, which preceded the main series of experiments. We sought for the place on the arm where tactile stimulation would be well perceived without any unpleasant sensations. This perception could be then reproduced as mental imagery. The subjects did not have any prior experience in imagery, and four of them participated later in the main experimental session. Vibrotactile stimuli were delivered with vibration motors, which were placed over several locations on the right arm, including index finger, thumb, palm, and inner wrist surface. We used 20 trials per stimulation site, followed by the participants' attempts to imagine the tactile sensation. The participants were then asked to describe their subjective experience and select the site where the stimulation-evoked sensation was the easiest to imagine.
Based on the results of these tests, the surface of the distal phalanx of the index finger and inner surface of the wrist were chosen as the most suitable places for TS and TI. Of these two locations we preferred the wrist and chose not to apply TS to the fingers for several reasons. First, finger stimulation led to long-term residual sensations that could continue into the other trials (Macefield et al., 1996;Ribot-Ciscar et al., 1996), which were also supported by the self-report from the subjects. Second, EEG modulations during finger stimulation are not hemispherically specific. According to Genna et al., 2017), mental imagery of finger movements, as well  control, learning and tactile imagery (TI) organized in runs. Each run was a random mixture of 20 reference state (rst) trials (counting visual objects) and 20 somatosensory trials (or control with the same visual cue but no stimulation). Specifically, the TS condition had TS and rst trials, the control condition had control and rst trials, the learning condition had TS trials with decreasing stimulus amplitude and rst trials, and the TI condition had TI and rst trials. The duration of each trial was 6 s. Vibrotactile stimulation was composed of pulses of vibration of variable frequency.
as tactile stimulation of the fingers, result in bilateral m-ERD patterns whereas stimulation of more proximal parts of the arm cause predominantly contralateral ERD patterns (Le Franc et al., 2021;Li et al., 2021). Third, we wanted to avoid any potential association of TS with motor movements which could have been the case for stimulation of the finger.

Tactile stimulation trials
We used a custom designed and computer-controlled vibrotactile stimulator based on an Arduino UNO. A flat vibratory motor (6 mm in diameter, max speed of 12 000 rpm) was placed on the inner surface of the right wrist. The stimulation continued for the entire 6-s trial duration. Variable-frequency stimulation patterns were used: stimulation was delivered as 100 ms with vibratory frequency picked from the range 3700-12,000 rpm, and the time interval between the pulses varied in the range 200-500 ms (Fig. 1). This random-frequency stimulation reduces tactile habituation and helped to avoid residual tactile sensations.

Control trials
Since the TS and TI trials were cued with the same visual stimulus (vibration pictogram), we had to control for the possible effects of this cue. Repeated presentation of such cues leads to the development of cross-modal association (Zhou and Fuster, 2000;Bor et al., 2014). As a control, we added the condition where participants observed the same visual cues during the TS and TI trials, but no stimulation occurred and imagery was not required. Twenty of these trials were randomly intermixed with 20 rst trials (N = 20).

Transition to tactile imagery
Transition to TI was conducted gradually during the learning session where 20 TS trials were intermixed with 20 rst trials. The TS amplitude gradually decreased from the parameters used in the TS session to zero. The subjects were instructed to focus attention on the skin sensations evoked by TS and to memorize them. As the TS intensity decreased, participants were asked to mentally compensate for the lack of stimulation by imagining the previously experienced sensation of stronger vibration. A similar learning strategy was previously used in MI studies Vasilyev et al., 2017;Liburkina et al., 2018).
Following the learning session, subjects provided a verbal report regarding task difficulty; they also described the imagery vividness. An objective measure of the imagery strength was given by the m-ERD value and topographic distribution. The imagery was considered stable if a participant reported that he/she could mentally reproduce the tactile sensations at the same place where TS was applied, the imagined sensation was not associated with the limb movements and muscle contractions, and the imagery was maintained for 6 s. The data from three participants were rejected from the analysis because one of them was unable to imagine tactile sensations and in the other two no ERD occurred during both TS and TI.

Tactile imagery
The TI condition followed the learning condition. Here, no tactile stimuli were applied; instead, participants were instructed to imagine vibrotactile sensations. TI trials were triggered with the same visual cues as the ones used for TS trials. Participants were asked to maintain imagery of vibrotactile stimulation for the entire 6-s interval during which the cue stayed on the screen. Like the TS, control and learning conditions, rst trials constituted half of the trials in the TI condition and TS trials constituted the other half. The TI condition was run twice to a total of 40 rst trials and 40 TI trials. We controlled for the absence of muscle activity by monitoring the EMG signals from the right flexor digitorum superficialis (FDS) muscle.

EEG and EMG recordings
We recorded 48-channel monopolar EEG data at a 500-Hz sampling rate using an NVX-52 DC amplifier (MCS). Passive Ag/Cl sensors were placed in accordance with the 10/10 international montage system. The TP10 electrode was used as the reference. The skin-electrode impedance for each of the electrodes did not exceed 20 kV. Bandpass filtering was conducted in the range 0.1-75 Hz using a FIR filter. An additional filtering was conducted using a 50-Hz Notch filter. Raw data collection was conducted using the NeoRec software (MCS) synced with the stimulus presentation environment. Stimulus presentation was conducted via a self-written code based on the PsychoPy python-module (Peirce, 2007). We also used one bipolar lead to record EMG from the FDS muscle.

Signal preprocessing
Raw EEG recordings were re-referenced using the common average reference (CAR), which also has spatial filtering effects (McFarland et al., 1997). Next, the signals were bandpass filtered in the range 1-30 Hz using a fourth-order Butterworth filter. Noisy channels were interpolated using the spherical spline method (Perrin et al., 1989). The preprocessed signal was divided into 8-s epochs (from À1 to 7 s relative to trial onset) for each trial. To assess the time-frequency dynamics of EEG oscillations amplitude, the Morlet wavelet transform with variable number of cycles was applied for all the extracted epochs. The frequencies of the wavelets ranged from 6 to 30 Hz with 0.4-Hz step. The full width at half maximum (FWHM) was equal to 180 ms corresponding to a spectral FWHM of 6 Hz. Then the time-frequency matrices with Morlet coefficients were converted to ERD values in percent with Equation 1, where Morlet coefficients for each timestamp within the trial were used instead of median power spectral dencity (PSD) value.
To plot the topographic distribution of ERD/S in each experimental condition, the power spectral density was calculated with the Welch method (Welch, 1967) for the epochs from 1 to 6 s relative to trial onset. The subjectspecific spectral peak was detected semi-automatically (i.e., under visual inspection) in the range of 8-15 Hz. This peak was consistent across different scalp locations and stable over time. Next, a frequency interval of 61.5 Hz around the peak frequency in the resting state was selected and used for ERD calculation for each recording channel (Eq. 1): Where PSD smr is the median spectral power across the epochs of the TS, TI or control trials in the range 1-6 s and subject-specific frequency subrange for each channel position. PSD rst defines the spectral power calculated for trials corresponding to the resting state.

Statistical analysis
To determine significant changes in the time-frequency dynamics of oscillatory activity and spatial distribution of the ERD/S values in all the experimental conditions (TS, TI, and control), the nonparametric cluster-level paired t test with 10.000 permutations was used (Maris and Oostenveld, 2007). This procedure is assumption free and is widely used for multichannel data and time-frequency analysis to avoid the multiple comparisons problem. The Friedman's test and a Wilcoxon signedrank test as the post hoc statistical test were performed to determine the statistically significant differences in total ERD/S value within the full trial in TS, TI, and control conditions, as well. The Bonferroni correction was applied for the number of tested hypotheses (adjusted to p , 0.003 for 15 comparisons). Statistical analysis was performed via the SciPy module v.1.4.1 (Virtanen et al., 2020). For signal processing and permutation methods, the MNE-Python v.0.23 was used (Gramfort et al., 2013). For data visualization, we used the Matplotlib graphics environment (Hunter, 2007).

Time frequency EEG dynamics during tactile stimulation and imagery
We started with the analysis of changes in EEG rhythms for the TS and TI trials. The Wavelet-Morlet transform was applied to the signal from channel C3 (i.e., the sensorimotor area contralateral to the stimulated hand), followed by an analysis of the event-related time-frequency perturbations (TFP). We used the C3 channel for the final analysis and visualization for all participants because for all of them the highest ERD response occurred for this channel. The choice of C3 was confirmed by statistical permutation tests results. With the Morlet transformation we obtained the time-frequency representation matrices for the TS, TI, and control conditions. Then, we baselined them to the resting state (data from rst trials), calculated the median over the group and then visualized. Figure 2 shows that the group-median temporal spectral dynamics were similar during the TS and TI trials: the m-rhythm amplitude gradually decreases relative to resting state and remains at the reduced level until the end of the trial (Fig. 2B). Then, it returned to the baseline. To discover significant differences between the somatosensory conditions (TS and TI) and the control state, we performed intrasubject subtraction of the time-frequency matrix corresponding to the control condition from the time-frequency matrices corresponding to TS and TI conditions. The nonparametric cluster-level paired t test was used to compare the obtained difference matrices with zero. We also used a similar procedure to perform comparison between the TS and TI conditions. Both the tactile stimulation and tactile imagery were characterized by a decrease across the a and b band power compared with the control condition. However, only the the decrease in the m-rhythm power was significant for the TI-condition when it was compared with the control state (e.g., the m-ERD clusters that persists for the entire trial duration in Fig. 2A), whereas a significant b -ERD occurred during the TS trials. No differences in TFP between the TS and TI were observed. Thus, the frequency range of the m-rhythm response appearing was similar for TS and TI (;8-14 Hz as can be defined in Fig. 2).

ERD/S topographical patterns
To determine the topographical localization of the sensorimotor response, we compared the m-ERD and b -ERD values for each electrode position across all experimental conditions using the nonparametric permutation tests. Median (N = 17) topographic distribution patterns of m-ERD/ S for all experimental conditions are shown in Figure 3A. The m-ERD induced by the tactile stimulation of the righthand wrist was more prominent in the central EEG sites over the sensorimotor cortical areas, with some contralateral dominance. The TI condition had a similar m-ERD distribution pattern. A different pattern was found for the control state.
Using a permutation t test relative to zero for each experimental condition, we determined the channels where significant m-ERD occurred. The permutation t test revealed a significant decrease in the m-rhythm amplitude over C3 and C1 channels for the TS condition (tC3 = À5.9; p = 0.001; tC1 = À6.5; p = 0.0007) and over FC5, C3, and C5 for the TI condition (tFC5 = À6.2; p = 0.0009; tC3 = À5.9; p = 0.0015; tC5 = À7.3; p = 0.0002). No significant m-ERD topographical origins were found in the control state, that is when the tactile stimulation was not applied, and participants did not perform imagery of tactile sensations. The obtained matrices of p-values were superimposed over the scalp EEG channels positions for visualization (Fig. 3B). Since the TS-trials were characterized by a significant eventrelated desynchronization in the ;20to 26-Hz frequency range, we performed spatial permutation t test for the b -ERD values. Figure 4 shows that a significant decrease in b -amplitude induced by the tactile stimulation of the right-hand wrist occurred contralaterally, predominantly over the fronto-central EEG channels. Significant spatial clusters where b -ERD emerged were discovered neither in the TI condition nor in the control state.
To analyze the global m-amplitude and b -amplitude decrease within the epoch duration, we calculated the median ERD value over all trials for the interval starting with the first s till the end of the trial, and over the individual frequency ranges where ERD was detected across all the participants. We rejected the first s of each active trial because of the TI onset jitter, which contributed to a gradually developing ERD (Fig. 4). The first s of the rst trials was rejected to avoid the residual effects of the tactile stimulation and imagery. There were significant differences between the experimental conditions according to Friedman's test in m-ERD value in the C3 channel (x 2 = 25.7; p = 0.000002). The paired comparisons revealed no significant difference between the ERD values in TS and TI conditions (W = 56; p = 0.33; see Fig.  5). On the other hand, as shown in Figure 5, the m-desynchronization in TS and TI conditions was significantly stronger compared with the control state (W = 0; p , 0.0001). The obtained statistical data are shown in Table 2.
A statistical analysis of the b -ERD values revealed significant differences between the three experimental conditions (x 2 = 15.6; p = 0.0004). Further analysis (Wilcoxon signed-rank tests) indicated that b -ERD was significantly stronger in the TS-condition as compared with the TI and control conditions (W = 13; p = 0.002; W = 1; p = 0.0003, respectively). The obtained statistical data are provided in Table 2.

Discussion
In this study, we examined the changes in sensorimotor rhythms during real and imagined vibrotactile stimulation.
We found that vibrotactile stimulation of the wrist as well as its imagery caused a prolonged decrease in the EEG power in the m-range. Time-frequency analysis showed a significant m-ERD in the 9-14 Hz range throughout the 6-s TS or TI trials. Additionally, TS (but not TI) caused significant b -ERD in the frequency range of 20-26 Hz. The frequency range where sensorimotor ERD appeared during TS and TI, as well as its spatial distribution, were similar to the well-known ERD patterns associated with real and imagined motor activities. This result raises the question regarding the similarities and differences in the ERD source localization for the motor versus somatosensory induced responses. We observed that cortical location of the m-ERD evoked by TI was similar to the location previously described for MI. Although it is tempting to interpret this result as a proof of a common neural substrate for both MI and TI needed for the processing of somatosensory sensations associated with movements, EEG recordings have a limited spatial resolution, so a more accurate mapping of ERD sources should be conducted in the future for different types of mental imagery. The previous literature on MI suggests that the m-rhythm ERD originates in the somatosensory cortex while b -ERD is mainly localized to the precentral areas Frolov et al., 2012;Frolov et al., 2014). Specifically for TI, Yao et al.   (2022a) suggested that ERD in the frequency range 8-26 Hz originates in sensorimotor cortex. They, however used Colin27 template for source localization, which potentially could have resulted in an overlap between the a-sources and b -sources. Yao et al. (2018Yao et al. ( , 2022a recently proposed using TS and TI-evoked ERD for controlling a BCI. They used vibratory stimulation that was like ours, and they also stimulated the wrist. While their work was mainly focused on constructing a BCI, we looked at the details of ERD associated with TI and added several new methodologies. First, we performed a systematic learning session where subjects familiarized themselves with the vibratory induced perceptions. Subjects were naive to mental imagery tasks in our study. Additionally, participants' imagery trials were not mixed with any kind of physical stimulation in our study, which is different from the work of Yao and colleagues who presented 200-ms pulse of vibration before each trial to precondition mental imagery, which could have added bottom-up effects to the ERD. Second, we used a mixed design with an equal number of somatosensory and reference trials, which also had equal durations. In the studies of Yao and colleagues, a short period before the start of the trial (from À2 to À1.2 s) was used as a reference for TI-related effects. However, such an individual reference for each trial is prone to drifts related to changes in the participant's physiological state during the experimental session. Our calculation of the reference was different from this method. We calculated the reference as the median oscillation power from all rst trials. This approach increased the statistical power of our findings and made the results more reliable. Additionally, using a visual scene as a resting state switched attention from the somatosensory task and reduced occipital a activity during the resting state, which is more robust for ERD estimation during imagery tasks (Vasilyev et al., 2017;Vasilyev et al., 2021). For the group-level analysis, we used the median calculation and nonparametric statistics to obtain more robust and reliable results for a relatively small sample size (N = 17).
Third, in the studies of Yao and colleagues, ERD was derived from a wide frequency band that comprised both a and b activity (8-26 Hz), whereas we analyzed ERDs for the m-bands and b -bands separately whose cortical sources are different. The effects of TI were different for these bands. Additionally, we analyzed individual m-rhythm ranges for each participant to account for the intersubject   Yao et al. (2018Yao et al. ( , 2022a and complement their work. Sensorimotor ERD during tactile stimulation and tactile imagery Electrical stimulation of the median nerve is a widely used method for exploring cortical effects of afferent stimulation. Many studies showed that median nerve stimulation leads to prominent contralateral ERD of the sensorimotor EEG rhythms (Salenius et al., 1997;Nikouline et al., 2000). Similar results were observed in the studies with cutaneous vibro-tactile stimulation (Cheyne et al., 2003;Vasilyev et al., 2017). Here, we used vibrotactile stimulation of the skin area on the right wrist. We expected to obtain results consistent with the results of aforementioned studies. Such effects of afferent stimulation suggest a general principle for the relationship between a-like rhythms and sensory events: before the presentation of somatosensory stimuli, synchronized a activity is observed, and the stimuli result in a-desynchronization, possibly because of disinhibition of thalamocortical circuits (Pineda, 2005).
In addition to m-ERD, tactile stimulation led to desynchronization in the b -frequency range. Yet, this effect was not observed for TI (i.e., was statistically insignificant). b -Oscillations have been linked to the activity of the primary motor cortex activity as it is well known that somatosensory inputs modulate b -oscillations because of motor-cortical mechanisms (Gaetz and Cheyne, 2006;Illman et al., 2021Illman et al., , 2022. These effects are driven by the ascending projections to M1 (Home and Tracey, 1979;Lemon, 1981) and by corticocortical connections with the somatosensory cortex (Jones, 1983). We suggest several reasons for the absence of b -ERD during TI. The first possibility is that TI is maintained by sustained activity of the primary and secondary somatosensory areas without the engagement of the motor cortex (Zhou and Fuster, 1997;Yoo et al., 2003). Indeed, participants did not imagine any movements in our experiments, so the activation of motor circuits could be weak. The second possibility is that the motor cortex responded during TS but not TI is that application of vibration to FDS generated a strong input to the motor cortex because of the activation of muscle spindle afferents (Edin, 1990). Stimulation of muscle spindles could evoke kinesthetic illusions, but this was not the case in our experiments particularly because the participants were instructed not to imagine limb movements. The third possibility is a relatively low level of alertness during TI compared with TS. Thus, Illman et al. (2021) suggested that alertness level has an effect on the sensorimotor b -desynchronization during tactile stimulation. Yet, the effects they observed were not statistically significant at the group level.
Notably, despite the absence of statistically significant b -ERD during TI, we observed a slight decrease in b power in this condition. This matches the general mental imagery model, which considers mental imagery as a reverse form of perception process with weaker cortical manifestations (Munzert et al., 2009;Dentico et al., 2014;Pearson et al., 2015). According to this model, mental imagery is based on the processing of information retrieved from memory by cortical areas. Following this idea, we propose that tactile stimulation can lead to memory formation, and this memory could be brought back to somatosensory areas when the previously experienced sensation is reproduced by imagining. Our results and the proposed interpretation agree with the simulation theory of motor imagery formulated by Jeannerod (2001).

Tactile imagery perspectives
The tactile sense is one component of the complex motor image (Kilteni et al., 2018), in fact, the tactile imagery is a part of the motor imagery, which includes kinesthetic sensations and visual image as well. Motor imagery is a widely used mental technique in sport to improve performance and motor imagery based BCIs are promising tools for the poststroke motor recovery. However poststroke patients often have difficulties in motor imagery learning. Tactile imagery similarly to motor imagery can be proposed as a BCI control paradigm based on m-ERD detection as well (Yao et al., 2018(Yao et al., , 2022aYakovlev et al., 2022). We suppose that somatosensory mental images are easier to perform, and the learning of tactile imagery is more regulated and understandable compared with motor imagery, thus TI-based BCIs could be more useful and lower barriers to entry compared with MI-BCIs. There are a lot of studies showing the usefulness of tactile feedback and tactile stimulation during motor imagery training and MI-BCI control (Zhang et al., 2021;Sakamaki et al., 2023;Zhong et al., 2023). We believe that the positive effect of TS on the learning on kinesthetic imagination of movements may be associated with the use of mental tactile sensations (i.e., participants can actually use the tactile imagery to increase the vividness of motor image and induce the sensorimotor rhythms depression more effectively).

Significance of the obtained results
Although several previous studies have already examined the effects of TI, our results are novel pertaining to EEG correlates of TI following learning to imagine vibrotactile stimulation. We observed m-rhythm desynchronization in the contralateral hemisphere during both TS and TI and b -desynchronization during TS only, which we interpreted as activation of both somatosensory and motor cortices during TS and activation of somatosensory cortex only during TI. Our results generally match the plentiful previous results on motor imagery (Pfurtscheller and Neuper, 1997;McFarland et al., 2000) assessed with EEG technique. The results we obtained also match the fMRI studies where tactile imagery was found to increase somatosensory cortical activity and enhance functional connectivity between the prefrontal and somatosensory cortical areas (Schmidt et al., 2014;Schmidt and Blankenburg, 2019). Our results also supplement recent EEG work on somatosensory imagery (Yao et al., 2018(Yao et al., , 2022a. The other study that our results agree with is the work (Bashford et al., 2021) where microelectrode techniques were used to localize cortical regions involved in the tactile image formation. Our work is also relevant to the research of spinal circuits where tactile and kinesthetic types of processing lead to increased spinal excitability (Bunno, 2019). Overall, our work adds to the existing tactile imagery literature and is also relevant to the field of brain-computer interfaces, where tactile imagery could be used to obtain EEG responses desired for communication, control of external devices and rehabilitation purposes (Yao et al., 2017a(Yao et al., , b, 2018(Yao et al., , 2022a.

Limitations
The results of the current study provide convincing evidence that TI leads to contralateral m-rhythm ERD, and this effect is similar to the one evoked by the real vibrotactile stimulation applied to the wrist. However it is important to mention several limitations to our research. First, EEG recordings allowed us to observe ERD/S responses to sensorimotor stimuli, but we were unable to precisely localize the source of ERD using this method. Highdensity EEG recordings combined with inverse solution computing (e.g., MNE, sLORETA) could provide more detailed information about the sources of imagery-related activity and could help to better compare the effects of tactile stimulation, tactile imagery and motor imagery. Other neurophysiological techniques besides EEG, like transcranial magnetic stimulation (TMS), MEG, functional near-infrared spectroscopy (fNIRS), could be useful for this purpose, as well.
We used a visual scene as a reference state and also used it in the control condition to exclude the conditioning effect of the visual pictogram used for the TS/TI tasks. The obtained results suggest that sensorimotor rhythm modulation was the specific effect of TI. Yet it could be useful to have an additional control for tactile attention in future studies. There could be a somatosensory attention condition, where the subject's attention is drawn to the stimulation site, but without performing imagery.
Another limitation in this study is participants' age range, since most of the subjects were relatively young (from 19 to 30 years old). It is important to complement our findings with a broader-age group in the future. Since a decline in sensorimotor functions is typically observed with aging, it would be of interest to study the effect of age on the EEG patterns.
In conclusion, based on EEG recordings conducted during tactile imagery, we conclude that the effects of TI (namely, m-rhythm event-related desynchronization) are similar to the effects of MI. Such mutuality of motor and somatosensory cortical processing, at least at the level of EEG, could indicate that the "imagery circuit" of the brain is not strictly localized. Yet, by changing imagery requirements it is possible to change activity distribution in somatosensory versus motor areas. Considering the sensory origin of the a rhythms in general, a-ERD during TI is yet another phenomenon where somatosensory processing is affected by thought. As such, it could be used in BCI design where thought needs to be communicated from the brain to the external devices.