No evidence for modulation of frontal brain activity asymmetry by a single session of EEG feedback

Abstract

Recent studies suggest that frontal hemispheric asymmetry may underlie various mental health disorders, and frontal alpha asymmetry (FAA), which reflects cortical activity and inactivity between the left and right frontal lobes, could be a potential biomarker for these conditions. In this research, we investigated whether a single session of electroencephalogram (EEG) feedback (EF) can modulate and shift FAA. We designed a preregistered, triple-blind randomized-controlled trial to address this gap in the literature. Sixty-five healthy individuals (M_age = 24.55, SD_age = 7.63) were recruited for the experiment. First, we assessed baseline resting-state FAA over a 10-min period, consisting of 5 min each under eyes-open and eyes-closed conditions. Subsequently, participants were assigned to one of two 30-min-long EF protocols, designed to modulate cortical activity by enhancing activation in either the right or left frontal hemisphere. FAA was reassessed immediately after the completion of the feedback intervention. The results indicated that a single session of EF does not modulate FAA; therefore, it should be considered with caution regarding causal inferences. However, the absence of effects may also be attributed to the form of feedback used, as well as individual differences in baseline brain activity and neurocognitive/psychological profiles.

1. Introduction

Previous research suggests that frontal hemispheric asymmetry may be associated with various mental health disorders (Giorgio Vallortigara, 2005; Kelley et al., 2017; Keune et al., 2011, 2012). This asymmetric frontal cortical activity can be measured using electroencephalogram (EEG), and frontal alpha asymmetry (FAA), the difference in alpha power (8–13 Hz (Hz)), which reflects cortical inactivity between the left and right frontal lobes, may serve as a biomarker of relevant disorders. However, the specific effects of a single-session EEG feedback (EF) on FAA have yet to be clearly established.

EF is considered one of the promising non-invasive brain modulation techniques. During EF, individuals receive continuous feedback displayed on a computer screen based on real-time analysis of their current brain activity. Specifically, to achieve a targeted pattern of brain activity, individuals are provided with rewarding feedback when the signals shift in the desired direction. This process enables individuals to learn the desired brain activity through operant conditioning via positive reinforcement (Enriquez-Geppert et al., 2017; Gruzelier, 2014).

Previous studies have suggested that alpha activity could be modulated by EF (Bazanova and Vernon, 2014; Ros et al., 2010). However, the effect of EF on FAA remains somewhat mixed. Some findings suggest that a single-session of EF may affect FAA within the session, but that these effects may not persist into post-intervention resting-state recordings (Peeters et al., 2014). In contrast, other studies report no immediate within-session changes, but instead observe alterations in resting-state FAA over time (Quaedflieg et al., 2015).

The current preregistered triple-blind randomized-controlled trial expands on a pilot investigation that offered preliminary evidence of EF-induced changes in resting-state FAA, utilizing a protocol adapted from previous studies (Peeters et al., 2014; Quaedflieg et al., 2015) and refined with a scoring-based incentive system to enhance reinforcement. We hypothesized that EF could be used to selectively modulate frontal brain asymmetry by targeting hemisphere-specific activity. In the left-frontal training condition (Protocol A), EF aimed to enhance activation in the left frontal regions, with asymmetry calculated as F3-F4. Participants in this condition were expected to show greater left-than-right frontal activation, resulting in an increased FAA score. Conversely, in the right-frontal training condition (Protocol B), feedback targeted increased activation in the right frontal cortex, calculated as F4-F3. This targeted modulation was expected to enhance right-over-left frontal brain activity, thereby producing a reduction in FAA.

2. Methods

2.1. Participants

A pilot study with 10 participants was conducted to validate our experimental procedure and inform the determination of the required sample size. A priori power analysis was performed using G∗Power (Faul et al., 2007, 2009), with a desired power of 80 percent, a significance level of 0.05, and a test-retest correlation of 0.6 for the effect of time (before and after EF) on FAA. Based on these parameters, an effect size of F > 0.237 (F > 0.237; η²p > 0.053) was found to be detectable with a sample size of 30 participants. A total of 65 participants (female = 45, M_age = 24.55, SD_age = 7.63) were recruited through social media and university courses. Inclusion criteria required participants to be at least 18 years old and to pass a screening that excluded individuals with psychological, psychiatric, or neurological disorders. Participants were also instructed to refrain from smoking and consuming caffeine for at least 2 h before the experiment. Eligibility was determined through self-reported responses. All participants gave written informed consent, and the study adhered to ethical principles. Each participant received either a voucher or course credit.

2.2. EEG data acquisition

Scalp voltage recordings were acquired using a 21-channel EEG cap equipped with Ag/AgCl electrodes, arranged according to the international 10–20 system. Data collection was conducted with the Nexus-32 system from Mind Media (Nexus-32, n.d.). EEG signals were recorded using a common average reference configuration, and were re-referenced to linked mastoids, both for pre- and post-intervention computation of FAA, and for online feedback. To monitor eye movements and blinks, vertical electrooculography (vEOG) was recorded from electrodes positioned above and below the left eye, while horizontal electrooculography (hEOG) was measured using electrodes placed at the outer corners of both eyes. Additionally, we instructed our participants to avoid eyes movements and blinks before each session. The sampling rate was 512 Hz.

2.3. Frontal alpha asymmetry

FAA scores were derived from EEG data collected under three distinct conditions. Initially, resting-state EEG recordings were obtained during two separate 5-min sessions, one with eyes open (EO) and the other with eyes closed (EC), administered both before and after the intervention. Capturing EEG under both EO and EC conditions allows for a more nuanced understanding of how sensory input, intrinsic brain activity, and cognitive processes contribute to neural function, thereby offering a broader perspective on the brain's functional organization (Barry et al., 2007). In addition to the resting-state sessions, FAA scores were also computed during EF interventions.

Preprocessing of FAA data was carried out using BrainVision Analyzer 2 (www.brainproducts.com), following established protocols (Smith et al., 2016). The initial steps included applying a high-pass filter at 0.5 Hz, a low-pass filter at 40 Hz, and a notch filter at 50 Hz to eliminate line noise. The first and last 10 s of each recording were excluded due to potential artifacts. The remaining data were segmented into 2-s epochs. For the EO condition, ocular artifacts were corrected using Independent Component Analysis (ICA), based on signals from vEOG and hEOG channels. Epochs containing residual artifacts, defined as exceeding ±75 μV in maximum amplitude relative to baseline, were excluded from further analysis. To assess spectral activity, Power Spectral Density (PSD) was calculated using Fast Fourier Transform (FFT) with a 10 % Hanning window, following baseline correction of each epoch. The resulting data were averaged, and mean alpha band power (8–13 Hz) was extracted for relevant electrode sites (i.e., F3 and F4).

In R (R Software: A Tool Analysing Experimental Data, 2016), alpha power values were log-transformed to correct for skewness. FAA was then computed by subtracting the alpha power at left frontal sites from their right-hemisphere counterparts, specifically F4-F3.

2.4. EEG feedback

We employed the Mind Media Nexus-32 feedback System (Nexus-32, n.d.) to modulate FAA, specifically targeting the activation of either the right or left frontal cortex in two separate intervention groups based on previous research (Quaedflieg et al., 2015). Participants were randomly assigned to one of the protocols, with the goal of regulating their brain activity for a duration of 30 min. We applied an 8–13 Hz IIR Butterworth Bandpass filter, and Root Mean Square (RMS) of the signal was computed online for 250 ms. Epochs for activity recorded at both F4 and F3 electrodes. Participants received continuous feedback for 30 min on the degree of asymmetry between F4 and F3. For both groups, the goal was to increase the height of a thermometer-like bar. For the group that received Protocol A, FAA was computed as alpha amplitude at F3-F4, for the group that received Protocol B, FAA was computed as the inverse, F4-F3. For positive reinforcement, participants acquired points for the time above the (bar) threshold, indicated by a black horizontal line that was updated every 15 s to maintain an 80 % above-threshold success.

2.5. Statistical analysis

Data analyses were conducted using R (R Software: A Tool Analysing Experimental Data, 2016) and Python 3.13 (Programming Language). Following the computation of FAA, as described previously, participants with missing data were excluded to ensure data quality. We first conducted a 2 × 2 repeated-measures analysis of variance (ANOVA) to examine the effect of EF on FAA. Additionally, a 2 × 3 was performed, incorporating FAA during the intervention (INT) as a time factor. We also conducted Tukey-adjusted pair-wise comparisons for statistically significant (i.e., p < 0.05) results in the main analyses. Exploratory analyses involved excluding participants with scores more than three standard deviations (SDs) from the mean, conducting linear mixed-effects model analyses, and performing spectral parameterization.

Specifically, linear mixed-effects models (LMMs) are a type of mixed model that include a random intercept for each participant, allowing to account for individual variability and ensuring acceptable reliability (Baayen et al., 2008; Schielzeth et al., 2020). We implemented LMMs using the lme 4 package in R, with the following model specification: FAA ∼ time x group + (1 | ID).

To account for individual variability in oscillatory peaks, we employed the Fitting Oscillations and One-Over-F (FOOOF) package in Python (Donoghue et al., 2020), a method for spectral parameterization that decomposes neural power spectra into periodic (oscillatory) and aperiodic (1/f-like) components (e.g., Monchy et al., 2024). We extracted oscillatory peaks (within the alpha frequency band) from the underlying aperiodic background activity for each participant and conducted the main analyses with the new power spectrum again.

3. Results

First, the results revealed a statistically significant relationship between EF and FAA, F (1, 106) = 4.34, p = 0.039, η²p = 0.039 (see Table 1). Specifically, Protocol A was associated with an increase in FAA, reflecting greater left frontal activity, whereas Protocol B led to a reduction in FAA, indicative of enhanced right frontal activity (see Fig. 1). However, this result may be attributed to the baseline difference that is reversed over time. To check this and examine potential online effects of EF, we included FAA measured during the EF task as an additional level of the time factor in the model. All together, the results did not show any difference between online FAA and post-intervention FAA (see Table 1 and Fig. 2), and the results were most likely due to the baseline difference.

Table 1.

Results of the frontal alpha asymmetry models.

Models	df	F	p	η2p
FAA F4-F3 (EO) (n = 55)
Time	1	0.2	0.648	0.001
Group	1	2.27	0.134	0.021
Time x Group	1	4.34	0.039∗	0.039
Residuals	106
FAA F4-F3 (EC) (n = 58)
Time	1	1.42	0.236	0.012
Group	1	0.31	0.579	0.002
Time x Group	1	0.65	0.419	0.005
Residuals	112
FAA F4-F3 (EO + INT) (n = 55)
Time	1	0.31	0.73	0.003
Group	1	2.62	0.107	0.016
Time x Group	1	2.69	0.07	0.032
Residuals	159

Note: Participants with missing data were excluded from the analyses. The EO + INT model incorporated online FAA during EF as a time factor, in addition to the pre- and post-intervention measurements.

Fig. 1.

The figure illustrates FAA values, calculated from F4-F3 electrode sites under the EO condition across the two intervention groups.

Fig. 2.

The figure illustrates FAA values, calculated from F4-F3 electrode sites under the EO condition across the two intervention groups. It also incorporates the online FAA during EF as a time factor, in addition to the pre- and post-intervention measurements.

We also conducted a post-hoc analysis to further examine the main conclusion. Specifically, Tukey-adjusted pairwise comparisons were performed, and the outcome supported our initial interpretation. The only notable trend toward significance was observed between baseline (pre-intervention) measurements in Group A and Group B, t (-2.54) = -0.234, p = 0.059 (see Table 2).

Table 2.

Results of the pairwise comparisons.

Contrasts	Estimate	SE	df	T-ratio	p
FAA F4-F3 (EO) (n = 55)
Pre Group A - Post Group A	−0.172	0.094	106	−1.827	0.266
Pre Group A - Pre Group B	−0.234	0.092	106	−2.54	0.059
Pre Group A - Post Group B	−0.134	0.092	106	−1.47	0.459
Post Group A - Pre Group B	−0.061	0.092	106	−0.664	0.91
Post Group A - Post Group B	0.037	0.092	106	0.406	0.977
Pre Group B - Post Group B	0.098	0.089	106	1.101	0.689
FAA F4-F3 (EO + INT) (n = 55)
Pre Group A - INT Group A	−0.083	−0.083	159	−0.962	0.929
Pre Group A - INT Group B	−0.125	−0.125	159	−1.481	0.676
INT Group A - Post Group A	−0.088	−0.088	159	−1.019	0.911
INT Group A - Pre Group B	−0.15	−0.15	159	−1.766	0.49
INT Group A - INT Group B	−0.041	−0.041	159	−0.493	0.996
INT Group A - Post Group B	−0.051	−0.051	159	−0.606	0.99
Post Group A - INT Group B	0.047	0.047	159	0.553	0.993
Pre Group B - INT Group B	0.108	0.108	159	1.309	0.779
INT Group B - Post Group B	−0.009	−0.009	159	−0.116	1

Note: INT indicates online FAA during EF added as a time factor.

Following our preliminary results, we conducted several exploratory analyses. First, we repeated the main analysis after excluding participants whose values exceeded three SDs from the mean. The results were consistent with the main analysis and did not provide additional evidence for the hypothesized effects (F (1, 100) = 5.79, p = 0.017, η²p = 0.054, see Table 3 in the supplementary materials). Visual inspection suggested that the observed significant effects were likely driven by baseline differences between groups (see Figs. 3 and 4 in the supplementary materials).

Secondly, we performed LMMs with random intercepts to account for individual variability, particularly in the baseline condition. The results indicated that baseline FAA was significantly higher in Group B compared to Group A, p = 0.012 and online FAA (during EF) was not significantly different than the baseline, p = 0.079 (see Table 4 in the supplementary materials).

Lastly, although most participants exhibited identifiable alpha oscillatory peaks based on spectral parameterization in each experimental condition (see Table 5 in the supplementary materials), additional participant exclusions were necessary because the asymmetry score was calculated based on electrode pairs. Additionally, the results from the spectral parameterization indicated an overall strong model fit (R² = 0.963). However, subsequent analyses, specifically the repeated-measures ANOVA, did not provide additional support for our hypotheses, as the interaction effects were non-significant (Table 6 in the supplementary materials). This may also be attributed to data loss (i.e., lack of alpha power) in certain electrodes, given that spectral parameterization could be considered a more conservative approach than excluding participants based on standard deviation criteria. The lack of observable effects from a single session of EEG neurofeedback may also have contributed to this outcome.

4. Discussion

The present study explored the effectiveness of EF in modulating the asymmetry of frontal brain activity. We implemented two distinct neurofeedback protocols targeting either right or left frontal cortical activity based on previous research (Peeters et al., 2014; Quaedflieg et al., 2015).

While prior studies have provided some support for the role of EF in modulating FAA, and our preliminary pilot data suggested a potential effect, our implementation did not yield the intended outcome. Notably, there has been ongoing debate regarding the reliability of auto-thresholding methods and the interpretation of pre-to post-session changes following a single session of EF (Peeters et al., 2014).

Alternatively, some previous findings may have been influenced by issues related to blinding or differences in feedback implementation, as highlighted in a study (Peeters et al., 2014) and later addressed in their follow-up study (Quaedflieg et al., 2015). For instance, in a study, participants were given different instructions across groups (e.g., to increase vs. decrease the feedback bar) (Peeters et al., 2014), raising concerns that observed effects could be attributable to differences in task demands rather than EF itself, compounded by the lack of experimenter blinding. In contrast, in our study, we controlled for these potential confounds by ensuring that all participants received identical instructions and tasks, with experimenters blind to condition.

The results may have been influenced by individual differences in neurocognitive abilities and psychological profiles. Some individuals may be inherently more or less responsive to the feedback intervention. Prior research has identified several potential predictors of feedback outcomes, including age, brain volume (Moretti et al., 2004), motivation (Kleih and Kübler, 2013), and mood states (Subramaniam and Vinogradov, 2013). Additionally, it was indicated that there is no single effective strategy (Roberts et al., 1989).

Second, the timing and nature of the EF protocol may have influenced the outcomes. EF involves a process of training individuals to increase or decrease power, and its effects may depend on the duration and intensity. Therefore, the effects of EF may be more evident over longer periods of training or in the context of repeated sessions, rather than in a single session as employed in our study (Quaedflieg et al., 2015). For instance, Mennella et al. (2017) found that 7-week FAA neurofeedback training results in changes in FAA, in addition to the reduction of negative feelings. However, it is important to acknowledge the risk of overtraining, noting that excessive practice can potentially diminish the overall effectiveness of neurofeedback (Strehl, 2014). On the other hand, Wang et al. (2019) found that even though there was no change in FAA after ten sessions of neurofeedback, it reduced the depression symptoms.

Future studies should carefully consider the design of the protocol. As noted earlier, variations in feedback type and session duration may lead to different outcomes. While our study focused on a single-session intervention, it is possible that the effects of feedback on FAA become more pronounced over time. Longitudinal studies examining the cumulative effects of feedback could provide more insight. Lastly, a more individualized approach could be an important factor. While recent studies started using machine learning in detecting brain asymmetries and EEG-based biomarkers (e.g., Yu et al. (2020), Herzog and Magoulas (2021), and Ryu et al. (2024)), to the best of our knowledge, no research has specifically examined EEG-feedback acceptance or individual variability in these contexts. Addressing these factors could be a valuable direction for future research, enabling more personalized and effective interventions.

CRediT authorship contribution statement

Atakan M. Akil: Writing – original draft, Visualization, Project administration, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization. Renáta Cserjési: Writing – review & editing, Validation, Supervision, Resources, Methodology, Investigation, Conceptualization. Tamás Nagy: Writing – review & editing, Validation, Supervision, Resources, Methodology, Investigation, Funding acquisition, Conceptualization. Zsolt Demetrovics: Writing – review & editing, Validation, Supervision, Resources, Methodology, Investigation, Conceptualization. H.N. Alexander Logemann: Writing – review & editing, Validation, Supervision, Resources, Project administration, Methodology, Investigation, Funding acquisition, Conceptualization.

Ethics information

The research was conducted following the ethical guidelines outlined in the Declaration of Helsinki and its later amendments. The study was approved by the Institutional Review Board at Eötvös Loránd University (protocol number: 2020/403). Each participant was given a voucher or course credit for their participation.

Data availability statement

Please access all the materials from our public repository:

https://osf.io/ktsud.

Declaration of generative AI and AI-assisted technologies in the manuscript preparation process

During the preparation of this work the author(s) used ChatGPT in order to improve readabilty. After using this tool/service, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the published article.

Funding information

AMA was supported by the OTKA FK 146604 research grant. TN was supported by the University Excellence Fund of Eötvös Loránd University, Budapest, Hungary (ELTE), and the János Bolyai research fellowship of the Hungarian Academy of Sciences. HNAL was supported by the Hungarian National Research, Development and Innovation Office (https://nkfih.gov.hu; grant no. K131635).

Declaration of competing interest

The authors have no conflicts of interest to declare.

Appendix A. Supplementary data

The following is the Supplementary data to this article:

Data availability

We have shared the link to our data/code in the manuscript.