We used F-test model with ANOVA in G*Power software (Version 3.1 for Mac) to calculate the sample size with a significant level of 0.05, an effect size of 0.25 and a statistical power of 0.8. The total required sample size was 24. Thirty-one university students participated in this study, and were randomly divided into two groups (i.e., congruent or incongruent). All participants had no neurological, skeletal, or muscular problems with normal or corrected vision. They signed a written informed consent form before the experiment. The Ethics Committee of Zhejiang University Psychological Science Research Center permitted our experiment.

The data from one participant in the incongruent group was excluded because of stepping during the experiment. Also, the data from three participants was excluded due to the severe body swaying (i.e., spread their arms or use the hip strategy) during the experiment (one in incongruent group, two in congruent group). Therefore, 14 participants (5 females) were included in the congruent group (mean age 22.93 ± 2.28 years, height 172.21 ± 8.80 cm, body mass 64.78 ± 11.24 kg) and 13 participants (6 females) were included in the incongruent group (mean age 24.85 ± 2.57 years, height 169.35 ± 7.05 cm, body mass 63.38 ± 12.45 kg) for the present study.

We combined a rotating platform and a VR headset with the visual scene to manipulate vestibular and visual inputs, respectively. Also, we used the Unity3D program to synchronize the control of the rotating platform and the visual scene to set “congruent” (i.e., natural visual stimulation) and “incongruent” (i.e., conflicted visual stimulation) experimental conditions (Figure 1).

In the congruent condition, the rotating platform was accelerated at 4°/s² for 10s to the right (clockwise) and kept rotating at 40°/s for 27 s, and then decelerated at 4°/s² for 10s, with no additional control in the visual scene. Thus, the visual scene moved counterclockwise relative to the participant. In the incongruent condition, the rotating platform was set to the same parameters as in the congruent condition. In this case, the visual scene moved clockwise relative to the participant, providing visual information that was opposite to the actual motion (Figure 1).

Before the experiment, demographic data (e.g., age, height, body mass, dominant side) was recorded for each participant. Participants were asked to stand on the stable platform and to cross their arms on the chest for 30 s (baseline) followed by 47 s of “congruent” or “incongruent” condition (rotation) and finally again on the stable platform for 30 s (recovery). Each participant was asked to complete five repeated tasks, either congruent or incongruent, with 5-min rest between each standing task. During the standing task, each participant was asked to cross their arms on the chest.

We recorded COP data, surface EMG (sEMG) data and EEG data from participants during the experimental tasks. A Wii balance board was placed in the center of the rotating platform and the BrainBlox program software was used to collect COP data with a 100 Hz sample frequency (Cooper et al., 2014). A 16-channel sEMG system (Trigno Wireless System, Delsys, United States) was used to collect the sEMG data of the left and right Tibialis anterior (TA) and Medial gastrocnemius (MG) with a 2,000 Hz sample frequency. The skin was shaven, abraded, and cleaned with alcohol prior to placing the electrodes. A double-sided tape was used to fix the electrodes. Also, an ANTNeuro EEG device containing 32 channels in the 10–20 standard regime was used to collect EEG data. The impedance of all electrodes remained below 5 k ohms throughout the experiment with a 1,000 Hz sampling frequency.

In the present study, we used the rotating platform to manipulate the vestibular information since the semicircular canals within the vestibular system are responsible for detecting the angular acceleration information. Thus, we divided raw data of COP, sEMG and EEG into two phases for further analysis: acceleration (0–10 s after platform start) and deceleration (37–47 s after platform start).

Since the postural control system is inherently dynamically nonlinear (Ivanenko and Gurfinkel, 2018) and previous studies showed that COP, sEMG, and EEG signals all have the positive Lyapunov exponent (Sbriccoli et al., 2001; Kannathal et al., 2005; Liu et al., 2015), complexity-based nonlinear methods can be used to identify mechanisms underlying variability in postural control (Ting et al., 2009; Kędziorek and Blażkiewicz, 2020). In the present study, fractal dimension (FD) analysis with the Higuchi’s algorithm was used for providing an indication of the complexity of a signal and quantifying the self-similarity of a pattern over multiple time-scale (Higuchi, 1988). Higher FD value suggest higher complexity of time series. Higuchi’s algorithm for FD calculation was shown below (Cui et al., 2017): first, construct k new signals from a given COP signal to a newly constructed signal S m k ; then, calculate the length L m ( k ) of S m k and compute the mean of L m ( k ) over m called L(k); finally, plot L(k) against k (ranging from 1 to k max ) on a double logarithmic scale and calculate the slope of this line as the FD index. To choose an appropriate k max , each FD index was plotted against the k max . The point at which the plateaus is considered a saturation point, and the k max should be chosen (Doyle et al., 2004).

To better understand the linear and nonlinear interactions of postural control system, we also calculated the traditional linear measures of COP, sEMG and EEG signals as complementary analysis.

Statistical analysis was performed using SPSS software (Version 24.0 for Mac). We combined the dependent variables from acceleration and deceleration phases for the further statistical analysis. Differences in measures were assessed through a mixed two-way ANOVA to examine the effect of practice (1st and 5th trial) and the group (Congruent and Incongruent). A series of post-hoc multiple comparisons with Bonferroni correction was used. To investigate the potential relationships between low- or high-frequency components of COP data and central adjustments (based on EEG data) or peripheral adjustments (based on sEMG data) respectively, we used two-tailed Pearson correlations between the complexity of low-frequency components of COP data and the complexity of EEG data, and between the complexity of high-frequency components of COP data and the complexity of sEMG data. Statistical significance was set at p < 0.05.

A two-way mixed ANOVA revealed a significant main effect of practice on the FD values of low- and high-frequency components of COP and on the SD values of COP in the AP and ML directions (Supplementary Table S1). However, no significant main effect of group or interaction effect of practice × group was observed. Post-hoc tests revealed that practice under the incongruent condition decreased FD_low values of COP in the AP and ML directions (p < 0.01) (Figure 2B), and decreased SD values of COP in the AP direction (p < 0.05) (Figure 2A). Practice under the congruent condition decreased SD values of COP in the ML direction (p < 0.05) (Figure 2A). Practice under congruent and incongruent conditions both increased FD_high values of COP in the AP and ML directions (p < 0.001) (Figure 2B).

A two-way mixed ANOVA revealed a significant main effect of practice on the FD values and on the normalized activity of TA and MG muscles in sEMG data. Also, there was a significant main effect of group on the FD values of right TA and on the normalized activity of right TA and left and right MG (Supplementary Table S2). However, no interaction effect of practice × group was observed. Post-hoc tests revealed that practice under the congruent and incongruent conditions increased FD values of sEMG in left and right TA muscles (p < 0.01) (Figure 3B), and decrease the sEMG activity in left TA muscle (p < 0.001) (Figure 3A). Besides, the FD value of right TA muscle in sEMG was greater under the 5th congruent condition compared with the 5th incongruent condition (p < 0.05) (Figure 3B). Left MG muscle activity was greater under the 1st incongruent then 1st congruent condition (p < 0.01) (Figure 3A).

A two-way mixed ANOVA revealed a significant effect of practice on the FD values of VC, TPJ, PPC, S1 and MC in EEG and on the alpha-band power value of VC, and a significant effect of group on the FD values of VC and TPJ in EEG. However, no interaction effect of practice × group was observed (Supplementary Table S3). Post-hoc tests revealed that practice under the congruent condition decreased the FD values of EEG in MC, S1, PPC, and TPJ (p < 0.01) (Figure 4B), while practice under the incongruent condition decreased the FD values of EEG in MC, S1, PPC, and VC (p < 0.05) (Figure 4B) and decreased the alpha-band power value of VC (p < 0.05) (Figure 4A). Besides, the FD value of EEG in VC was greater under the 1st incongruent condition compared with the 1st congruent condition (p < 0.01), and the FD values of EEG in TPJ under the 5th incongruent condition was greater compared with the 5th congruent condition (p < 0.01) (Figure 4B).

The two-tailed Pearson correlation tests found that (1) significant positive correlations between the complexity of high-frequency component of COP data in AP and ML directions and the complexity of sEMG data of left and right calf muscles (Table 1); (2) significant positive correlations between the complexity of low-frequency component of COP data in AP direction and the complexity of EEG data of VC, TPJ, and PPC (Table 2).

Our fractal analysis of high-frequency components of COP signals revealed a significant increase after repeated balance practice under congruent and incongruent conditions (Figure 2B). This increased complexity of postural sway is in agreement with van den Hoorn et al. (2018), suggesting that increased complexity of postural sway indicates an improved ability to adapt to changing environments with effective strategies in postural control (Kędziorek and Blażkiewicz, 2020).

Our fractal analysis of sEMG signals showed a significantly increase in the fractal exponent of TA muscle after repeated practice under congruent and incongruent conditions (Figure 3B, top). According to these present results and previous findings that more unstable standing conditions induced lower sEMG complexity of calf muscles (Murillo et al., 2012), we suggest that higher complexity of sEMG data probably indicate the greater capacity of postural muscles to adapt to the perturbation. However, our present results showed no significant changes in the fractal exponent of MG muscle activity after repeated practice under congruent or incongruent condition (Figure 3B, bottom). Accordingly, we might suggest that TA muscles adapted to the perturbation in a more effective way than MG muscles. This is consistent with Schmid et al. (2011), who found that activity of TA decreased more significantly than that of MG during a 3-min backward and forward horizontal oscillations of the support base. It is reasonable to assume that muscles responsible for dorsiflexion (i.e., TA) and plantar flexion (i.e., MG) of the ankle play different role in postural control. Changes in muscle length of TA better reflect the changes in ankle angle, thus providing a better source of proprioceptive inputs; whereas MG actively contributes to stabilizing continuous postural sway (Merletti et al., 2010).

The changes in muscle activation and the complexity of sEMG signals are not always align across different tasks. For example, muscle fatigue decreased the muscle activity (%MVC) and also decreased the complexity of sEMG data with lower FD values or lower sample entropy values (Rampichini et al., 2020). However, our present results showed that practice under incongruent condition decreased the left TA muscles activity but increased the complexity of left TA sEMG data with higher FD values. This results are consistent with previous studies, where showed that when balance task difficulty increased, the calf muscle activity increased (Gebel et al., 2019) and the complexity of sEMG data decreased (Murillo et al., 2012; Pakniyat and Namazi, 2021). Besides, we also observed that practice under incongruent condition reduced the left MG muscle activity without any significant changes of fractal exponent of left MG muscle (Figure 3). Thus, when exploring changes in the sEMG data caused by different factors like muscle fatigue or postural perturbations, it is possible to comprehensively interpret the results from linear and nonlinear measures.

Although we found an increase in the fractal exponent of sEMG data from TA and high-frequency of COP data, the results of the analysis of EEG signals showed a reverse trend that repeated balance practice under congruent and incongruent conditions reduced the fractal exponent of EEG signals in different ROIs (Figure 4B). This reverse trend aligns with the findings in Pakniyat and Namazi (2021), where the complexity of EEG signals was higher in walking than standing in response to perturbation whereas the complexity of sEMG signals exhibited the opposite pattern, being higher in standing than walking. We could speculate that complexity environment might increase the complexity of EEG (Kamal et al., 2020; Pakniyat and Namazi, 2021). Accordingly, the decrease of complexity of EEG signals after habituation balance practice in the present study might correlate with the decrease of level of central volitional control or level of attention (Ming et al., 2009).

However, our fractal analysis of low-frequency COP signals revealed a significant decrease after repeated practice under incongruent condition (Figure 2B). Moreno et al. (2022) recently argued that the long-time latency changes in low-frequency of COP signals align with the idea that regulating the COP to the desired state is primarily governed by alterations in the body’s reference configuration linked to central volitional control. Accordingly, the decreased complexity of low-pass components of COP suggested a decreased level of central volitional control after repeated practice under conflicting environments. Our correlation results are consistent with this hypothesis, showing that the complexity of low-frequency component of COP data was positively related with the complexity of EEG data from visual cortex and posterior parietal cortex (Table 2). Furthermore, we also found a significantly positive correlation between the complexity of high-frequency component of COP data and the complexity of sEMG data from calf muscles (Table 1). These results supported the previous findings that low- and high-frequency components of COP might be related to peripheral adjustments (such as reflex mechanisms and mechanical muscular properties) and central adjustments (such as cortical response), respectively (Zatsiorsky and Duarte, 1999; Moreno et al., 2022).

Previous studies suggested that the increased complexity of EEG signals could reflect a good cortical response to stimuli (Takahashi et al., 2009). Accordingly, we observed that repeated practice under incongruent condition both reduced the fractal exponents of visual cortex EEG signals and the alpha-band power values of visual cortex (Figure 4). This result is consistent with the sensory reweighting theory (Peterka, 2002), suggesting that the brain probably down-weighted the less reliable visual information in our experimental setting for postural control. In contrast, our results showed a significantly decrease in the fractal exponents of TPJ EEG signals after repeated practice under congruent condition rather than incongruent condition, while there was no significant change of alpha-band power value after practice under congruent or incongruent condition (Figure 4). Although the superior temporal lobe has both congruent neurons and opposite neurons (Gu et al., 2006), the complexity of EEG data from superior temporal lobe reduced after repeated practice under congruent condition whereas the superior temporal lobe continuously played a key role in processing conflicting sensory information (Wang et al., 2022). These results suggested nonlinear dynamics of cortical activity differed after balance practice under congruent and incongruent environments.

Since the present study aimed to investigate the effect of visual-vestibular conflicts on the dynamics of postural control, we chose the data from acceleration and deceleration phases, necessitating the stimulation of the vestibular system. Thus, one limitation is that we did not further explore the dynamic changes from other phases, such as baseline and recovery phases. However, it would be beneficial for further studies to investigate the dynamic changes after perturbations (i.e., the recovery phase) using complexity-based nonlinear methods, since previous studies showed that perturbations had a noticeable after-effect on postural control among young and older adults (Doumas and Krampe, 2010; van den Hoorn et al., 2018).