This study was conducted in accordance with the Declaration of Helsinki. All experimental protocols and procedures were approved by the Research Ethics Committee in Shonan Fujisawa Campus, Keio University (Approval Number 167). The examiners provided a detailed explanation of the purpose, experimental procedures, potential benefits, and risks involved. After receiving all of the relevant information, participants provided written informed consent before participating in the experiment.

We recruited 16 gymnasts (11 men, 5 women, aged 18–24 years) and 16 healthy adults (8 men, 8 women, aged 19–22 years) as a non-gymnast group. All participants were right-handed. All gymnasts had been practicing at least for 7 years (range: 7–16 years) and had participated in an all-Japan intercollegiate gymnastic championship at least once. Note that two gymnasts were members of the Japanese national gymnastics team. The non-gymnasts group had no experience of gymnastic training. None of the participants had experienced any neurological and musculoskeletal disorders.

Two-sided unpaired t-tests were performed on VMI and KMI scores of KVIQ between groups (non-gymnasts vs. gymnasts), to confirm differences in the subjective vividness of motor imagery between them. To test the differences in αERDmax and βERDmax during motor execution or KMI, we performed a two-way mixed-model ANOVA between participant groups (gymnasts and non-gymnasts) and tasks (wrist dorsiflexion and shoulder abduction). If the interaction was significant, we performed a two-sided unpaired t-test for groups (gymnasts vs. non-gymnasts) and a two-sided paired t-test for tasks (wrist dorsiflexion vs. shoulder abduction). To test the differences in the EEGα-PSD or EEGβ-PSD between resting-state EEG and relax-phase EEG during tasks, we also performed two-way ANOVA between participant groups (gymnasts and non-gymnasts) and conditions (resting sate and relax phase). The p-values of 0.05 were used to indicate statistical significance. All statistical analyses were performed using SPSS statistics software (IBM SPSS Statistics 25, IBM developerWorks, Tokyo, Japan).

Figure 2 shows group data (mean ± S.D.) for KVIQ scores obtained from VMI and KMI tasks between gymnasts and non-gymnasts. The KVIQ scores were significantly greater in gymnasts, both in VMI (gymnasts, 42.68 ± 6.22; non-gymnasts, 35.44 ± 8.73, p = 0.011) (Figure 2A) and in KMI (gymnasts, 43.50 ± 6.78; non-gymnasts, 35.94 ± 8.24, p = 0.008) (Figure 2B). The KVIQ results indicate that gymnasts subjectively evaluated how vividly they could imagine their own movements.

Typical examples of EEG signals, EEG time-frequency maps, and ERD time courses during wrist dorsiflexion from a non-gymnast and a gymnast are shown in Figures 3A,B, respectively. From these time-frequency maps, a decrease in EEG power can be observed around 12 and 22 Hz in the contraction phase (0 to 5 s) compared with the relax phase (−6 to −3 s) in both participants when performing motor execution.

Figure 3C shows group data for αERDmax during wrist dorsiflexion and shoulder abduction motor execution. An ANOVA on αERDmax during motor execution revealed no significant effects of group [F_{(1, 30)} = 0.209, p = 0.651] and task [F_{(1, 30)} = 0.831, p = 0.369], while a significant interaction was obtained [F_{(1, 30)} = 4.654, p = 0.0391]. An unpaired t-test for group revealed no significant difference in αERDmax between gymnasts and non-gymnasts for wrist dorsiflexion (gymnasts, 64.41 ± 20.96; non-gymnasts, 65.58 ± 13.97, p = 0.855) and shoulder abduction task (gymnasts, 70.13 ± 21.07; non-gymnasts, 63.26 ± 16.77, p = 0.316). A paired t-test for task revealed no significant difference in αERDmax between wrist dorsiflexion and shoulder abduction execution both for non-gymnasts (wrist dorsiflexion, 65.58 ± 13.97; shoulder abduction, 63.26 ± 16.77, p = 0.303) and for gymnasts (wrist dorsiflexion, 64.41 ± 20.96; shoulder abduction, 70.13 ± 21.07, p = 0.078). Figure 3D shows the group data for βERDmax during wrist dorsiflexion and shoulder abduction motor execution. An ANOVA on the βERDmax during motor execution with groups and task revealed no significant effects of group [F_{(1, 30)} = 0.571, p = 0.456] and task [F_{(1, 30)} = 1.249, p = 0.273], and interaction [F_{(1, 30)} = 1.815, p = 0.188]. The results revealed no effects of sports experience and body part on ERD magnitude during motor execution.

Typical examples of EEG signals, EEG time-frequency maps, and ERD time courses during wrist dorsiflexion KMI from a non-gymnast and a gymnast are shown in Figures 4A,B, respectively. From the time-frequency map for a non-gymnast participant, we did not observe clear ERD in the imagery phase (0–5 s) compared with the relax phase (−6 to −3 s) in both the α-band and β-band (Figure 4A). Conversely, clear ERD can be observed in the time-frequency map for the gymnast participant around 12 and 22 Hz (Figure 4B).

Figure 4C shows the group data for the αERDmax during wrist dorsiflexion and shoulder abduction KMI. An ANOVA on αERDmax during KMI showed significant effects of group [F_{(1, 30)} = 5.437, p = 0.027] and task [F_{(1, 30)} = 10.975, p = 0.002]. No significant interaction effect [F_{(1, 30)} = 0.266, p = 0.610] was observed. Figure 4D shows the group data for βERDmax during wrist dorsiflexion and shoulder abduction KMI. An ANOVA on βERDmax during KMI showed no significant effects of group [F_{(1, 30)} = 0.876, p = 0.357], but significant effects of task [F_{(1, 30)} = 8.019, p = 0.008] and a significant interaction [F_{(1, 30)} = 7.421, p = 0.010]. An unpaired t-test for group revealed a significant difference in βERDmax between gymnasts and non-gymnasts for the shoulder abduction task (gymnasts, 46.08 ± 18.09; non-gymnasts, 32.27 ± 17.84, p = 0.021) but not for the wrist dorsiflexion task (gymnasts, 46.42 ± 22.76; non-gymnasts, 49.56 ± 17.84, p = 0.666). A paired t-test for task revealed a significant difference in βERDmax between the wrist dorsiflexion and shoulder abduction KMI conditions for non-gymnasts (wrist dorsiflexion, 49.56 ± 17.84; shoulder abduction, 32.27 ± 17.84, p = 0.002), but not for gymnasts (wrist dorsiflexion, 46.42 ± 22.76; shoulder abduction, 46.08 ± 18.09, p = 0.938). These results indicated that gymnastics experience affected ERD magnitude during KMI of sports-non-specific movements.

Figure 5A shows the group data for the EEGα-PSD in resting-state EEG and relax-phase EEG during the tasks. An ANOVA examining EEGα-PSD data revealed no significant effects of group [F_{(1, 30)} = 0.486, p = 0.491] or condition [F_{(1, 30)} = 4.135, p = 0.051]; however, a significant interaction [F_{(1, 30)} = 6.382, p = 0.017] was observed. An unpaired t-test for group revealed no significant difference in EEGα-PSD during resting-state EEG (gymnasts, 0.473 ± 0.155; non-gymnasts, 0.482 ± 0.166, p = 0.878) and that during the relax phase (gymnasts, 0.482 ± 0.155; non-gymnasts, 0.403 ± 0.125, p = 0.878; p = 0.122) between gymnasts and non-gymnasts. In non-gymnasts, a paired t-test revealed significant differences in the EEGα-PSD between conditions (resting state, 0.482 ± 0.166; relax phase, 0.403 ± 0.125, p = 0.008). However, in gymnasts, no significant differences in EEGα-PSD were observed between conditions (resting state, 0.473 ± 0.155; relax phase, 0.481 ± 0.155, p = 0.715). Figure 5B shows the group data for EEGβ-PSD. An ANOVA on EEGβ-PSD showed no significant effects of group [F_{(1, 30)} = 0.106, p = 0.747] and condition [F_{(1, 30)} = 1.787, p = 0.191], and no significant interaction [F_{(1, 30)} = 0.063, p = 0.804]. These results indicate that in non-gymnasts, the EEGα-PSD was smaller in the relax-phase EEG than during resting-state EEG, while such a difference was not observed in gymnasts. Conversely, EEGβ-PSD did not differ between resting-state EEG and relax-phase EEG in both gymnasts and non-gymnasts.

We evaluated ERD as a physiological indicator of KMI ability in the present study because it is considered to reflect changes in corticospinal excitability (Takemi et al., 2013a) and is associated with the subjective vividness of KMI measured by KVIQ (Toriyama et al., 2018). It should be noted that the present results revealed greater ERD magnitude during KMI of sports-non-specific movements in gymnasts than in non-gymnasts, although differences in neural activity between athletes and non-athletes have not been reported in motor imagery of sports-non-specific movements in previous studies using EEG (Di Nota et al., 2017), MEG (Kraeutner et al., 2018), TMS (Fourkas et al., 2008), or fMRI (Wei and Luo, 2010). This may be related to the fact that gymnasts perform motor imagery including KMI frequently as a part of their daily practice to reduce the risk of serious injury in their practice. Furthermore, a previous psychological study showed that the vividness of motor imagery of sports-non-specific movements was higher in athletes engaged in individual and/or non-contact sports compared with athletes engaged in the team and/or contact sports (Di Corrado et al., 2019). Thus, as gymnasts have superior motor imagery ability among athletes, they may provide a particularly suitable population for highlighting differences in neural activity during KMI of sports-non-specific movements compared with non-athletes.

It is possible that the present task protocol, in which participants performed KMI following motor execution repeatedly in the order of seconds, led to the current finding of greater ERD magnitude in gymnasts. In psychological questionnaires, the conventional procedure for measuring motor imagery ability is to examine participants while they perform motor execution, then motor imagery in one trial, and subjectively evaluate the vividness of the motor imagery of the preceding movement (Malouin et al., 2007). In physiological experiments, however, the conventional procedure involves evaluating neural activity while participants perform only motor imagery (Fourkas et al., 2008; Wei and Luo, 2010; Di Nota et al., 2017; Kraeutner et al., 2018). Thus, there has been a methodological gap in the approach for examining motor imagery between psychological questionnaire studies and experimental physiological studies. In response to direct questioning in the current study, gymnasts reported that they usually perform actual movements and KMIs alternately in their daily practice. Therefore, the method for measuring KMI in the present physiological experiment was designed in accord with the procedure of psychological questionnaire measurement. We provided no special prior training or instruction to the participants to remove its effect on motor imagery (Wriessnegger et al., 2018; Meng and He, 2019; Corsi et al., 2020). Participants were asked to imagine the sensation of the motor execution they had performed just before, similar to the way in the KVIQ. The current physiological findings may have been due to differences between gymnasts and non-gymnasts in the ability to flexibly modulate corticospinal excitability when imagining their own movements, by referring to the actual movement.

As shown in equation (1), we were able to confirm that the ERD was determined by both the degree of synchronization during the relax phase [R(f)] and the degree of desynchronization during KMI [A(f, t)]. As shown in Figure 5, first, gymnasts could return their sensorimotor α-rhythm during the relax phase in the task to the same power level as during the resting state for 60 s, whereas non-gymnasts could not. Thus, gymnasts appeared to be good at relaxing deeply by making their sensorimotor rhythms more synchronized within a short period of time. However, higher EEGα-PSD during the relax phase does not appear to be the only factor involved in gymnasts' greater ERD magnitude in the α-band. As shown in Figure 2, second, differences in αERDmax between groups were not observed in motor execution but were found in KMI. Thus, gymnasts also appeared to be good at increasing corticospinal excitability by making their sensorimotor rhythms more desynchronized, even in KMI. Overall, the present results demonstrate that gymnasts have the ability to generate a clear contrast in the state of the sensorimotor cortex, when required to repeatedly switch across relaxing, motor execution, and KMI conditions. On the basis of the current findings, we believe that the ability to modulate the brain state without any movement is a core aspect of superior KMI ability in gymnasts.

Interestingly, this study demonstrated different results between ERD magnitude in the α-band and β-band. Several previous studies reported that functional roles played by the sensorimotor rhythms are different between frequency bands. During actual muscle contraction with weak-to-moderate intensity, the sensorimotor rhythm is known to be coherent with EMG activity only in the β-band, with no significant coherence in the α-band (Ushiyama et al., 2010, 2017; Suzuki and Ushiyama, 2020). In addition, sensorimotor β-oscillations have been proposed to be the basis for large-scale communication across sensorimotor, other areas, and the periphery (Kilavik et al., 2013). When focusing on oscillatory power itself, the EEG spectral power in the sensorimotor area contralateral to the contracted/imagined limb was decreased in both the α- and β-bands, while that in task-irrelevant cortical regions was increased in the α-band (Pfurtscheller, 1992; Pfurtscheller and Lopes da Silva, 1999), but not in the β-band (Stolk et al., 2019). During KMI, the ERD magnitude was increased by increasing task demand in the β-band, but not in the α-band (van Elk et al., 2010; Brinkman et al., 2014). Taken together, these findings suggest that the functional roles of sensorimotor rhythms for movement/imagery should be distinguished between the α-band and β-band.

First, αERDmax was larger in gymnasts than in non-gymnasts during both KMI tasks (i.e., wrist dorsiflexion and shoulder abduction). In task-relevant cortical regions, neural populations are assumed to be disinhibited by the ERD of the sensorimotor area in the α-band, which would allow reallocation of computational resources (Brinkman et al., 2014). However, task-irrelevant cortical regions are assumed to be inhibited by enhancing their α-oscillations (Pfurtscheller et al., 1997). As gymnasts are required to perform skilled movements successively, they are trained to quickly switch their attention across their body parts by facilitating task-relevant regions and inhibiting task-irrelevant regions. The present results regarding ERD in the α-band would reflect such an ability of gymnasts.

Second, βERDmax was larger in gymnasts only during shoulder abduction KMI, but not during wrist dorsiflexion KMI. This task specificity in βERDmax may be caused by ERD in the β-band playing a role in the calculation of specific motor commands. In general, wrist movement is used frequently in daily life, which makes it easy for most people to perform wrist dorsiflexion KMI. Thus, βERDmax would not differ between gymnasts and non-gymnasts in the wrist dorsiflexion KMI task. However, as isometric shoulder abduction is a movement rarely used in daily life, it may be difficult for most people to perform this KMI. Conversely, gymnasts are well-trained to move their upper limbs, including the shoulder joints, both dynamically (i.e., giant swing) and statically (i.e., handstand and rings). Therefore, it would be easy for gymnasts to imagine shoulder abduction because they are skilled at adjusting the movement parameters of their shoulder joints. We assume that ERD in the β-band is an indicator for how precisely a person can imagine their own movement kinesthetically.

In the present study, only upper limb movements (i.e., wrist dorsiflexion and shoulder abduction) were examined. Gymnasts use their upper limb muscles specifically as anti-gravity muscles for postural control, such as handstand and pommel horse. This usage of the upper limbs is unique relative to the movements of non-gymnasts. The uniqueness of gymnasts' upper limb usage may lead to superior KMI ability regarding upper limb movements. Thus, we cannot clearly predict whether similar results would be obtained when performing similar experiments for other body parts. However, the KVIQ results demonstrated that gymnasts tended to show higher scores for all movements. In addition, most gymnasts perform motor imagery of various body parts in their daily practice. On the basis of these findings, we speculate that gymnasts have superior KMI ability irrespective of body parts, although confirming this possibility will require further investigation.

The present study is the first to observe differences in physiological indices between athletes and non-athletes during KMI of sports-non-specific movements. This means that the present study can bridge the gap between psychology and physiology studies regarding differences in KMI ability of sports-non-specific movements between athletes and non-athletes. However, because only gymnasts participated in this study, it is unclear whether the present results are specific to gymnasts or apply generally to athletes performing any sports. Because differences in motor imagery ability of sports-non-specific movements would be expected among athletes, further investigation is needed to elucidate sports-specific differences in motor imagery ability for sports-non-specific movements.

In the field of brain–computer interfaces (BCIs), conflicting results have been reported for the relationship between BCI performance and psychological questionnaire score. In particular, some studies have demonstrated a significant correlation between them (Vuckovic and Osuagwu, 2013; Marchesotti et al., 2016), while the other has not (Rimbert et al., 2019). The present study supports the former one because gymnasts reported higher subjective vividness of KMI with a greater magnitude of ERD compared to non-gymnasts. This result might be due to the consistency of imagery method between physiological and psychological measurements of this study (i.e., imagining the sensation of the movement performed just before). Thus, it is still unclear whether gymnasts always excel in the ability to modulate sensorimotor rhythms in other imagery methods such as repeating only motor imageries and/or imagery with online neurofeedback. In any case, the present study indicated the importance of comparing corticospinal excitability measured by ERD for evaluating KMI ability. In future studies, imagery training using bioelectrical signals may provide a useful tool for improving the motor imagery ability of athletes.