Sensorimotor rhythm neurofeedback training relieves anxiety in healthy people

Abstract

Timely relief of anxiety in healthy people is important, but there is little research on this topic at present. Neurofeedback training allows subjects to regulate their specific brain activities autonomously and thus alter their corresponding cognitive functions. Inattention is a significant cognitive deficit in patients with anxiety. Sensorimotor rhythm (SMR) was reported to be closely related to attention. In this study, trainability, frequency specificity, and brain-behavior relationships were utilized to verify the validity of a relative SMR power protocol. An EEG neurofeedback training system was developed for alleviating anxiety levels in healthy people. The EEG data were collected from 33 subjects during SMR up-training sessions. Subjects attended six times neurofeedback training for about 2 weeks. The feedback value of the neurofeedback group was the relative SMR power at the feedback electrode (electrode C3), while the feedback values for the control group were pseudorandom numbers. The trainability index revealed that the learning trend showed an increase in SMR power activity at the C3 electrode, confirming effects across training. The frequency specificity index revealed only that SMR band activity increased significantly in the neurofeedback group. The brain-behavior relationships index revealed that increased SMR activity correlated negatively with the severity of anxiety. This study indicates that neurofeedback training using a relative SMR power protocol, based on activity at the C3 electrode, could relieve anxiety levels for healthy people and increase the SMR power. Preliminary studies support the feasibility and efficacy of the relative SMR power protocol for healthy people with anxiety.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11571-021-09732-8.

Introduction

Anxiety disorders are the most prevalent mental disorders and are associated with immense healthcare costs and a high burden of disease (Lu et al. 2017). Continued excessive worry, difficulty concentrating, sleep abnormalities, emotional lability, fatigue, and restlessness are significant clinical manifestations in patients with anxiety (Blaskovits et al. 2017). In the light of large population-based surveys, up to 33.7% of the population are affected by anxiety disorders during their lifetime in the twenty-first century (Bandelow and Michaelis 2015). Only one out of six (16%) patients with anxiety received effective treatment, owing to the limitations of maladaptive cognitive and behavioral therapy (Jonge et al. 2017; Weerdmeester et al. 2019). Thus, a better understanding of interventions clinically that ameliorate anxiety and stress is urgently needed, given their negative consequences on human health (Danielsson et al. 2012). Healthy people who are in a high level of stress and anxiety for a long time may lead to anxiety disorders or other symptoms (Wells and Matthews 2006; Wiseman et al. 2015). Hence, timely intervention to treat anxiety in healthy people is a promising and possible way to reduce the prevalence of anxiety disorders.

There are two conventional treatment modalities in clinical psychiatry: pharmacological treatment and psychotherapy. Pharmacological treatment employs drugs to reduce psychiatric symptoms, and its effectiveness varies across disorders and medications with adverse outcomes, such as dry mouth, headache, or nausea, which can hamper the quality of life (Cheon et al. 2015; Damsker et al. 2009). Another conventional treatment modality is psychotherapy, which shows empirical effectiveness, although there is some difficulty in verifying the effectiveness through experimental studies (Schueller et al. 2017; Cuijpers et al. 2008). However, the era of the ‘quantified self’ is upon us (Lupton 2016). In the field of medicine and mental health treatment, people have become increasingly interested in self-monitoring technology in recent years. (Piwek et al. 2016; Schueller et al. 2017; Weerdmeester et al. 2019; Yetisen et al. 2018). To enhance treatment effectiveness and address the limitations of conventional treatment modalities, many complementary treatments have been proposed, of which neurofeedback is one of the most sophisticated methods (Cheon et al. 2015).

The ability for training to control brain electrical activity has been demonstrated since 1960 (Kamiya 1968, 1969). Owing to its excellent temporal resolution (milliseconds or less) (Sitaram et al. 2016; Thibault et al. 2015), electroencephalography (EEG) biofeedback, known as EEG neurofeedback has been utilized to make some improvements in some disorders (Bonnet et al. 2017; Coben et al. 2010; Micoulaud-Franchi et al. 2015; Nigro 2019; Yan et al. 2019); it is a therapeutic technique in which subjects are tasked with regulating their brain activity autonomously (Johnston et al. 2010; Maurizio et al. 2014; Nazari et al. 2011; Thibault et al. 2015). In clinical, EEG neurofeedback has been used in attention deficit and hyperactivity disorders (Arns et al. 2009, 2014; Gevensleben et al. 2009), depressive disorders (Choi et al. 2011; Hammond 2005a), anxiety disorders (Hammond 2005b), and sleep disorders (Arns and Kenemans 2012; Cortoos et al. 2010).

The sensorimotor rhythm (SMR) training protocol is known to enhance attention which emerges when one is motionless yet remains attentive and is suppressed by movement in the human brain recorded over central scalp regions. (Lubar and Lubar 1984; Pfurtscheller 1981; Reichert et al. 2016; Sterman et al. 1970). It has a frequency range of 12–15 Hz and it has also been proven to be an effective frequency with benefit for anxiety but not widely in clinical due to its long duration, slower effects, and individual difference (Gadea et al. 2020; Gomes et al. 2016; Gruzelier 2014b; Reichert et al. 2016; Ros et al. 2009). A large body of literature showed that the SMR neurofeedback training reduces inattentive and hyperactive/impulsive symptoms in attention-deficit hyperactivity disorder (ADHD) children clinically (Enriquez-Geppert et al. 2019). Studies suggest that increasing SMR activity voluntarily employing neurofeedback training (NFT) has also positive effects on the attentional performance of healthy subjects (Egner and Gruzelier 2004; Vernon et al. 2003). One possible reason is that the circuitry of SMR is a thalami-cortical, bottom-up mechanism, and the SMR neurofeedback training acts within the inhibitory mechanism of the thalamic circuitry. Driven by the increase in SMR, it improves the body's inhibition ability and reduces the interference of somatosensory information (Egner and Gruzelier 2004; Reichert et al. 2016). There is another reason that the improvements in ADHD symptoms following the SMR training might be the result of the vigilance stabilization mediated by the regulation of the locus coeruleus noradrenergic system of which activation has been shown to impact the sleep spindle circuitry (Sinha and Saurabh 2011). The SMR training increases sleep spindle density and improve sleep quality in healthy adults and ADHD patients trained with the SMR protocol showed decreased sleep onset latency and improved sleep quality responsible for the improved inattention (Arns et al. 2014; Enriquez-Geppert et al. 2019; Schabus et al. 2014; Veen et al. 2010).

Studies have shown that training aspects of attention can relieve anxiety (e.g., yoga, meditation) (Kiken et al. 2015; Shreve et al. 2020; Simon et al. 2020; Wuthrich et al. 2021). Besides, inattention is one of the clinical manifestations of anxiety (Blaskovits et al. 2017). It was hypothesized that improving attention-related SMR activity alleviates anxiety. In most cases, the SMR training protocol clinically used as feedback value to relieve anxiety is absolute power (Gomes et al. 2016; Gruzelier 2014a, 2014b; Ros et al. 2009). However, absolute power is affected confounded by scalp thickness and electrical resistance (Allen et al. 2004; Knyazev et al. 2004). To decrease data variability and better reflect cortical activity (Allen et al. 2004; Cook et al. 1998; Knyazev et al. 2004), the relative power value was proposed and used as the feedback value aim to increase SMR power designed to ease anxiety in a healthy population.

Concerning the reliability of NFT effects, trainability, independence, and interpretability first proposed by Zoefel in 2011 were three parameters for optimal research by which to validate the efficacy of NFT (Zoefel et al. 2011). In that case, these three parameters include effects concerning the EEG observed at all and expectations as well as significant behavioral impact. A change in the targeted band caused by NFT was defined as trainability. If the trainability is successful, the frequency specificity as we called independence in this study, which is to reflect the target frequency band, not other frequency bands of the neurofeedback group has to be assessed. Finally, interpretability namely brain-behavior relationships indicating that the targeted band was generated and was related to reliable behavioral effects need to be determined.

In this study, we proposed the relative power as the feedback value to increase the subjects’ SMR power and developed an individual adaptive training system based on EEG neurofeedback to alleviate anxiety in healthy people. The three parameters mentioned above were utilized to validate training effects in treating healthy people with anxiety during the SMR up-training.

Materials and methods

Subjects

A total of 33 right-handed adults (sixteen men and seventeen women, [mean ± standard deviation (SD) age, 22.21 ± 1.60 years]), without any history of neurological illness or psychiatric disorders, were recruited from Tianjin University in this study. A single-blind placebo-controlled design was applied in which subjects were randomly assigned to the neurofeedback groups (seven men and ten women, mean age = 23.07 years, SD = 1.77, range 20–26), who were treated using feedback from electrode C3 (Gruzelier 2014b) in a relative SMR power protocol or the control group (ten men and six women, mean age = 21.88 years, SD = 1.36, range 18–24), who were treated using feedback from pseudorandom numbers. All subjects were informed about the procedure of the experiment before EEG data recording and signed a consent form in advance; five of the subjects had had neurofeedback relaxation training over a year previously (two of the five subjects were in the neurofeedback group and three in the control group in this study). They all received a small financial remuneration after completing the experiment. There were no significant differences between groups in the Gender, Age, and Self-Rating Anxiety Scale (SAS). The experimental protocol was approved by the Ethics Committee of Tianjin Anding Hospital.

Neurofeedback system design

The system is mainly composed of three parts: an EEG acquisition module, a feature extraction module, and a feedback module. The neurofeedback system is shown in Fig. 1.

Fig. 1.

Visual neurofeedback experimental system based on brain-computer interface

Acquisition module

Each subject was seated comfortably in a chair about 65 cm away from a 23.5-inch LCD screen; EEG signals were acquired by a 64-channel SynAmps² system (Neuroscan, Australia) with standard Ag/AgCl electrodes placed on the scalp according to the international 10–20 system. The reference electrode was located at the right mastoid and the ground electrode was placed on the forehead. The impedance for all electrodes was kept below 10 kΩ. The sampling rate was 1000 Hz.

Feature extraction module

For the preprocessing, the EEG raw data were varied with reference to binaural averaging and downsampled at 500 Hz. To filter out baseline drift, power line interference, and high-frequency noise, an online band-pass filter between 0.5 and 50 Hz and a 50 Hz notch filter were enabled in the amplifier (Wang et al. 2019). In this study, we implemented an independent component analysis filter using the EEGLAB toolbox in MATLAB 2016b. The neurofeedback interface was written using MATLAB 2016b and Java Script. The neurofeedback training focused on increasing the relative SMR power referred to electrode C3 as a feedback electrode. The power spectrum estimation method used in this study was Welch algorithm spectral with a 3 s Hamming window, 50% overlapping, and was zero-padded to 512 points (Akbar et al. 2016; Feng et al. 2010). We used a Welch algorithm to decompose EEG data to the following frequency bands: delta (0.5–4 Hz), theta (4–8 Hz), alpha (8–12 Hz), SMR (12–15 Hz), beta (15–30 Hz), and gamma (30–45 Hz) (Mirifar et al. 2017). Then we calculated the power value of each frequency band. The overall process of Welch spectrum estimation is to segment the EEG data firstly, then calculate the power spectrum of each data segment, and finally calculate the average power spectrum of all data segments in which data segments are allowed to overlap, and perform windowing operation on each data segment. The data with a total length of N is divided into K segments (overlapped), and the length of each segment is L. The calculation formula is as follows:

$$P\omega =\frac{1}{k}\sum _{i=1}^k\left(\frac{1}{\mathit{LV}}{\left|\sum _{n=1}^l\omega \left(n\right)x_i\left(n\right)e^{-jwn}\right|}^2\right)$$ 1

where the power of the ω(n): $\mathrm{V}=\frac{1}{L}{\sum }_{n=1}^L{\left|\omega ,(,n,)\right|}^2$

Feedback module

During online training, EEG data were online filtered from blinking artifacts (through the independent component analysis filter) and visual feedback was then displayed every 2 s on the LCD screen in the form of a sunset video. Linear mapping was used to convert between the relative SMR power and the sunset video progress. Subjects were not given any instruction in how to control the feedback screen but were only told to stay mentally focused and physically relaxed. They were not given any mental strategy, nor were they aware of the EEG-trained parameter (Ioannides 2018). No additional information was displayed during training. Subjects were instructed to let the sunset video play for as long as possible. The relative SMR power value was calculated based on the following formula (Finnigan and Robertson 2011; Reichert et al. 2015; Wang et al. 2013):

$$P_{\text{r}}\text{(SMR)}=\frac{P_{\text{a}}\text{(SMR)}}{P_{\text{a}}\text{(total bandwidth)}}$$ 2

where P_r = relative power; P_a = absolute power; total bandwidth = 0.5–45 Hz.

Experimental design

Figure 2a illustrates the experimental paradigm procedure. The severity of anxiety was evaluated using the SAS at the beginning and the end of the experiment (about 10 days apart) (Olatunji et al. 2006). The neurofeedback group underwent six times neurofeedback training (a single training every other day) for about 2 weeks (Quaedflieg et al. 2016), all in the same period (Fig. 2b). Figure 2c illustrates a single neurofeedback training. The duration of a single training was 42 min, comprising an 8-min resting-state period, a 2-min baseline trial, six 4-min feedback trials, and a second 8-min resting-state period.

Fig. 2.

Experimental design. a The SAS was used at the beginning and the end of the experiment (which intervened about 2 weeks). b , c The neurofeedback group underwent a total of six times neurofeedback training (one every other day), which were composed of two resting-state periods (8 min each), a baseline, and six training trials (4 min each) namely a session. d Resting-state was recorded as a block of eight one-minute periods following an ‘OCCOCOOC’ sequence of eyes-open and eyes-closed states (where ‘O’ = eyes-open and ‘C’ = eyes-closed).

Neurofeedback procedure

Resting-state

The resting-state measurement in this study used an “OCCOCOOC” sequence (“O” = “open eyes”; “C” = “closed eyes”) for a total of 8 min since the EEG signals are more affected by blinking when the eyes are open and more stable when the eyes are closed and resting (Adolph and Margraf 2017). During the resting-state, the subjects wore headphones; they kept their eyes open when a prompt tone was played to indicate “Please open your eyes,” and closed their eyes when the prompt tone indicated “Please close your eyes.” The details are shown in Fig. 2d.

Baseline and trial

A 2-min baseline and six 4-min trials are under the online feedback module. A 2-min baseline was recorded, in which the subjects saw the screen and were instructed to relax without trying to control the feedback screen voluntarily. The baseline data recorded was used to calculate an individual threshold. There are 60 relative SMR values in the 2 min baseline as 2 s is a sample. The minimum value of relative SMR power at baseline was the individual threshold of that day. The individual threshold did not adjust at the beginning of each trial. The sunset video started playing when the feedback value exceeded the individual threshold. Otherwise, the screen displayed the initial image of the sunset video. During each trial, when the feedback value exceeded 80% of the previous one, the current threshold was increased by one. By contrast, when the feedback value was less than 20% of the former, the current threshold was reduced by one. The sunset video was presented on an LCD screen: the progress of the video linearly corresponds to the relative SMR power. Subjects were instructed to try their best to let the sun sink as low as possible.

Offline EEG processing

The EEG data were cleaned of artifacts using a three-step procedure. Initially, the reference was varied and downsampling was conducted in the same way as in online EEG processing. A Butterworth filter was used to filter the EEG data between 0.5 and 50 Hz. Then the EEG data were carefully inspected for the presence of artifacts, such as eye blinks, eye movements, and body movements. We applied an independent component analysis filter in MATLAB 2016b to remove the eye blinking component. Finally, we computed the power values for each epoch in the bands 4 to 30 Hz, which were commonly not affected by ocular and muscular artifacts (Delorme et al. 2007).

Self-rating anxiety scale

The SAS is a self-reported questionnaire that contains 20 items to assess the intensity of anxiety. Each question gives a choice of four answers (each scoring 1 to 4 points). The SAS score is used to determine whether a subject has minimal (20–39 points), mild (40–49 points), moderate (50–59 points), or severe (60–80 points) anxiety (Tao and Gao 1994). The test–retest-reliability in the SAS score (r = 0.760, p < 0.01) has also been examined.

Statistical analysis

Tiredness is a largely unrecognized feature of neurofeedback learning. Gruzelier et al. (2014) mentioned that subjects reported tiredness after a 15-min SMR protocol session (Gruzelier 2014b). In this study, as a result of the 24-min SMR training, the second resting-state period in a single training did not reflect the training effect well; therefore, we only analyzed the first resting-state period. The first resting-state period of the six times neurofeedback training was defined as resting 1 to resting 6. Baseline, trial, and session are all defined in the same way. For each subject, the SMR power at the feedback electrode (C3) was calculated for the first resting-state, baseline, trial, session of each training. Moreover, the power values were processed by the min–max normalization, and outliers at any electrode were removed (White and Thomas 2005). All data were analyzed and eliminated by the Kolmogorov–Smirnov test and triple standard difference method using SPSS (Version 20.0. Armonk, NY: IBM Corp.2011) and MATLAB 2016b.

Concerning this study on healthy people with anxiety, data analysis excluded those pre-SAS values above 40, including one in the neurofeedback group and one in the control group. We also excluded subjects who may have an impact on the results due to events or physical conditions during training, including three in the neurofeedback group and two in the control group. Thus, the final sample for the EEG analysis and SAS scale consisted of 26 subjects (i.e. the neurofeedback group: 13, the control group: 13).

1. Trainability is reflected by the trend of the SMR power over training days. SMR power at the C3 electrode in resting-state was tested. To assess trainability, repeated measures analysis of variance (ANOVA) was performed, with two within-subject factors of Group (neurofeedback group and control group) and Time (resting 1 to resting 6), which used Greenhouse–Geisser’s degrees of freedom correction for violations of the sphericity assumption (as indicated by significant Mauchly s test of sphericity) and Least—Significant Difference (LSD)'s corrected probability values (p-values) for post-hoc comparisons. If trainability was successful, the frequency specificity, as well as the effect on the behavioral level of the neurofeedback training, were assessed.

2. In the neurofeedback group, we determined whether only the SMR frequency band was significantly increased. To assess frequency specificity, resting 1 and resting 6 were collapsed within four frequency bands (theta (4–8 Hz), alpha (8–12 Hz), SMR (12–15 Hz), beta (15–30 Hz)) across subjects and compared using a paired t-test. Statistical significance was assumed at the 0.05 alpha level (two-tailed).

3. To judge whether the intervention training affects the emotional experience, the raw scores of the SAS scale were utilized before and after training in two groups, respectively, using a paired t-test. Then, we assessed brain-behavior relationships, and the Spearman correlation was calculated between the change in SMR power and the change in symptoms of anxiety (SAS score).

Results

Demographics for both groups are given in Table 1. There were no significant differences between groups in sex and age (p > 0.05). Repeated measures ANOVA did not reveal significant main effects of Group (neurofeedback group and control group) and Time (baseline 1 to baseline 6) on SMR power or relative SMR power at C3 (SMR power (Group: F(1, 24) = 1.794, p > 0.05; Time: F(5, 20) = 0.446, p > 0.05); Relative SMR power (Group: F(1, 24) = 0.727, p > 0.05; Time: F(5, 20) = 0.437, p > 0.05). Furthermore, there was no significant interaction between Time and Week (SMR power: F(5, 20) = 0.946, p > 0.05; Relative SMR power: F(5, 20) = 1.003, p > 0.05). Note that preliminary analysis (independent samples t-tests and repeated measures ANOVA) revealed no differences in behavioral parameters and electrophysiological parameters between the neurofeedback group and the control group.

Table 1.

Demographic variables of the subjects in this study

	Neurofeedback group n = 13	Control group n = 13	Neurofeedback vs. control, p
Sex (male:female)	4:9	7:6	0.234
Age, years (mean ± SD)	22.38 ± 1.71	22.08 ± 2.15	0.917
SAS scale (mean ± SD)	28.85 ± 4.22	29.23 ± 5.18	0.837

Trainability

As depicted in Fig. 5, the averaged SMR power across all subjects during six times neurofeedback training at the C3 electrode in the resting-state was calculated, demonstrating that the SMR power increased in the neurofeedback group but that there was no upward trend in the control group. Repeated measures ANOVA revealed a significant interaction between Group (neurofeedback group and control group) and Time (resting 1 to resting 6) on SMR power at C3. Furthermore, a significant simple effect of Time by Group in the neurofeedback group was found. Also, the results revealed a significant simple effect of Group by Time. LSD post-hoc comparisons revealed the SMR power at C3 in resting 6 was higher than resting 1, resting 2, and resting 4 of the neurofeedback group and the control group is higher than the neurofeedback group in resting 3. (Table 2 and Fig. 3).

Fig. 5.

EEG changes between resting 1 and resting 6 at the C3 electrode in the neurofeedback group: *p  <  0.05. Error bars indicate SD

Table 2.

Learning curves for SMR power at C3 during the resting-state

	F
	Group (p)	Time (p)	Group × Time (p)	LSD's post-hoc comparisons
C3	3.833 (0.062)	1.620 (0.160)	2.538* (0.032)	Group: Con > NF in resting 3
				Time: resting 6 > resting 1, resting 2 and resting 4 in NF

Bold number indicates repeated measures ANOVA revealed a significant interaction between Group (neurofeedback group and control group) and Time (resting 1 to resting 6) on SMR power at C3

Bold words indicate a significant simple effect of Time by Group in the neurofeedback group was found. Also, the results revealed a significant simple effect of Group by Time. LSD post- hoc comparisons revealed the SMR power at C3 in resting 6 was higher than resting 1, resting2, and resting 4 of the neurofeedback group and the control group is higher than the neurofeedback group in resting 3

Con: Control Group; NF: Neurofeedback Group; *p < 0.05

Fig. 3.

The learning curve for averaged SMR power at feedback electrode (C3) in resting-state across all subjects during six times training in the neurofeedback group and control group. The x-axis shows the resting-state range, whereas the y-axis shows averaged SMR power: *p < 0.05. All data have been min–max normalized individually. Error bars indicate SD. The black dotted line indicates zero

Furthermore, it is necessary to report on changes in SMR power within- and between- neurofeedback training. Figure 5b illustrates the time course of SMR power over the training trials for both groups. The regression analysis of absolute SMR power (predictor variable = trial; dependent variable = SMR power) revealed a significant positive slope across trials only in the neurofeedback group. In terms of within-training, repeated measures ANOVA revealed no significant main effects of Group (neurofeedback group and control group) and Time (trial 1 to trial 6) on SMR power at C3, but there was a significant interaction between Group and Time. In addition, a significant simple effect of Time by Group in the neurofeedback group was found. Besides, the results revealed a significant simple effect of Group by Time. LSD post-hoc comparisons revealed higher SMR power at C3 in trial 6 than trial 1 and trial 4, trial 3 than trial 1 of the neurofeedback group, and the neurofeedback group is higher than the control group in trial 6. (Table 3 and Fig. 4). In terms of between-training, repeated measures ANOVA revealed no significant main effects of Group (neurofeedback group and control group) and Time (session 1 to session 6) on SMR power at C3(Group: F(1, 24) = 0.545, p > 0.05; Time: (F(5, 20) = 0.916, p > 0.05)), and there was no significant interaction between Group and Time (F(5, 20) = 0.959, p > 0.05).

Table 3.

Learning curves for SMR power at C3 within-session

	F
	Group (p)	Time (p)	Group × Time (p)	LSD's post-hoc comparisons
C3	0.762(0.391)	1.240(0.328)	5.202 **(0.003)	Group: NF > Con in Trial 6
				Time: Trial 6 > Trial 1 and Trial 4 in NF Trial 3 > Trial 1 in NF

Bold number indicates there was a significant interaction between Group and Time

Bold words indicate a significant simple effect of Time by Group in the neurofeedback group was found. LSD post- hoc comparisons revealed higher SMR power at C3 in trial 6 than trial 1 and trial 4, trial 3 than trial 1 of the neurofeedback group. Besides, the results revealed a significant simple effect of Group by Time. LSD post- hoc comparisons revealed higher SMR power at C3 in the neurofeedback group than the control group in trial 6

Con Control Group, NF Neurofeedback Group

**p < 0.01

Fig. 4.

Time course of SMR power over the neurofeedback training trials, averaged over all 6 neurofeedback training sessions, presented separately for the neurofeedback and control groups. The x-axis shows the trial range, whereas the y-axis shows averaged SMR power: *p  <  0.05; **p  <  0.01. The black dashed line represents the regression line. All data have been min–max normalized individually

Frequency specificity

As the trainability analyses revealed that the trained frequency changed significantly in the neurofeedback group, frequency specificity was subsequently assessed in the neurofeedback group. A paired t-test revealed that there was a significant increase in SMR power between resting 1 and resting 6 at the C3 electrode (SMR: t(12) = -2.701, p < 0.05,), while other frequency bands (theta: t(12) = -2.062, p > 0.05, alpha: (t(12) = -0.069, p > 0.05, beta: (t(12) = -0.986, p > 0.05) were not significantly affected. These results indicate that using the relative SMR power protocol, the neurofeedback training increased SMR power independent of the other frequency bands in the neurofeedback group (Fig. 5).

Brain-behavior relationships

On average, subjects in the neurofeedback group decreased their SAS score from 28.85 ± 4.22 to 26.69 ± 4.27 after training (t(12) = 2.545, p < 0.05). For the control group, the SAS scale after six times training was decreased slightly but not significantly (t(12) = 2.030, p > 0.05) from 29.23 ± 5.18 to 27.00 ± 6.23. Furthermore, the correlations between changes in emotional symptoms and SMR power at electrode C3 in the first and last training of resting-state were examined. A significant negative correlation was found between the decrease in SAS score and the increase in SMR power (r =  0.398, p < 0.05). That is, the greater the SMR power, the greater the improvement in symptoms of anxiety.

Discussion

The objective of this work was to propose the relative SMR power protocol as the feedback value used to improve absolute SMR power and develope an individual adaptive training system based on EEG neurofeedback to alleviate anxiety in healthy people. Trainability, frequency specificity, and brain-behavior relationships were used to validate training effects in treating healthy people with anxiety during the SMR up-training. Anxiety is associated with inattention in terms of clinical manifestation. Given the rationale beyond this protocol, the relation between SMR power and attention performance has been found in a previous study (Lubar and Lubar 1984). To decrease data variability and better reflect cortical activity, we hereby proposed the relative SMR power protocol as the feedback value and explored the application of a neurofeedback protocol based on SMR up-regulation in healthy people with anxiety. The results showed that the subjects who had NFT relieved their anxiety and increased their SMR power compared with subjects who had sham training. This suggests that neurofeedback supports the feasibility and efficacy of the NFT protocol based on relative SMR power for healthy people with anxiety. Thus, this neurofeedback protocol has the potential to alleviate anxiety in a healthy population.

Regarding trainability, we first assessed the effectiveness of the neurofeedback protocol for all subjects by evaluating the change in SMR power. The analysis of learning effects during training and within-session illustrated a gradual enhancement in SMR power. In addition, we have failed to report significant changes in SMR power between-session which is consistent with the results of Kober et al. (2015), Vernon et al. (2003), and Vernon, (2005) (Kober et al. 2015; Vernon 2005; Vernon et al. 2003). For the neurofeedback group, a linear increase of SMR power during the first resting-state overtraining was visible, suggesting the transfer of the previous learning experience to the next training. The mechanism could be that subjects accessed their perception to regulate the EEG signals, mastered techniques to control EEG signals, and reinforced the techniques by repeated practice of neurofeedback tasks (Lacroix and Gowen 2010; Schwartz and Shapiro 1976; Sitaram et al. 2016). This result is significant because it proves that the training of each stage is based on the training times and experience of the previous days. As with any type of experiment, the amount of training is crucial to determine its effectiveness. Hammond (2011) believes that at least 5–10 times of training are needed to prove the effectiveness of neurofeedback (Hammond 2011). In this study, we performed only six times of training, which limited the performance of the results. If more training times were added, the results of trainability should have been better. In the recent neurofeedback literature, it is still highly discussed whether neurofeedback training should lead to tonic changes in the background EEG (as assessed during resting measurements) or to phasic changes (Gruzelier 2014c). Only the analysis of changes in SMR power within- and between- sessions can reveal the second one. The neurofeedback group receiving real feedback showed an increasing trend in absolute SMR power at the C3 electrode within-session. Subjects showed receiving sham feedback no changes in SMR power over time within-session. But beyond that, the change in SMR power between-session was not found. It shows that the subjects can also improve the SMR power during short-term training. But each session, that is, the state of each day is not the same, so there is no trend of SMR growth between sessions during the training. These results reflect successful NFT based on the relative SMR feedback value in the neurofeedback group (Gruzelier et al. 2006; Kerstin et al. 2008; Schabus et al. 2014; Vernon et al. 2003). Accordingly, our subjects were able to learn how to adapt and generalize mental alertness, physical relaxation, which corresponded to increasing SMR activity (Marzbani et al. 2016; Xiang et al. 2018).

Regarding frequency specificity, only SMR band activity significantly increased in the neurofeedback group after training; other frequency bands, namely theta, alpha, and beta were unaffected. In other words, the neurofeedback group significantly increased the target frequency band, which is also consistent with previous research (Chen and Lin 2020; Quaedflieg et al. 2016; Zoefel et al. 2011). Similarly, we also want to explore the changes in the control group. A paired t-test revealed that there was a significant increase in theta power between resting 1 and resting 6 at the C3 electrode (theta: t(12) =  2.184, p < 0.05), while other frequency bands (alpha: t(12) = 0.194, p > 0.05; SMR: t(12) = 0.057, p > 0.05; beta: t(12) = 0.224, p > 0.05) were not significantly affected possibly due to theta indexing low arousal, tiredness, and inattention (Gruzelier 2014b). Besides, the feedback value in this study was relative power which included theta, alpha, SMR, and beta bands, and the subjects in the control group felt tired or scatterbrained because they were unable to control the feedback interface easily. These results indicate that using the relative SMR power protocol, the neurofeedback training increased SMR power independent of the other frequency bands in the neurofeedback group (Fig. 6).

Fig. 6.

EEG changes between resting 1 and resting 6 at the C3 electrode in the control group: *p  <  0.05. Error bars indicate SD

Regarding Brain-behavior relationships, the increase in SMR power correlated only with the decrease in SAS score. Hence, the increase in SMR activity was correlated only with the intensity of anxiety according to the subjects in this study. The SMR training protocol is known to enhance attention. In this study, attention has been improved probably from the following two reasons. Subjects have learned the mental state promoted by SMR power up-training, namely a peaceful, relaxed state, and remaining focused leads to improve the body's inhibition ability and reduces the interference of somatosensory information, possibly a result of subjects shifting their attention from negative thoughts to the neurofeedback tasks (Hammond 2011; Pimenta et al. 2018; Sitaram et al. 2016). Another reason for increased attention may be due to improved sleep quality. Regarding the SAS, we explored the items about sleep according to the following phrases: 'I fall asleep easily and get a good night's rest ', 'I have nightmares'. On average, subjects in the neurofeedback group decreased their score from 3.46 ± 0.78 to 3.00 ± 0.91 after training (t(12) = 2.521, p < 0.05). For the control group, the scale after six times training was decreased slightly but not significantly (t(12) = 0.485, p > 0.05) from 3.08 ± 1.39 to 2.92 ± 1.12. As mentioned above, training aspects of attention can relieve anxiety (e.g., yoga, meditation). This training helps to learn focus attention (Kiken et al. 2015; Simon et al. 2020). Therefore, training with the SMR protocol improved the body's inhibition ability and sleep quality, thereby improving attention (Enriquez-Geppert et al. 2019; Reichert et al. 2016), which may be the reason for the improved intensity of anxiety.

There are some limitations to this study. The main limitation is that the sample size should be justified with a power analysis based on the smallest effect size of interest or another method (Online Resource 1) (Ros et al. 2020). A further limitation is that we did not include a cognitive performance test that assessed attention based only on the SMR power. Without such a cognitive performance test, we cannot completely exclude the possibility that subjects gradually became familiar with the experiment process during the training or daily life. Third, long-term training could have a better impact than short-lived state changes (Engelbregt et al. 2016). Therefore, future protocols might require more training or follow-up studies to ensure more robust effects of NFT based on EEG.

In future research, we aim to find an appropriate method to exclude the responder to verify the role of NFT. This would be beneficial for future trainer selection and treatment outcome prediction. Additionally, combining other reward protocols with concurrent relative SMR power up-regulation might generate more positive outcomes. Neurofeedback based on relative SMR power could be a time-effective treatment for anxiety and an interesting new option to consider in anxiety therapy, especially for clinical applications. However, whether absolute or relative SMR power is better as a neurofeedback value will be discussed in the future.

Conclusion

Taken together, the relative SMR power protocol we proposed has been reported as the feedback value to increase SMR power rather than relative SMR power within healthy subjects who learned self-regulation based on NFT. Moreover, three parameters—trainability, frequency specificity, and brain-behavior relationships—were utilized to validate training effects in treating healthy people with anxiety during the SMR up-training. Hence, our results imply that NFT using a feedback protocol based on relative SMR power at the C3 electrode could significantly decrease the SAS score and increase SMR power; thus, anxiety could be relieved for healthy people. This study supports the potential of neurofeedback as a promising novel experimental therapy for neurological and psychiatric conditions. This preliminary study supports the feasibility and efficacy of a relative SMR power protocol for healthy people with anxiety.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

NFT

Neurofeedback training

SMR

Sensorimotor rhythm

SAS

Self-rating anxiety scale

Author contributions

SL and XH contributed equally to the study conception, literature search, and writing. All authors contributed to manuscript revision, read and approved the submitted version.

Funding

This research was funded in part by National Natural Science Foundation of China, Grant Nos. 81925020, 81630051.

Declarations

Conflict of interest

The authors declare that they have no conflict of interest.

Ethical approval

This study was approved by the Ethics Committee of Tianjin Anding Hospital. (Number: NCT04562324 (Online Resource 2)).