Data were collected from 20 healthy participants (9 female, 18–35 years, M ± SD: 24.9 ± 4.2 years), all right-handed with normal or corrected vision, and no neurological or psychiatric conditions. Participants had not taken part in previous MI-NF studies and provided written informed consent. Handedness was assessed using the Edinburgh Handedness Inventory (Oldfield, 1971). To evaluate participants’ ability to engage in visual and kinesthetic MI, we used the short version of the Kinesthetic and Visual Imagery Questionnaire (KVIQ-10), which measures the vividness of both imagery modalities separately (Malouin et al., 2007). Three participants were excluded for not meeting inclusion criteria (two for signs of depression; one for ambiguous handedness), leaving data from 17 participants (8 female, 18–35 years, M ± SD: 25.2 ± 4.2 years) for analysis. These data was compared to 17 previously collected datasets (10 female, 23–32 years, M ± SD: 25.8 ± 2.5 years) (Daeglau et al., 2021b).

Each participant completed three sessions: One in the morning, one in the evening on the same day, and one on the following morning (Figure 1). In the first session, participants practiced a sequential finger-tapping task for MI, moving each finger to touch their thumb at a rate of 1 Hz with both hands. Each session started with a 2-min resting-state measurement with eyes open, followed by an MI block consisting of three runs. During MI, participants were asked to kinesthetically imagine the finger-tapping task from a first-person perspective, focusing on the sensations they would feel if they were physically performing the task.

In the first run of the MI block, participants mentally rehearsed the finger-tapping task without receiving NF. The EEG data collected in the first run were used to train the NF parameters for the second run and is therefore referred to as training run (T). In the following two NF runs (NF1 and NF2), MI was supported by real-time NF, which visualized individual ERD strength.

For the quiet rest group, NF2 was followed by a 10-min quiet rest period, during which participants relaxed and focused on a fixation point on the screen. For the documentaries group, the overall session structure was identical to that of the quiet rest group, except for the post-MI condition. Here, NF2 was followed by two 30-min blocks, during which participants passively watched a nature documentary. The quiet rest period was designed to provide a controlled post-MI resting phase, aligning with motor learning research suggesting that short post-training waking rest supports skill consolidation (Hotermans et al., 2006; Brokaw et al., 2016; Humiston and Wamsley, 2018). Extending this period to 60 min, as in the documentaries group, was deemed impractical due to the potential for boredom and disengagement.

The second session of the study replicated the structure of the first session. In the third session, participants only completed a 2-min resting-state measurement and one MI block (three runs: T, NF1, NF2). Participants were instructed to maintain their regular daily routines between sessions, avoiding exhausting exercise and naps.

The experiment was controlled using OpenViBE 0.17.1 (Renard et al., 2010), which was also used in the original study by Daeglau et al. (2021b). The procedure was identical for both the quiet rest and documentaries groups. Each 10-min run consisted of 40 trials (20 left-hand trials, 20 right-hand trials) in pseudorandomized order (Figure 2). Each trial began with a 5-s baseline period, followed by a 3-s preparation cue. During the 5-s MI interval, blue graphics appeared on either the left or right side, indicating which hand should be used for the MI of the finger-tapping task. Trials were separated by an inter-trial interval ranging from 0 to 4 s.

When feedback was provided, a white ball moved along the x-axis (representing ERD lateralization) and the y-axis (representing contralateral ERD strength) based on two linear discriminant analysis classification results. Participants used kinesthetic MI of the finger-tapping task to steer the ball to the upper left corner for left-hand trials and the upper right corner for right-hand trials. They were instructed to focus on the fixation symbols, remain relaxed, and minimize physical movement.

Settings and procedures for group quiet rest were matched as closely as possible to those of group documentaries in all aspects except for the post-MI condition. Both groups followed the same experimental procedure, including inclusion criteria, identical laboratory environment, pre-experiment assessments (Edinburgh Handedness Inventory, KVIQ-10), MI task and structure of MI blocks. A standardized instruction sheet was used to ensure consistency in explaining the experimental procedure and visualizing the task for participants. The sessions for both groups were scheduled at the same times of day, and the overall session structure was kept as identical as possible. The only systematic difference between groups was the post-MI condition and, consequently, the session duration. Additionally, data collection for the two groups was conducted by different female experimenters.

EEG and surface electromyography data were recorded using a BrainAmp Amplifier System (Brain Products GmbH, Gilching, Germany). EEG settings included a 0.1 μV amplitude resolution, 500 Hz sampling rate, and 0.016–250 Hz online analogue filter. Impedance was reduced to below 10 kΩ. EEG data were collected from 64 sintered Ag/AgCl electrodes arranged in an equidistant layout on a customized, infracerebral electrode cap (EasyCap, Herrsching, Germany; caps that include electrode positions below the cerebrum), with a central frontopolar electrode as ground, a nose-tip reference, and two electrodes below the eyes to capture eye movements.

Electromyography data, measuring muscle activity of both hands, had impedance reduced to below 100 kΩ. Electrodes were placed over the muscle belly and the proximal base of the Flexor digitorum superficialis and the Abductor pollicis longus, with clavicle electrodes serving as reference and ground. Data acquisition during MI was managed with OpenViBE Server 0.17.1, while BrainVision Recorder (version 1.20.0506) was used for resting-state recordings.

Online and offline processing of the EEG and electromyography data followed the same protocol and utilized the same software as in Daeglau et al. (2021b), ensuring methodological consistency and allowing for a direct comparison between the newly recorded dataset and the existing one. The primary distinction between the two groups was the altered post-MI condition.

Paired t-tests were used to compare training run ERDs (T-MI-ERD) with averaged NF run ERDs (NF-MI-ERD) to assess the impact of NF on MI-ERD, under the assumption that T-MI-ERD would be less negative than NF-MI-ERD (measure 1 > measure 2). A repeated measures ANOVA with block as the repeated measures factor and NF-MI-ERD as level was used to evaluate across-block gains in NF-MI-ERD. To compare NF-MI-ERD between the quiet rest and documentaries groups, another repeated measures ANOVA was conducted with blocks as repeated measures factor, NF-MI-ERD as level, and group as a between-subjects factor. Mauchly’s test (Mauchly, 1940) was used to assess sphericity, and Greenhouse–Geisser corrections were applied when necessary. Interaction effects were investigated with independent t-tests comparing NF-MI-ERDs between groups within each block. Effect sizes were reported as eta-squared (η²) for ANOVAs (90% CI) and Cohen’s d (d) for t-tests (95% CI) (Steiger, 2004). The Bonferroni-Holm method was applied to correct for multiple comparisons (Holm, 1979). Additionally, a baseline power analysis was conducted. The Shapiro–Wilk test was used to assess normality, and Levene’s test examined the equality of variances. Independent samples t-tests were performed to compare baseline power between the quiet rest and documentaries groups within the same MI block.

The effects of NF on MI-ERD within each block were assessed using paired t-tests, comparing the means of NF-MI-ERD with those of T-MI-ERD (Table 1). The results revealed a significantly stronger expression of ERD in all three blocks when MI was supported by NF (df = 16, p < 0.001, d > 0.9 for each block), indicating that NF reliably enhanced ERD.

The NF-MI-ERD time courses for all participants across each block are illustrated in Figure 3. The characteristic desynchronization of brain activity during MI was observed consistently across all participants and blocks (Figure 3A).

To further assess the development of NF-related ERD gains across blocks, a repeated measures ANOVA was conducted. The analysis showed no statistically significant changes in NF-MI-ERD over the three blocks (F_(2,32) = 1.735, p = 0.193, η² = 0.098). Notably, NF-MI-ERDs were highly variable, with only two participants showing a linear increase in ERD over time, as highlighted by the solid grey lines in Figure 3B.

The time courses of NF-MI-ERD for both the quiet rest and documentaries groups, along with the distribution of single-subject mean NF-MI-ERD values are illustrated in Figure 4. ERD was evident in both groups throughout the experiment, with the most pronounced differences occurring in block 2 (Figure 4A). However, these differences were subtle and did not display a clear trend across the blocks. The raincloud plots reveal that the distribution of mean NF-MI-ERD values is broader in the documentaries group compared to the quiet rest group, indicating greater variability in the ERD response across all blocks (Figure 4B).

A baseline power comparison was conducted to test for potential differences between groups. Prior to this, the Shapiro–Wilk test indicated that baseline power values were normally distributed in both groups across all blocks (all p > 0.05) and Levene’s test confirmed that variance was homogenous between groups (all p > 0.05). No significant differences in baseline power were found within the same MI blocks (Block 1: df = 32, p = 0.28 d = −0.377; Block 2: df = 32, p = 0.106, d = −0.571; Block 3: df = 32, p = 0.321, d = 0.346). These results confirm that baseline EEG activity was comparable between groups.

To assess the effect of NF across blocks and compare the two groups, a repeated measures ANOVA was conducted. The analysis showed no statistically significant main effect of NF across blocks (F_(2,64) = 1.831, p = 0.169, η² = 0.010), nor was there a significant difference between groups (F_(1,32) = 0.055, p = 0.815, η² = 0.001). However, a significant interaction effect between block and group was observed, albeit with a small effect size (F_(2,64) = 3.526, p = 0.035, η² = 0.019). This interaction was further investigated by post-hoc independent t-tests to determine if the NF-MI-ERDs differed between groups within each block (Table 2). Bonferroni-Holm corrections for multiple comparisons were applied. No significant differences were found between the quiet rest and documentaries groups in any of the individual blocks (Block 1: df = 32, p = 0.7, d = 0.325; Block 2: df = 32, p = 0.885, d = −0.365; Block 3: df = 32, p = 0.822, d = −0.078).

One possible explanation for the observed results is the simplicity of the finger-tapping task, which was chosen as the movement sequence for MI. Participants were instructed to imagine their thumb sequentially touching the index, middle, ring, and little finger at a rate of one touch per second. Importantly, the task did not require participants to memorize a complex sequence of finger movements. Before the first MI block, the task was demonstrated and physically practiced until participants felt comfortable with the sequence and the required pace of 1 Hz. Due to the simplicity of the task, participants learned it quickly. Additionally, during the KVIQ assessment, all participants reported an understanding of the difference between visual and kinesthetic MI and indicated being capable of applying kinesthetic MI. Given this, along with the presence of a typical ERD across all participants in all MI blocks, particularly in the first MI block (see Figure 3A), we are confident that participants were able to perform the kinesthetic MI of the finger tapping task effectively.

Research shows that complex or varied movements can enhance motor performance and learning after MI training (Allami et al., 2008; Rozand et al., 2016; Ruffino et al., 2017). The finger-tapping task, used in various MI-NF studies (Korman et al., 2007; Debarnot et al., 2009; Zich et al., 2015), may have allowed participants to quickly transition into the automatic phase of motor learning. This would reduce cognitive load, potentially diminishing the increase in ERD. As Pfurtscheller and Lopes da Silva (1999) reported, ERD is strongest when a movement sequence is newly learned but decreases once the movement becomes automatic. Similarly, Wolpaw and McFarland (2004) showed that brain-computer interface users initially rely on MI to generate control-relevant EEG activity, but as their performance improves, control becomes more automatic. Although their study involved a larger training paradigm with more sessions and trials, the underlying principle, that learned motor tasks require less cognitive effort and exhibit reduced ERD, might also apply for the present study. This could explain the slight trend towards a less strong ERD observed when averaging NF-MI-ERD across participants in the quiet rest group (Figure 3B).

Additionally, the type of NF used could also have impacted ERD development. Studies have shown that more realistic and multidimensional NF approaches can promote stronger ERD (Ono et al., 2013; Liang et al., 2016; Braun et al., 2016; McWhinney et al., 2018) compared to two-dimensional, object-based NF. For instance, Ono et al. (2013) demonstrated that a three-dimensional anatomically congruent NF elicited greater ERD than conventional bar-based NF. Similarly, Liang et al. (2016) found that adding multimodal sensory feedback, such as vibrotactile stimulation, enhanced MI-based NF performance. Braun et al. (2016) showed that virtual reality NF improved motor-related brain activity more effectively than traditional two-dimensional feedback, while McWhinney et al. (2018) reported that task engagement was higher when NF provided richer and more interactive cues. Therefore, NF methods that incorporate more complex, dynamic feedback or that engage multiple sensory modalities may provide richer information and thereby enhance the brain’s ability to modulate ERD more effectively. The relatively simplistic NF design used in this study, with limited feedback dimensionality, might have constrained the participants’ ability to fully engage with the task and optimize ERD. Incorporating more challenging MI tasks alongside a more sophisticated, multidimensional NF approach could potentially improve the development of contralateral ERD across blocks, leading to more pronounced changes in brain activity over time.

Another important consideration is the pronounced inter-and intra-individual variability in ERD, a well-known phenomenon in MI research (Ahn and Jun, 2015; Wriessnegger et al., 2020; Leeuwis et al., 2021), that could affect MI/MI-NF outcomes. Such differences may arise from context factors within and beyond the experimental task, including character traits, imagery strategies and capability, task engagement, time-variant cognitive factors, brain structure, and session structure (Rulleau et al., 2015; Jeunet et al., 2015; Ahn and Jun, 2015; Rulleau et al., 2018; McWhinney et al., 2018; Lee et al., 2019; Saha and Baumert, 2019; Daeglau et al., 2021a; Leeuwis et al., 2021).

Given this variability in ERD responses, we applied CSP filtering as a preprocessing step to enhance the spatial specificity of MI-related neural activity by maximizing the discriminability of oscillatory patterns associated with MI. This is a common approach in MI-NF studies (Ramoser et al., 2000) because it optimally extracts the most relevant task-related EEG components, reducing volume conduction effects and improving the signal-to-noise ratio compared to single-channel analyses.

In this study, all participants in the quiet rest group exhibited detectable NF-MI-ERD in each block, yet substantial variability was evident across individuals, although CSP filtering was employed. When analyzing averaged values across blocks, the progression of ERD was notably inconsistent. Some participants showed a reduction in ERD from Block 1 to Block 2, while others displayed the opposite trend, with an increase in ERD. Similarly, transitions from Block 2 to Block 3 reflected comparable variability. Only two participants demonstrated the expected pattern, where NF-MI-ERD became progressively more negative from Block 1 to Block 3, aligning with the anticipated NF-gain. A similar pattern was observed in the documentaries group, where variability in ERD expression was even more pronounced (see Figure 4B). In particular, Block 2 and Block 3 revealed substantial differences between participants, with some exhibiting strong ERD while others showed markedly weaker responses.

This pronounced inter-and intra-individual variability underscores the complexity of achieving consistent ERD progression during MI-NF training and highlights the necessity of individualized approaches to optimize training outcomes.

While this study provides valuable insights into ERD development during MI-NF training, there are several factors to consider for future research. The sample size, though adequate for initial analysis, could be expanded to capture more subtle effects. One notable difference between the two groups was the length of the post-MI condition. The MI blocks were identical in both groups, but the documentaries group had a longer post-MI session, with 30 min of passively watching nature documentaries used as both a break and control condition, whereas the quiet rest group had a 10-min resting-state measurement. Despite the shorter session length after the MI blocks in the quiet rest group, our results showed no significant differences in ERD development between the two groups. This suggests that the length of the session following MI-NF may not have a strong impact on ERD outcomes in this context.

Furthermore, some limitations should be considered when comparing the two groups. From the study’s inception, we ensured that all procedures in the quiet rest group paralleled those of the documentaries group as closely as possible. Both groups were matched in terms of sample size, age, and gender distribution to minimize demographic biases. Additionally, the same NF protocol and analysis pipeline were applied to both groups to maintain consistency. However, some differences remain. Data collection was conducted by two different experimenters, but both were of the same gender, allowing us to rule out potential differences in gender-related experimenter effects (Roc et al., 2019). Additionally, there was a time gap between the data collection periods of the two groups, which could have introduced slight variations in the population from which participants were recruited or laboratory conditions. While every effort was made to standardize the experimental setup, these factors may have contributed to variability in the results and should be considered when interpreting the findings.