Combining neurofeedback with source estimation: Evaluation of an sLORETA neurofeedback protocol for chronic tinnitus treatment

2.1Participants

Fifty-three chronic tinnitus patients were recruited at the Department of Otorhinolaryngology (University Hospital Zurich). Participants had to be diagnosed with chronic subjective tinnitus (>0.5 years), be between 18 and 75 years old, have adequate knowledge of the German language, wear no cochlear implants, suffer from no psychiatric or neurological disorder, have no acute suicidal tendency, show no signs of drug or alcohol addiction, and have no current prescriptions for tranquilizers, neuroleptics, or anti-epileptics. Three participants decided to leave the trial before finishing the full study procedure (described in section 2.2) and were thus excluded from data analysis. Furthermore, two patients who completed the full study procedure had to be excluded because their depression scores at all four time points suggested clinically relevant depressive symptoms (i.e., BDI sum-scores of more than 18 points). Unfortunately, one patient died due to an unrelated health issue before completion of the last follow-up measurement. The obtained data for this patient was however included in the analysis as they were successfully gathered for the neurofeedback training, as well as the post-testing and the first follow-up measurement.

Table 1 shows the demographic and clinical details of the 48 participants included in the final analysis. Participants of the final sample were between 24 and 75 years old with a mean age of 46.83 (SD = 12.8). The sample consisted of 38 males and 10 females.

Thirty-five participants described their tinnitus percept as tonal, three as noise-like, and ten indicated an undefinable type. Most participants (n = 39) perceived tinnitus in both ears, however 16 subjects of this group indicated a left- while 11 specified a right-sided tendency. Of the remaining subjects, four indicated a clear left lateralization, three a right lateralized tinnitus, and two experienced it diffusely inside their heads.

Mean tinnitus loudness was rated as 54.04 (SD = 25.17), while mean distress measured with the THI consisted of 32.58 points (SD = 17.01), and of 26.33 points (SD = 14.06) points measured with the TQ. These values suggested “mild tinnitus” according to the THI and “slight tinnitus” according to the TQ for the overall group on average.

Participants were randomly assigned to one of two study groups that followed identical study procedures (described in detail in section 2.2) with the only difference being the tomographic (TONF) or non-tomographic (NTNF) neurofeedback application (details in section 2.4). The groups were matched according to age, gender, and hearing loss, and were of equal size (n = 24). Participants were not informed of their group assignation, and this project is thus considered a single-blind randomized controlled trial. The study was approved by the appropriate Ethics Committee (Kantonale Ethikkommission Project KEK-ZH-Nr. 2014-0594) and was registered online at ClinicalTrials.gov (NCT02383147) and kofam.ch (SNCTP000001313).

2.2Procedure

Prior to the start of the neurofeedback training period, participants gave their informed consent, underwent audiometric screening, and participated in the baseline measurements (t1) of the study. This screening assessment consisted of resting-state EEG measurements (details in section 2.3) and the completion of a set of questionnaires covering demographics, tinnitus-related symptoms, as well as other psychological and health-related questions. The forms for this subjective measurement were chosen according to the guidelines of the Tinnitus Research Initiative (TRI) (Landgrebe et al., 2012; Langguth et al., 2007). In particular, an adjusted version of the Tinnitus Sample Case History Questionnaire (TSCHQ) was used to obtain information about demographics, tinnitus properties, prior treatment attempts, and other tinnitus-related issues. Participants evaluated their tinnitus distress using the German version of the Tinnitus Handicap Inventory (THI) (German version by Kleinjung, Fischer, et al., 2007) and the Tinnitus Questionnaire (TQ) (German version by Goebel & Hiller, 1994). For these two forms, sum-scores range between 0–100 in the former and 0–84 in the latter case. Additionally, participants completed German versions of Beck’s Depression Inventory (BDI) (Hautzinger, Bailer, Worall, & Keller, 1995), Beck’s Anxiety Inventory (BAI) (Prinz & Petermann, 2015), the short form of the WHO Quality of Life scale (WHOQOL-BREF) (Angermeyer, Kilian, & Matschinger, 2000), Symptom Check List (SCL-K-9) (Klaghofer & Brähler, 2001), and Short Form Health Questionnaire (SF-36) (Bullinger, Kirchberger, & Ware, 1995). The main behavioral outcome measures of this study were defined as self-reported tinnitus loudness (rated from 1 “very low” to 100 “very high”) and sum-scores of THI and TQ.

After the screening, participants were randomly assigned to one of two study groups (TONF vs. NTNF) and participated in 15 weekly neurofeedback training sessions (details in section 2.4). Post-measurements (t2) were performed immediately after the completion of the training period by repeating resting-state EEG measurements as well as questionnaire surveys. Two follow-up measurements were conducted 3 and 6 months after the training period. While questionnaires were completed at both follow-up time points, resting-state EEG was only recorded at the first (t3). Participants completed the final survey for the last follow-up measurement (t4) at home after all preceding measurements had been conducted at the Department of Otorhinolaryngology and the University of Zurich.

2.3EEG recording

For recording of resting-state EEG at t1, t2, and t3, we used a BrainAmp DC amplifier system in combination with active channel actiCap electrode caps (Brain Products, Munich, Germany). The 64 silver/silver chloride electrodes were arranged according to the 5/10 electrode position system (Oostenveld & Praamstra, 2001). FCz electrode served as reference for the recording while the ground electrode was placed at AFz.

Data was sampled with a sampling rate of 1000 Hz. Furthermore, direct current (DC) mode with a high-cutoff filter of 1000 Hz and a slope of 12 dB/octave was used for recording. Electrodes were prepared with conductive paste prior to recording in order to keep impedance below 10 kΩ. The measurement was performed in a sound-proof and electromagnetically shielded room, and patients were given minimal instructions: to sit upright and to avoid excessive movements in order to minimize potential artifacts.

Resting-state EEG was then recorded over a time span of 8 minutes during which subjects were asked to open (EO) and close (EC) their eyes in regular intervals. For playback of these instructions, Presentation software (Neurobehavioral Systems, Inc., 2010) was used, and a fixation cross was presented during eyes-open segments. After the first measurement, patients were given some time to relax and were subsequently informed about the second part of EEG recordings. This time, participants were explicitly instructed to allow their tinnitus to come naturally during the 8 minutes of measurement and to try not to suppress it. This clear instruction given before the second resting-state measurement (as opposed to the minimal before the first) has been shown to make the measurement situation more comparable with the neurofeedback situation because a patient’s tinnitus is usually more salient during neurofeedback treatment (Güntensperger et al., 2019). Therefore the resting-state activity of eyes-open segments of this second recording was chosen as main electrophysiological outcome measure.

2.4Neurofeedback training

Thirty-one silver/silver chloride electrodes according to the 10/20 system and a NeuroAmp amplifier (BEE Medic GmbH, Singen, Germany) were used for acquiring neurofeedback EEG. The signal was referenced to electrodes placed on the earlobes of subjects, and ground electrode was placed at AFz. The sampling rate of the recording was set at 500 Hz, and impedance was kept below 20 kΩ. The EEG signal was processed in real-time using the software Cygnet 2.0.3.34 (BEE Medic GmbH, Singen, Germany). While only activity measured at four fronto-central electrodes (FC1, FC2, F3, and F4) was used for feedback generation of the NTNF group, the signal of all 31 electrodes was considered for the TONF group. In the latter case, the implemented sLORETA algorithm estimated the signal at four sources over the primary auditory cortex (55/–25/10 and 55/–30/10 for right Heschl’s gyri, –55/–25/10 and –55/–30/10 for left Heschl’s gyri). Only activity over these sources was set as relevant for the neurofeedback of the TONF group. It is important to note that, even though only four electrodes were feedback-relevant for participants of the NTNF group, all 31 electrodes were equally placed and prepared for these participants in order to disguise group affiliation. Visual feedback was implemented in the computer simulation Inner Tube (Somatic Vision, Encinitas, CA, USA). Patients observed a space ship, the speed, navigation accuracy and visibility of which changed according to their brain activity.

While an increase in alpha power was rewarded with acceleration of the ship, an increase in delta deteriorated autopilot accuracy and visibility of the simulation. The rewarded alpha frequency band was set in the range of±2 Hz around an individual alpha peak determined for each participant, and the inhibited delta band was set in the range of 3–4 Hz. During the feedback, the Cygnet software allowed for automatic filtering and excluded movement artifacts and system voltage (45–55 Hz) from feedback. Each session lasted for 15 minutes and was repeated once a week.

2.5EEG data analysis

Brain Vision Analyzer 2 (Brain Products, Munich, Germany) was used for preprocessing the EEG data. Band-pass filters with Butterworth zero-phase filters between 0.1 Hz and 80 Hz with slopes of 24 dB/octave at the low, and 48 dB/octave at the high cutoffs were first applied to the data. A band-rejection filter with a central frequency of 50 Hz, a bandwidth of 1 Hz, and a slope of 24 dB/octave was used in order to eliminate possible line noise. Next, very noisy or dead channels were identified by visual inspection and temporarily excluded from the data. An independent component analysis (ICA) with a restricted Infomax algorithm was performed and, in doing so, the EEG signal decomposed into its independent components. The resulting components were inspected to identify if they were indicative of eye-blinks, pulse, or other regular artifacts. Obvious cases were subsequently removed from the data with the inverse ICA procedure. After this step, previously excluded channels and electrodes with high amounts of remaining noise were replaced by using spline-type topographical interpolations. As a last step, data were carefully visually inspected and any remaining irregular artifacts (e.g., muscle movements) removed. Before segmenting data into eyes-closed and eyes-open conditions, an average reference over all channels was applied. The implicit reference of data recording (FCz) was re-included into the data and used for subsequent analysis.

In the next step, data were imported into MATLAB Statistics Toolbox 2017a (The MathWorks Inc., Natick, Massachusetts, United States) and EEGLAB 14.1.1b (Delorme & Makeig, 2004) for surface-based analysis. A hamming window with 2 s window length and 1s overlap was first applied on the data of eyes-open segments. Next, Fast Fourier Transforms (FFT) for each 2 s-segment were computed, logarithmized and averaged over all segments for each patient resulting in power values in decibels (dB) with a frequency resolution of 0.5 Hz. Subsequently, power values in the rewarded (individual) alpha range were divided by those in the inhibited (3–4 Hz) delta range. The resulting alpha/delta ratio was then averaged over the four electrodes (FC1, FC2, F3, F4) used for training of the NTNF group.

Analysis on source level was performed using the LORETA toolbox (version 20150810). Technical details of implemented power analysis and source estimation are provided in Pascual-Marqui (2002). The lead field behind the sLORETA algorithm is described in Fuchs, Kastner, Wagner, Hawes, and Ebersole (2002) and the integrated electrode position system in Jurcak, Tsuzuki, and Dan (2007). Standard electrode positions were registered with the integrated tool and the transformation matrix was regularized according to an estimated signal to noise ratio of 100. The analysis of log-transformed sLORETA data in the frequency domain resulted in mean current density values (mA/mm²) that were exported and averaged over the four voxels used for neurofeedback training of the TONF group (see section 2.4).

2.6Statistics

Data analysis was performed using the statistical environment R (R Core Team, 2017) including packages “ggplot2” (Wickham, 2009), “ggsignif” (Ahlmann-Eltze, 2017), “Hmisc” (Harrell Jr, 2017), “jtools” (Long, 2017), “multcomp” (Hothorn, Bretz, & Westfall, 2008), “nlme” (Pinheiro, Bates, DebRoy, Sarkar, & R Core Team, 2017), and “xtable” (Dahl, 2016). Time effects for behavioral (THI and TQ sum-scores, tinnitus loudness) and EEG-related data were estimated by means of repeated-measures mixed model analyses of variance (ANOVA). Using the “nlme” package in R, the model was specified as lme(X∼time, random=∼1|subject, data, method=“ML”, ”na.action=na.exclude) and compared to a baseline model using only the intercept as predictor lme(X∼1, random=∼1|subject,data,method=“ML”,na.action=na.exclude).

Additionally, a model with group-term (group + time) was calculated to test for group effects within time points: lme(X∼time + group, random = ∼1|subject, data, method=“ML”, na.action=na.exclude). An additional model with group*time interaction was fitted on the data in order to reveal potential differences between the TONF and NTNF group between time points. This model was specified as: lme(X ∼time*group, random =∼1|subject, data, method=“ML”, na.action=na.exclude). In order to evaluate training success and stability of potential changes, a priori defined contrasts comparing t1 with all other time points (t2, t3 and t4 for behavioral, t2 and t3 for EEG data) were analyzed for all these models using Bonferroni corrected one-tailed t-tests. Furthermore, effect sizes r for these analyses were calculated. According to Field, Miles, and Field (2012), we converted them directly from respective t-values and labeled r = 0.1 a small, r = 0.3 a medium, and r = 0.5 a large effect (Cohen, 1988). Finally, in order to reveal any other potential differences between time points, post-hoc Tukey tests were performed comparing each of the four time points with each other. The alpha threshold was set at p = .05 for all statistical tests.

3.1Behavioral results

Results concerning changes in tinnitus-related symptoms of both neurofeedback groups (TONF: group with tomographic neurofeedback, NTNF: group with surface-based neurofeedback) are summarized in Fig. 1 and in Table 2. As described in section 2.6, three repeated-measures ANOVA mixed models with the factor time were calculated and compared to the baseline model (model without factor time): (a) a model without group-term (data is not dependent on neurofeedback group); (b) a model with group-term but no interaction (data within but not between time points is dependent on neurofeedback group); and (c) a model with group-term and group*time interaction (data within and between time points is dependent on neurofeedback group).

Fig. 1

Plots showing tinnitus-related symptoms for the two neurofeedback groups (dark: tomographic neurofeedback, gray: surface-based neurofeedback) before (t1), 1 week after (t2), 3 months after (t3), and 6 months after (t4) training. Tinnitus Handicap Inventory (THI) and Tinnitus Questionnaire (TQ) were analyzed as measures for tinnitus-related distress, and complemented with subjective tinnitus loudness (0–100). Error bars represent±1 standard error for within-subjects designs according to Morey (2008).

For the tinnitus questionnaire THI and for tinnitus loudness, the model without group-term showed the best fit on the data. This suggests a significant effect of the factor time on tinnitus distress and tinnitus loudness but no effect of group affiliation.

In the following paragraphs, results of a priori defined contrasts of these significant models only will thus be reported. For TQ, none of the models showed an improved fit compared to the baseline model suggesting no significant time effects in this case.

3.2EEG results

Data of resting-state EEG measurements for the two neurofeedback groups are summarized in Fig. 2 and Table 2. Repeated-measures mixed model ANOVA were performed for the trained alpha/delta ratio on both surface and source level data.

Fig. 2

Plots showing alpha-delta ratio for the two neurofeedback groups (dark: tomo- graphic neurofeedback, gray: surface-based neurofeedback) before (t1), 1 week after (t2), and 3 months after (t3) training. EEG data was analyzed on source level (over the four voxels used for neurofeedback of the ToNF group) and on surface level (over the four electrodes used for neurofeedback of the NTNF group). Error bars represent±1 standard error for within-subjects designs according to Morey (2008).

Results of the analyses indicate the best fit on the data for the model without group-term, suggesting significant effects of the factor time but no effect of group affiliation. In the following paragraphs, results of a priori defined contrasts will thus be reported for the model without group-term.

4.1Tinnitus-related distress

Results of the trial suggest that tinnitus patients of both groups benefited from the neurofeedback intervention (see Table 3 and Fig. 1). Tinnitus-related distress measured with the THI decreased over the course of the training and remained stable on a lower level up to 3 months after training completion (see section 3.1.1). Even though 6 months after the training no significant difference was found compared to baseline measures, Tukey post-hoc tests suggested no statistically meaningful increase in the follow-up period. Furthermore, tinnitus-related distress measured with TQ was on a lower level after the training compared to baseline. In this case, neither the results of the main ANOVA model nor contrast analysis of the four time points revealed significant differences, but data did suggest a statistical trend (see section 3.1.2).

On average, the study group was able to reduce THI scores by –5.29 and TQ scores by –2.48 points. But are these differences, statistically significant or not, also clinically relevant? As simple as this question may sound, there is still no clear consensus in the tinnitus literature about which size of difference should be termed clinically meaningful for patients. Regarding THI, recommended changes range from 6 points (Zeman et al., 2011) up to 20 points (Newman, Sandridge, & Jacobson, 1998). For TQ, the work by Kleinjung, Steffens, et al. (2007) indicated a minimal difference of 5 points to be clinically relevant but Hall, Mehta, and Argstatter (2018) recently suggested a higher cutoff of 12 points. Whichever of these we adopt, it seems that tinnitus-related distress of patients in our study on average did not change in a clinically meaningful way. It should be mentioned, however, that neither of these questionnaires were initially developed to account for measuring longitudinal treatment-related changes in distress. While the THI more generally measures impact of tinnitus on everyday function, TQ was designed to assess various psychological aspects of tinnitus (e.g., including impact on auditory perception), and both thus only partly cover tinnitus-related distress.

Categorizing the questions of these two questionnaires, Kennedy, Wilson, and Stephens (2004) showed that only 68% of questions concern psychological or emotional tinnitus distress. In the case of TQ, only 47% of questions serve this purpose which might explain the higher amount of inter-individual variance (see Fig. 1) and, thus, the non-significant results. The Tinnitus Functional Index (TFI), which has been specifically developed as a treatment-responsive measure (Meikle et al., 2012) and recently validated (Peter et al., 2017), might have been more suitable to cover treatment-related changes in this context.

Figure 1 illustrates the time course of tinnitus symptoms for the two neurofeedback groups. Even though results of the two groups seem to differ within and across time, repeated-measures mixed models considering group affiliation of the patients as a factor in the model did not show an improved fit on the data. This suggests that the neurofeedback treatment had roughly the same effect on changes in tinnitus-related distress regardless of whether the feedback was applied on the basis of only four active electrodes (NTNF) or was acquired with 31 electrodes and source estimation (TONF). The question that arises in this context is why the neurofeedback protocol designed to evoke spatially specific effects over auditory brain areas—that is, regions were the maladaptive processes leading to tinnitus have been proposed to originate from (De Ridder, Vanneste, Langguth, & Llinás, 2015; Llinás et al., 1999; Weisz, Dohrmann, & Elbert, 2007)—does not show different effects when compared to the unspecific NTNF protocol. Possible reasons for the absence of this between-group difference might be found in the high inter-individual variability of tinnitus manifestation. Recent research in the tinnitus field suggests that tinnitus should not be considered a homogeneous phenomenon that sounds and feels the same for all sufferers (e.g., Landgrebe et al., 2010; Langguth, Kreuzer, Kleinjung, & De Ridder, 2013; van den Berge et al., 2017). Rather, it has consistently been shown that many aspects of the tinnitus percept vary greatly across individuals according to their specific tinnitus history. Current brain models regarding tinnitus emergence and manifestation take this view into account (e.g., De Ridder et al., 2014; Sedley, Friston, Gander, Kumar, & Griffiths, 2016). Most of these models are distanced from the idea that tinnitus emerges exclusively in auditory areas but rather consider it to be coded in various sub-networks distributed across the whole brain. This holds in particular for individuals with a long history of tinnitus symptoms. In this context, a distress network has been proposed that includes structures of the limbic system (e.g., anterior cingulate cortex and amygdala), prefrontal areas (e.g., dorsolateral prefrontal cortex), and also the insula (De Ridder et al., 2014). Furthermore, Rauschecker, Leaver, and Mühlau (2010) found that the limbic system, in particular amygdala, nucleus accumbens, and ventromedial prefrontal cortex (vmPFC), are part of an inhibitory frontostriatal gating mechanism. They found that these structures via the thalamic reticular nuclei, are part of a noise-cancelling feedback loop, which is able to protect the auditory central system from any unpleasant sounds under normal conditions (Rauschecker, May, Maudoux, & Ploner, 2015). Husain (2015) were able to confirm the involvement of limbic structures and found that the lack of habituation to the constant tinnitus tone might be explained by a suppressed response from amygdala or insula when conveying emotions. An even more comprehensive framework has recently been proposed by Sedley et al. (2016) on the basis of predictive coding. According to this group higher brain areas are responsible for decreasing top-down prediction errors and for resetting the default prediction of the auditory system to expect tinnitus.

Following these considerations, it is not surprising that neurofeedback training aimed at influencing activity in auditory areas exclusively does not always lead to additional benefits regarding tinnitus-related distress. Based on this indication, future tomographic neurofeedback treatments should target the large scale, whole-brain networks involved in tinnitus generation and maintenance instead of confining the training to auditory areas. For instance, previous EEG studies have shown that low theta and a high amount of beta oscillations in insula and anterior cingulate cortex might be linked to tinnitus (e.g., De Ridder, Congedo, & Vanneste, 2015; Meyer et al., 2017). Based on these findings, it may be beneficial to extend the auditory alpha/delta protocol with theta-up/beta-down training in insula and ACC.

4.2Tinnitus loudness

Data analysis revealed that, across the whole study group, neurofeedback led to a significant decrease in tinnitus loudness. However, this decrease was followed by an obvious (see Fig. 1) and, according to the Tukey post-hoc test, statistically meaningful return to baseline levels in the follow-up period. A possible explanation for this issue might be that the number of 15 training sessions was not high enough to affect the loudness of the percept in a sustainable way. However, two studies working with the same protocol (Crocetti et al., 2011; Dohrmann et al., 2007) both reported a stable decrease of tinnitus intensity with only 10 and 12 sessions, respectively. Another reason for this inconsistency might be the length of individual neurofeedback sessions which was 15 minutes in our study and 30 or 20 minutes in the aforementioned studies. Furthermore, patients in our project trained only once a week while the frequency was higher (2–3 times per week) in the previous studies. Therefore, higher intensity (in frequency or length) of neurofeedback training might be considered a crucial factor for longer-lasting effects regarding tinnitus loudness.

Also in the case of tinnitus loudness, the repeated-measures mixed model designs with group-terms did not show improved fits on the data, suggesting no between-group differences within or across time (see Table 3). This effect could not be shown despite the noticeable difference between groups in Fig. 1 and the exploratory analysis revealing significant effects for NTNF but not for TONF (see section 3.1.3). Therefore, the logical conclusion is that both groups reacted in a similar way regarding their tinnitus loudness (i.e., a decrease after the training, and an increase in the follow-up phase). As in the previous case with tinnitus distress, it should be noted that tinnitus sub-networks also need to be taken into account for tinnitus intensity. Regarding the perception (experience) of the tinnitus sound, De Ridder, Elgoyhen, Romo, and Langguth (2011) have suggested that activation in auditory areas alone is insufficient. For conscious perception to arise, activity in sensory areas must be connected to networks composed of regions in the cingulate, parietal, frontal and most importantly the insular cortex that code the importance and salience of percepts (see also Meyer et al., 2016). Sensory stimuli are only consciously perceived after they have been classified as relevant and important by means of this sub-network. Since the recruitment of additional brain circuits is thus necessary for initial perception, also the intensity of a percept (in this case tinnitus loudness) is dependent on this co-activation. Neurofeedback focused specifically on auditory areas ignores these networks, a fact that should be considered in future studies.

4.3Electrophysiological parameters

An analysis of EEG-related parameters was performed in order to have proof of concept that the neurofeedback intervention presented here indeed leads to the proposed effects in the participants’ brain activity. Source data over the four voxels used for TONF and alpha/delta ratio over the four active electrodes of the NTNF group suggested that learning of the proposed EEG rhythm occurred in both groups. Repeated-measures mixed model ANOVA performed over both groups combined revealed significant effects of the factor time and, furthermore, significant ratio increases between pre and post measurements. In addition, these effects were confirmed 3 months after completion of the training period, suggesting a persistent change in resting-state brain activity.

When a group-term was included in the multilevel model, the fit was not improved (see Table 3). This suggests no meaningful differences between electrophysiological changes of the TONF and the NTNF group on both, surface and source levels. This contradicted our hypotheses as we expected surface-related changes to be more distinct for the NTNF group and changes on source level to be predominant in the TONF group. It is important to keep in mind, however, that both neurofeedback protocols were developed in order to target activity over auditory areas. The four fronto-central electrodes (FC1, FC2, F3, F4) of the NTNF group were chosen because of their high probability to reflect activity in auditory brain regions according to Pantev et al. (1995, 1) (as cited in Dohrmann et al., 2007). Hence it is reasonable to assume that the high validity of these electrodes for representing the auditory cortex might explain the unexpected training-induced alterations on source level for members of the NTNF group. This might serve as a possible explanation as to why electrophysiological parameters changed equally on surface and source levels for participants of both tomographic and non-tomographic neurofeedback groups.

4.4Limitations

One major limitation of the study design is the lack of a sham control group due to restrictions of time, funding, ethical reasons and other arguments discussed previously (Güntensperger et al., 2017). The risk of unspecific effects is particularly high in the context of chronic tinnitus where many participants hopefully turn to any new treatment approach after unsuccessfully trying a wide variety of therapies on their own. Expectations of researcher and participant and the effect of simply participating in a study may thus serve as alternate explanations for the reductions in THI and TQ scores. Kreuzer et al. (2012) for instance found around a 5 point decrease of TQ values and a 6 point THI reduction in a mere waiting-list control group. However, apart from a sole evaluation of subjective tinnitus symptoms, we focused our analysis extensively on not voluntarily modifiable neurophysiological data recorded with resting-state EEG. As has been extensively discussed in our recent publication (Güntensperger et al., 2019), this data at least partially speaks against a purely unspecific explanation. However, without a sham control group it is not possible to properly distinguish specific and unspecific effects of the treatment and, therefore, to draw final conclusions about the effectiveness of this neurofeedback protocol. In this context, a recently published “Trials” article may soon provide novel insights (Jensen et al., 2020). The author outline a randomized controlled trial comparing an alpha/delta neurofeedback group with an active beta/theta training control group and a passive diary control group. The comparison of the results of these three groups may lead to a better understanding of specific and non-specific effects in neurofeedback treatment for tinnitus.

4.5Conclusion

Overall, alpha/delta neurofeedback seems to be a valid treatment addition for chronic tinnitus. Surface-based as well as tomographic neurofeedback both lead to sustainable effects on tinnitus-related distress. Effects on tinnitus loudness are transient but might be prolonged with higher training frequency (2–3 sessions per week) and longer individual training sessions (min. 20 minutes). It should be noted, however, that the higher specificity of tomographic neurofeedback might be disadvantageous in some cases as it exclusively focuses on auditory areas and ignores other tinnitus-relevant areas (i.e., areas of the distress and/or salience network). Future studies should try to include these regions (e.g., anterior cingulate or insular cortex) as targets for tomographic neurofeedback in order to develop more specific neurofeedback protocols.