Neurofeedback training in major depressive disorder: a systematic review of clinical efficacy, study quality

and reporting practices

Lucas R. Trambaiolli1, Simon H. Kohl2,3, David E.J. Linden4, David M.A. Mehler5
1
Division of Basic Neuroscience, McLean Hospital, Harvard Medical School, Boston, USA
2
JARA Institute Molecular neuroscience and neuroimaging (INM-11), Jülich Research Centre, Germany
3
Department of Child and Adolescent Psychiatry, Medical Faculty, RWTH Aachen University, Germany
4
School for Mental Health and Neuroscience, Faculty of Health, Medicine and Life Sciences, Maastricht University, The Netherlands
5
Department of Psychiatry, University Hospital Münster, Germany

Abstract: Major depressive disorder (MDD) is the leading cause of disability worldwide. Neurofeedback training has been suggested
as a potential additional treatment option for MDD patients not reaching remission from standard care (i.e., psychopharmacology and
psychotherapy). Here we systematically reviewed neurofeedback studies employing electroencephalography, or functional magnetic
resonance-based protocols in depressive patients. Of 585 initially screened studies, 24 were included in our final sample (N=480 patients
in experimental and N=194 in the control groups completing the primary endpoint). We evaluated the clinical efficacy across studies
and explored its relationship with the number of sessions as a potential proxy for a dose-effect response. We also attempted to group
studies according to the control condition categories currently used in the field that affect clinical outcomes in group comparisons. In
most studies, MDD patients showed symptom improvement superior to the control group(s). However, most articles did not comply
with the most stringent study quality and reporting practices. We conclude with recommendations on best practices for experimental
designs and reporting standards for neurofeedback training.

Keywords: Neurofeedback; Biofeedback; real-time fMRI; electroencephalography; functional magnetic resonance imaging, Major
depressive disorder; Self-regulation; Neuroimaging

1. Introduction

Major depressive disorder (MDD) is a serious mental disorder characterized by at least one depressive episode lasting for two or more
weeks (Association, 2013). This episode includes symptoms such as changes in cognition, reduced mood, interest or pleasure, and
vegetative complaints (Otte et al., 2016). MDD has been recognized as a major public health challenge because of the increasing number
of cases worldwide. For Western countries, it is estimated that MDD affects one in every five to six adults (Bromet et al., 2011; Patten,

1
2009). MDD represents a major risk factor for suicide attempts (Hoertel et al., 2015; Olfson et al., 2017). Moreover, MDD patients often
suffer from comorbid psychiatric conditions (Alonso and Lépine, 2007), which increases the burden on patients and their families.
Current treatments mainly include psychotherapy or pharmacotherapy (Kupfer et al., 2012). The most widely used type of
psychotherapy for depression is cognitive behavioral therapy (CBT), which aims to identify the cognitive factors leading to depressive
symptoms and develop mental and behavioral strategies to cope with these (Otte et al., 2016). Another psychotherapeutic approach
developed for the treatment is cognitive bias modification, which aims to readjust negative attention biases commonly observed in
depressed patients (Fodor et al., 2020). The mainstay of current pharmacotherapy for depression are monoaminergic antidepressant
drugs (Sharp, 2012). However, around one third of depressed patients do not respond to these conventional treatments (Fava and
Davidson, 1996; Rush et al., 2006). Other therapeutic options include non-invasive brain stimulation such as transcranial magnetic
stimulation (TMS) and electroconvulsive therapy (ECT), for which several stimulation protocols have been developed that show
superiority compared to sham stimulation (Mutz et al., 2019). However, TMS and ECT can yield aversive effects, including local pain,
headache and discomfort (Cusin and Dougherty, 2012; Rossi et al., 2009). Some ECT patients report acute but partly also persistent side
effects of amnesia and cognitive disturbances following treatment (Sackeim et al., 2007). Lastly, invasive electrical deep brain
stimulation (DBS) of subcortical and cortical areas is currently explored for its clinical potential (Delaloye and Holtzheimer, 2014),
although most recent findings remain inconclusive and have sparked a debate in the field (Bari et al., 2018). One common feature of
electrical or magnetic brain stimulation treatments, shared with pharmacological treatment, is that patients remain passive recipients of
the intervention.
In contrast, non-invasive neurofeedback training is a neuromodulation technique that involves patients as protagonists of their
treatment. Patients learn self-regulating particular features of brain activity (Sitaram et al., 2017) by actively engaging in processes
which are often adopted from techniques used in psychotherapy (Arns et al., 2017; Fovet et al., 2015). However, given the current
discussion regarding the specificity and efficacy of neurofeedback protocols across psychiatric disorders (Thibault et al., 2018), a formal
evaluation within specific conditions is much needed. MDD can be considered one of the most extensively studied applications of
neurofeedback training, with the first case studies reported more than two decades ago (Baehr et al., 1997; Earnest, 1999; Rosenfeld et
al., 1996). This systematic review pursues three main goals: First, we describe the different neurofeedback protocols that have thus far
been explored with MDD patients and the main clinical and neural outcomes of these studies. Second, we summarize reported clinical
changes and evaluate their efficacy. Lastly, we assess the study design and reporting quality of published research articles. We discuss
limitations and open challenges, closing with a set of recommendations for future neurofeedback studies in MDD that may help
advancing the field.

1.1 Description of a neurofeedback system

Neurofeedback is a non-invasive technique that provides the user with real-time feedback about their neural self-regulation performance.
Feedback is commonly provided from areas that are thought of as putative neural substrates underlying specific behaviors or pathologies

2
(Kim and Birbaumer, 2014; Sitaram et al., 2017). For instance, one well described and commonly found symptom in MDD is low mood.
Several neurofeedback studies trained patients on neural correlates of emotion regulation with the aim to improve this capacity and the
depressive symptom(s) related to low mood. Different imaging modalities have been used to train self-regulation in healthy participants
and/or patients, including electroencephalography (EEG), magnetic encephalography (MEG), functional magnetic resonance imaging
(fMRI), and functional near-infrared spectroscopy (fNIRS) (Thibault et al., 2016). Irrespective of the imaging modality, neurofeedback
interventions usually consist of four main steps (Paret et al., 2019; Sitaram et al., 2017): 1) identifying the neural target (i.e., correlate
of a symptom or skill) either by the means of functional data during a so-called localizer session or based on previous anatomical
hypotheses using masks, 2) recording the neural activity of this neural target, 3) processing these measures while ideally controlling for
potential artefacts and 4) presenting real-time feedback of this signal to the user.
At the recording stage, i.e. step one and two, the nature of brain signals needs to be considered as it differs between imaging
modalities such as EEG and fMRI. For instance, EEG has been frequently used in depressed patients to search for neural correlates of
mental states and later explored to develop neurofeedback protocols (Enriquez-Geppert et al., 2017; Gruzelier, 2014). EEG uses scalp
electrodes, which similar to MEG, measure local field potentials (LFPs). LFPs represent the summed activity of local neural populations
reflecting the electric potential in the extracellular space. Hence, EEG signals are largely determined by post-synaptic activity providing
a direct measure of neural activity (Da Silva, 2009). fMRI is another neuroimaging technique that is increasingly used for neurofeedback
experiments in depression (Watanabe et al., 2017; Weiskopf, 2012). This technique uses the blood oxygen level dependent (BOLD)
contrast, a measure for the relative changes in local blood oxygenation that result from the metabolism of brain cells. fMRI hence
provides an indirect measure of neural activity. More recently, fNIRS has gained the attention of the neurofeedback community (Kohl
et al., 2020). fNIRS uses near-infrared light to measure local changes in oxygen concentrations in cortical gyri (Hoshi, 2003; Strangman
et al., 2002; Villringer et al., 1993); these strongly correlate with the fMRI BOLD signal (Cui et al., 2011; Huppert et al., 2006;
Strangman et al., 2002). Different neuroimaging technologies have their advantages and disadvantages, in particular, for real-time
experiments (Thibault et al., 2016). For example, EEG provides higher temporal resolution and reduced cost compared to fMRI and
fNIRS, and wireless-EEG systems provide new perspectives for portable therapeutic applications in the near future (De Vos et al., 2014;
Ries et al., 2014). Conversely, fMRI possesses a higher spatial resolution, which allows the development of protocols that target both
cortical and subcortical areas composing the circuitry of interest (Sulzer et al., 2013; Weiskopf, 2012). Multi-modal neurofeedback
approaches attempt to bridge these advantages and compensate for some disadvantages by combining two or more neuroimaging
techniques (Mano et al., 2017). For instance, these studies may benefit from the spatial resolution of fMRI and temporal resolution of
EEG, combining these with the aim to achieve higher self-regulation performance (Perronnet et al., 2017).
The third step involves data processing methods. However, in real-time experiments, data preprocessing and data acquisition are
quasi-simultaneous (depending on the delay of the respective imaging technique) such that recorded brain signals are continuously
converted to an output system (Sitaram et al., 2017). For all brain imaging modalities noise-reduction methods are essential to increase
the validity of the feedback; they are ideally applied to filter non-neural signal sources, such as the electrooculography (EOG) and
electromyography (EMG) in EEG-based protocols (Moretti et al., 2003), or respiration and pulse waves in fMRI-based protocols

3
(Murphy et al., 2013). Artifact corrected time-series are subsequently processed to calculate values that are subsequently used for
feedback presentation. These values can be based on signal changes with respect to baseline in individual brain areas, correlations
between time series of different brain areas (connectivity based feedback) (Koush et al., 2013; Ramot et al., 2017), or the output of more
complex algorithms that classify different brain states based on variations in brain activity patterns (Watanabe et al., 2017). However,
all these processing methods vary according to the neurofeedback paradigm and are subject to ongoing methodological research and
development (Brigadoi and Cooper, 2015; Heunis et al., 2019; Heunis et al., 2020; Hinterberger et al., 2003; Krusienski et al., 2006;
Lotte et al., 2007).
The fourth and final design step concerns the presentation of real-time feedback. Although visual feedback is the most common
approach, other feedback modalities can also be used in this stage and include auditory, vibrotactile, electrical or proprioceptive
stimulation (Sitaram et al., 2017). The feedback setup should be carefully designed because it can cause distraction, frustration, or even
induce negative emotions in users (Birbaumer et al., 2013; McFarland et al., 1998). The feedback system should constantly update the
trainee about the targeted neural activity. Such real-time feedback allows the trainee to create, correct and optimize a mental or
behavioral control strategy and thereby to achieve the desired level of proficiency in self-regulating neural activity (Birbaumer et al.,
2013; Curran and Stokes, 2003).

1.2. Study design and non-specific effects of neurofeedback protocols

When conceptualizing this systematic review we were guided by a recently published consensus statement that discussed different
mechanisms responsible for driving the outcomes of a neurofeedback experiment (Ros et al., 2020). The authors identified five potential
contributors (Micoulaud-Franchi and Fovet, 2018; Ros et al., 2020): neurofeedback-specific effects, which are related to the actual
training of a target neurophysiological variable (e.g., increased or decreased functional connectivity between trained ROIs); non-specific
neurofeedback effects, associated to the neurofeedback context, but not to the trained neural signals (e.g., the high-tech
environment); general non-specific effects, which are caused by psychosocial influences (e.g., believe-based expectations); repetition
related effects, referring to the recurrence of training (e.g., test-retest improvements due to mental imagery tasks employed in
neurofeedback paradigms); and, finally, natural effects, associated to natural events in life (e.g., natural recovery or remission). The
extent to which these factors contribute to overall clinical effects as observed in experimental (and to some degree also control) groups
remains subject to ongoing and future research. Given that all potentially contributing factors as listed above likely play a role, and that
they even interact with each other, some authors recently described neurofeedback as a complex intervention when studied in a clinical
context (Craig et al., 2008; Sorger et al., 2019).
Similar to other interventions, developing a neurofeedback paradigm for clinical application requires several phases.
Uncontrolled single-group designs are suitable for the early phase, for instance, to assess technical feasibility and acceptability of the
paradigm in a healthy or patient sample. "Exploratory trials" may also serve to optimize the intervention in the targeted patient population
(similarly to Phase I Clinical Trial designs) (Sorger et al., 2019). However, single-group designs cannot control for non-specific effects.

4
Thus, further experiments with appropriate control conditions are needed during later phases to disentangle the neurofeedback-specific
outcomes from those caused by other (psychosocial) mechanisms (Thibault et al., 2018; Thibault and Raz, 2016).
One main challenge that the neurofeedback field currently faces is that standards for the design of randomized controlled trials
are traditionally based on the requirements that pharmacological studies need to fulfill. This challenge pertains in particular to the design
of control conditions and risk of unblinding. For instance, in pharmacological studies the control group can receive a highly comparable
treatment that omits the active component to drive improvement (Linden, 2014; Thibault et al., 2018). In such trials where participants
receive so-called passive treatment (i.e., the intervention does not require a specific engagement in a task), the design of control
conditions mainly need to account for belief-based expectations (commonly referred to as “placebo effects”) rather than a range of
contributing factors as listed above. Moreover, complex interventions such as neurofeedback that involve active engagement of the
participant have their own requirements to ensure blinding (Linden, 2014; Sorger et al., 2019) (noteworthy, the design of appropriate
placebo control conditions also remains subject for discussion in the pharmacological literature (Jensen et al., 2017; Moncrieff et al.,
2004)).
Recent discussions in the field have therefore resulted in new best-practice research recommendations for different control
conditions (for a detailed framework, please refer to (Sorger et al., 2019)). For this review, we grouped control conditions into three
main categories:
 Passive control: this category includes control conditions that involve continued standard care only. Passive control conditions
can reveal whether the neurofeedback has clinically significant benefit as a stand-alone, or add-on, intervention compared to
standard care, for instance (Choi et al., 2011; Escolano et al., 2014; Wang et al., 2019; Wang et al., 2016). While this design
controls for natural effects (e.g., regression to the mean), it does not control for any general or neurofeedback training related
non-specific effects.
 Active control outside the scanner: this category includes control conditions where the participant is engaging in a similar mental
task, but outside of the neuroimaging scanner (Jaeckle et al., 2019; Linden et al., 2012). This condition is also referred to as a
mental-rehearsal control (Sorger et al., 2019). In addition to natural effects, it also allows to control for repetition related effects
that occur by merely engaging in the behavioral/cognitive strategy.
 Active control inside the scanner: this category includes a variety of approaches in which the patient is actively performing a
task inside the scanner and that may either control for neurofeedback specific effects, non-specific neurofeedback effects, or both
(for a more detailed overview, please refer to [61]). For example, patients in the control group are trained to self-regulate their
brain activity in the same ROI but in the opposite direction of the experimental group. In an alternative design, they receive
feedback from a different ROI or network using an alternative strategy (Mehler et al., 2018). Such designs match groups for
some general non-specific and non-specific neurofeedback effects such as motivation, received reward during training, the high-
tech environment, or the interaction with the experimenter and allow testing for neurofeedback specific effects (Thibault and
Raz, 2016; Wood and Kober, 2018). Other approaches present feedback based on signals from a different brain source not
associated with the brain function targeted in the experimental group (Young et al., 2017b; Young et al., 2014), sham signal

5
 (e.g., randomly generated), non-neural sources (e.g., other biological features) or yoked data (e.g., data from a different
participant (Hamilton et al., 2016)). However, it is essential to match groups for perceived rewards and evaluate to which extent
patients remain "blind", considering that previous studies report that they could detect non-contingency and experienced adverse
effects such as frustration and reduced motivation (Sorger et al., 2019).
Finally, it is also relevant to design "double-blind" or “triple blind” trial designs where the rater, participant and neurofeedback
operator are blinded to the treatment condition. While only a few neurofeedback software packages are currently capable to blind the
neurofeedback operator (Ros et al., 2020), double-blinding/triple-blinding can alternatively be achieved with two experimenters (these
are either responsible for operating, and if necessary programming, the experiment or interacting with participants (Arnold et al., 2013))
in addition to independent and blinded research or clinical staff who assess the outcomes (Ros et al., 2020).

1.3. Rationale for the use of neurofeedback for MDD

Neurofeedback feasibility studies have yielded first promising results in different non-clinical and clinical applications ranging
from athletic performance (Mirifar et al., 2017) to motor rehabilitation for neurodegenerative disorders and stroke (Krucoff et al., 2016;
Linden and Turner, 2016). In neuropsychiatry, small randomized controlled studies have shown benefits for different disorders. For
example, EEG-based training protocols were successfully applied to substance abuse disorders, eating disorders, attention-
deficit/hyperactivity disorder, autism spectrum disorder, tinnitus, and obsessive-compulsive disorder, while fMRI-based training
protocols have been successfully applied to attention-deficit/hyperactivity disorder (ADHD), post-traumatic stress disorder,
schizophrenia, Alzheimer’s disease, Tourette Syndrome, autism spectrum disorder, overweight/obesity, chronic pain, spider phobia, and
obsessive-compulsive disorder (Arns et al., 2017; Sitaram et al., 2017; Thibault et al., 2018). These studies provide preliminary data
suggesting that neurofeedback training may be effective in changing brain function and treating some neuropsychiatric symptoms,
including those related to disturbances in the reward system. Noteworthy, clinical effects have been reported to last also during
longitudinal follow-ups (Becerra et al., 2006; Gevensleben et al., 2010; Goldway et al., 2019; Mehler et al., 2018; Rance et al., 2018).
Most neurofeedback training paradigms are informed by neurophysiological or computational models suggested to explain the
genesis of depressive symptoms. Thereby, this technique provides potentially a new way to directly test for the causal validity of reported
biomarkers (Mehler and Kording, 2018; Micoulaud-Franchi et al., 2019). Similar to brain stimulation protocols (e.g. TMS or DBS) most
neurofeedback protocols aim to modulate local activity. Importantly, the neurofeedback acts as an “endogenous” stimulator, reducing
issues related to safety or side effects from conventional neuromodulation approaches. Additionally, such a form of non-invasive,
endogenous neuromodulation puts the patient at the center of the intervention and may hence result in beneficial psychophysiological
and psychosocial effects (see below) (Linden, 2014).
Besides local functional changes, several studies have also reported remote effects of neurofeedback training at the network
level. For instance, neurofeedback training has been reported to alter intrinsic functional connectivity (Hampson et al., 2011; Scheinost
et al., 2013) and directed effective connectivity (Zotev et al., 2011; Zotev et al., 2013). Moreover, these alterations (and related

6
symptomatic improvements) were partly found to persist for months, supporting the idea of long-term changes in network organization
(Megumi et al., 2015). Reports of long lasting network changes and associated clinical improvement render neurofeedback training a
particularly promising approach to treat patients that present with abnormal connectivity patterns in brain networks relevant for affective
and cognitive processing (Hamilton et al., 2015; Mulders et al., 2015).
Cognitive processes that have been suggested to contribute to the pathophysiology of depression include biased attention and
processing of negative stimuli, recall of negative memories (Lewinsohn and Rosenbaum, 1987; Sato and Kawahara, 2011), and recurrent
negative thoughts (rumination) (Beck, 2008; Clark and Beck, 2010; Disner et al., 2011). These cognitive processes share underlying
brain structures which are commonly reported as showing abnormal activity, or connectivity, in patients with depression, such as the
lateral and medial prefrontal cortex (PFC), anterior cingulate cortex (ACC), amygdala, hippocampus, and striatum (Groenewold et al.,
2013; Kaiser et al., 2015). Thus, the majority of neurofeedback protocols for MDD aim to directly or indirectly rebalance these networks.
For instance, in fMRI-based protocols these areas are the ones commonly targeted for self-regulation, as individual ROIs (Jaeckle et al.,
2019; Young et al., 2017b; Young et al., 2014) or as multi-ROI networks (Linden et al., 2012; Mehler et al., 2018). Although some
studies using EEG-based neurofeedback claim to target some of these brain structures (e.g. as the dACC and the amygdala (Walker and
Lawson, 2013)), we note that the relatively low spatial resolution and fidelity of EEG imposes substantial limitations and requires
validation. The most common approach relies on recordings from frontal channels to measure potential asymmetries in the alpha
frequency band (Choi et al., 2011; Hammond, 2005; Peeters et al., 2014; Ramirez et al., 2015; Wang et al., 2019; Wang et al., 2016).
This approach assumes that the hyper- and hypoactivation of opposite hemispheres indicate the valence experienced during emotion
regulation (Harmon-Jones et al., 2010) and that this marker may reflect symptoms of dysfunctional emotion regulation as commonly
observed in depressed patients (Thibodeau et al., 2006). Of interest, neurofeedback protocols that use simultaneous EEG-fMRI
recordings showed that EEG frontal asymmetry was correlated with activity in brain structures involved in emotion regulation in healthy
subjects (Zotev et al., 2013) and patients suffering from depression (Zotev et al., 2019; Zotev et al., 2016). However, the relationship
between these biological markers and the cognitive mechanisms for depression is still debatable. In this context, neurofeedback protocols
may provide additional validation of these mechanisms (Linden, 2014).
Apart from targeting neural correlates of MDD, neurofeedback training paradigms have also been designed to tap into the
interaction between psychological and biological aspects of the disorder (Deldin and Chiu, 2005). For instance, some neurofeedback
protocols incorporate aspects from cognitive therapy such as cognitive restructuring approaches and means of emotional self-regulation,
including training to self-regulate the response for valenced figures, autobiographical memories, or affective imagery (MacDuffie et al.,
2018; Skottnik and Linden, 2019). Thereby, mental imagery based neurofeedback training can potentially aid patients in developing
coping strategies to mitigate negative thoughts and value positive experiences (Clark and Beck, 2011). Further, the task engagement
itself in combination with contingent positive reinforcement during a neurofeedback session may result in behavioral activation and
modulate self-efficacy (Dimidjian et al., 2011), i.e. an individual’s sense of being in control of their environment and to cope with
challenges (Bandura, 1982; Mehler et al., 2018). Moreover, such psychophysiological effects are particularly relevant for the treatment
of depressed patients (but also other psychiatric patient populations) who often show deficits in these capacities. Of interest, it has been

7
noted that psychotherapy approaches and neurofeedback strategies may be mutually translatable (MacDuffie et al., 2018; Skottnik and
Linden, 2019). As such, neurofeedback training may provide a promising add-on tool to augment standard care treatment, supporting
patients in the process of cognitive restructuring and other learning processes initiated in psychotherapeutic sessions.

1.4. Aim of this review

Given the growing number of studies investigating neurofeedback applications as a treatment for MDD over the last decades, we aim
here to (I) summarize and compare current findings, (II) evaluate the quality of these studies, and (III) provide guidelines for future
research that can accelerate the field. Different from previous reviews (Linden, 2014; Sacchet and Gotlib, 2016; Young et al., 2018b),
we note that the present study comprises to our knowledge the first attempt of a systematic investigation of EEG and fMRI neurofeedback
training protocols in MDD patients. Also, to assess study design and reporting quality, we employed the Joanna Briggs Institute (JBI)
critical appraisal tools (Tufanaru et al., 2017) and “Consensus on the Reporting and Experimental Design of Neurofeedback studies”
(CRED-nf) checklist (Ros et al., 2020).

2. Methods

2.1 Systematic search

A systematic search on English peer-reviewed journal articles published until March 6th, 2020, was performed for this review. The
PubMed bibliographic database, and pre-print servers including life science papers (arXiv, medRxiv, psyArXiv, and bioRxiv) were
queried using the following search rule:
(biofeedback OR neurofeedback) AND (depression OR depressive)
Resulting articles were selected or rejected based on the criteria described in Table 1.

Table 1 - Eligibility criteria.

Inclusion criteria:
1) Studies presenting original results in human adults (> 18 years
old)
2) Studies including patients with a formally diagnosed current
depressive episode
Exclusion criteria:
1) Studies including patients with other psychiatric disorders (but
not major depressive disorder) in the experimental sample, or
targeting depressive symptoms in other disorders

8
 2) Studies exclusively evaluating healthy participants
3) Studies applying biofeedback based only on non-neural signals
4) Studies without voluntary control of brain activity
5) Studies with animal models
6) Case reports (n<5), reviews, commentaries, or editorials

As shown in Figure 1, a total of 577 journal articles were found in the PubMed and pre-print databases, and eight other papers
were included from other sources (papers cited in articles screened) (the list of articles is available on: https://osf.io/k76g2/). Through
relevance screening, 539 articles were rejected as they did not meet the inclusion criteria. After full-text examination, only 24 articles
were included in this systematic review.
To collect relevant information, a data extraction sheet was created including 23 data items which were extracted and grouped
into four categories: Study Design (diagnostic criteria, symptom scales, existing comorbidities, parallel treatments, randomization,
blinding, experimental paradigm, control paradigm, feedback modality, number of sessions, and follow-up), Clinical Outcomes (within
group differences post-NF, between groups differences post-NF, within group differences at follow-up, between groups at follow-up,
exclusions and drop-outs), Other Significant Outcomes (within group differences post-NF, between groups differences post-NF, within
group differences at follow-up, between groups at follow-up). One co-author (LRT extracted data from studies and another co-author
screened the extraction results (SHK). Disagreements between the reviewing authors were resolved by discussion.

9
 Figure 1 - Search decision flow diagram according to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (Moher et al., 2009).

2.2 Assessment of clinical efficacy

When evaluating clinical outcomes between studies, three other design aspects (besides differences in the experimental paradigm) make
it difficult to compare clinical effects as reported across studies: (1) Studies vary greatly in their control conditions, ranging from no
control or passive control conditions, where patients merely engage in training mental coping strategies, to active neurofeedback control
conditions that are closely matched for various psychosocial factors including reward, successful self-regulation experience, regular
interactions with practitioners/researchers and practicing mental coping strategies (Thibault et al., 2018). (2) Studies vary in their
inclusion criteria, which may also pertain to baseline severity levels of depressive symptoms. For established treatments in depression,
it is well known that baseline differences account for part of the variance of clinical improvements and superiority of treatment over
non-specific psychosocial effects (Fournier et al., 2010; Kirsch et al., 2008) and hence differences in baseline severity may bias results.
(3) Studies used different outcome measures (i.e., numerical scales). While calculating standardized effect sizes can account for

10
differences in used outcome measure (see 3), they do not account for differences in baseline severity (see 2). To address the first aspect
(heterogeneity in control conditions), we grouped studies into one of four categories: no control, passive control, active control outside
the scanner, or active control inside the scanner. To address the second and third aspect (i.e., potential baseline differences and
heterogeneity in clinical scales), we normalized clinical percentage changes reported for individual studies by the maximum score of
the respective clinical scale that was being used (Figure 2A). Symptom improvement scores were computed as the percentage change
of the primary outcome measure with respect to baseline. For studies that did not declare their primary outcome (Choi et al., 2011;
Hammond, 2005; Paquette et al., 2009; Peeters et al., 2014; Ramirez et al., 2015; Walker and Lawson, 2013; Wang et al., 2016; Yuan
et al., 2014), we considered all clinical outcome measures (if multiple were reported) as secondary and adopted a conservative approach
selecting the symptom scale with the least percentage change.
Lastly, to compare clinical effects between neurofeedback and other interventions, we computed the number needed to treat
(NNT) for studies reporting remission rates. Specifically, following Altman’s recommendation, we refer here to the “number needed to
treat for one additional patient to benefit, or to be harmed” (i.e., worsening of depressive symptoms), i.e. NNTB and NNTH, respectively
(Altman, 1998), and report these point estimates alongside their 95% confidence intervals using the Wilson score method. In contrast to
the widely used Wald method, the Wilson score method is expected to yield less biased results for studies with relatively small sample
sizes or unbalanced designs (Bender, 2001; Newcombe, 1998) and thus seemed more appropriate for the current sample. Calculations
were performed modifying a custom written script originally created by Bender (Bender, 2001) using the statistical software “Statistical
Analysis System” (SAS, version 9.4). The SAS script that also includes the extracted data for reported remission rates is available:
https://osf.io/jw7mu/.

2.3 Assessment of experimental design and reporting quality

To enhance reporting standards in the neurofeedback field, the recently published “Consensus on the Reporting and Experimental Design
of Neurofeedback studies” (CRED-nf) checklist suggests “essential” and “suggested” items around design and reporting aspects,
including pre-experiment registration, control groups and measures, feedback specifications, outcome description and data
storage/publishing (Ros et al., 2020). Two of the coauthors (LRT and SHK) independently rated the studies included in this review
according to the 23 criteria of the CRED-nf checklist. Disagreements between the reviewing coauthors were resolved by discussion.
Moreover, we also assessed the methodological quality of included studies based on the checklist for quasi-experimental studies
of the Joanna Briggs Institute (JBI) critical appraisal tools (Tufanaru et al., 2017). The JBI checklist has been used in various
experimental fields and thus allows comparing standards between neurofeedback studies but also entire fields. The JBI checklist includes
items such as clarity of cause and effect, similar participants, similar treatment in compared groups, existence of a control
group/condition, multiple measurement points of the outcome, completion of follow-up, similar outcome measurements in compared
groups, reliability of outcome measurements, appropriate statistical methods. The same coauthors who rated studies according to the
CRED-nf evaluated the studies included in this review according to the nine criteria of the JBI checklist. Similar to Kohl et al. (Kohl et

11
al., 2020) we adapted some of the CRED-nf items to account for the fact that most studies were published before these guidelines (see
Supplementary Material for details).

2.4. Contacting authors

The data extraction sheets, JBI and CRED-nf score tables, and a preprint version of the manuscript were shared with all corresponding
authors of the included studies to ask for corrections. Five of 16 (~31%) corresponding authors replied to our enquiry and either approved
the data extraction or suggested minor corrections.

3. Results

Thirteen of the 24 studies included in this review used EEG neurofeedback protocols: six studies including only frontal channel (targeting
the structures from the frontal cortex); four studies combining frontal channels with other portions of the scalp; and three studies looking
at regions other than the PFC. The remaining eleven studies applied real-time fMRI neurofeedback protocols: six studies targeting the
amygdala exclusively; three studies targeting different networks (two studies including emotion processing network, one the salience
network); and two studies with two or more distinct regions. Following this review's first aim, and given the heterogeneity of protocols,
we first provide a detailed overview of the various experimental protocols used for EEG (Section 3.1) and fMRI (Section 3.2)
neurofeedback studies, respectively. In these sections, we emphasize on study designs and training paradigms (Table 2), clinical
outcomes at primary endpoint and at follow-up (if reported; Table 3) as well as other statistically significant cognitive or neural effects
(if evaluated; Table 4). As far as reported by the authors of the original studies, we also extracted information about co-occurring
standard treatment for depression, including psychopharmacological medicine (patients are referred to as medicated) and psychotherapy.
We follow with an investigation of clinical effects grouped by control condition (Section 3.3) as well as drop-out rates and side-
effects (if reported; Section 3.4). Specifically, we compared baseline scores and changes of the primary outcome measure (or if not
declared, we used the secondary outcome measure with the least improvement; Figure 2) to evaluate clinical efficacy. We further
explored the relationship between the number of neurofeedback training sessions and clinical improvement (Figure 3), and computed
the number needed to treat for one additional patient to benefit (or to be harmed) [NNTB/NNTH] (Altman, 1998) for studies that declared
a primary outcome measure and reported remission rates (Table 5).
Lastly, following this review's second aim, we report quality scores for study design and reporting (Figure 4) and discuss these
findings in the context of best practice recommendations (Section 3.5).

3.1. EEG neurofeedback paradigms and clinical effects

12
In the first controlled, non-blinded pilot EEG neurofeedback study in MDD, Schneider et al. (1992) compared the ability of medicated
patients (N=8) and healthy controls (N=8) to regulate slow cortical potentials (SCP) at the central electrode (Cz). The patient group
showed significantly higher control of the system, and the authors reported a negative correlation between on-task SCP and the onset of
illness, and a correlation in the opposite direction between SCP and the number of hospitalizations pre training (Schneider et al., 1992).
However, no clinical changes after the neurofeedback training were reported by the authors.
Later, neurofeedback researchers became interested in protocols that exploited spectral lateralization observed in frontal
electrodes in response to mood induction (Harmon-Jones et al., 2010; Palmiero and Piccardi, 2017). Four neurofeedback studies reported
frontal alpha asymmetry as the main feature. Alpha asymmetry is calculated as the difference in the alpha spectral power between left
and right frontal channels F3 and F4. The first randomized, controlled and non-blinded study that used this approach in non-medicated
patients compared changes in (self-rated and clinician-rated) depression scales between a group that engaged in EEG frontal alpha
asymmetry neurofeedback training (N=12) and a control group that received psychoeducation (N=11) (Choi et al., 2011). After 10
training sessions, the neurofeedback group showed a significant improvement of the depressive symptoms (reduction of more than 60%
for the HDRS-17 and BDI-II scores), which persisted in the neurofeedback group at one-month follow-up (no follow-up reported for
the control group). However, one main limitation of the study was the lack of blinding and the fact that patients were already partly
remitted at enrolment (five in the experimental and two in the control group). Another research group later employed an uncontrolled,
single-arm study with nine patients (medication status unclear) suffering from moderate to severe depression who underwent a similar
training for a maximum of 30 sessions (three per week) (Peeters et al., 2014). Partial (defined as at least 50% reduction) and total
remission (a score of ≤ 6) were reported for four and one patient based on self-rated depression scale (QIDS-SR16). Moreover, this study
reported a significant correlation between symptom improvement and the normalization of the frontal alpha balance.
A similar alpha asymmetry EEG training protocol was tested later by Wang et al. (2016) in a randomized non-blinded controlled
pilot study with medicated patients. However, the authors found no significant difference between patients performing neurofeedback
(N=7) and those undergoing psychoeducation (N=7). Later, the same research group expanded this design by another control group and
recruited additional patients. Specifically, they compared in a non-randomized, non-blinded two-arm follow-up study the efficacy of
alpha asymmetry neurofeedback (N=24) and beta parietal asymmetry neurofeedback training (N=23) in medicated patients. Wang et al.
(2019) found that both groups showed significant improvement in depressive (more than 10%) and anxiety related (around 9%)
symptoms. When compared to a control group receiving placebo therapy (N=23), the alpha neurofeedback showed significant
improvement of anxiety symptoms, while the beta neurofeedback showed a significant effect of depressive symptoms (Wang et al.,
2019). Exploratory post-hoc analyses suggested that changes in depressive symptoms (BDI scale) were positively correlated with the
beta variation in left and right parietal electrodes P3 and P4 (Chen and Lin, 2020).
Two independent studies followed a different approach combining the frontal alpha asymmetry with other EEG features: in an
uncontrolled, non-blinded single-arm pilot study, Hammond (2005) combined beta up-regulation with alpha and theta down-regulation
over left and right frontal electrode channels F3 and F4 in a small sample (N=8). Seven patients showed improvement in the Minnesota
Multiphasic Personality Inventory (MMPI) scale (with an average reduction of about 30%). Of interest, these improvements in

13
depressive symptoms were largely maintained at follow-up one year later (Hammond, 2005). Another uncontrolled, unblinded single-
arm pilot study explored the up-regulation of alpha and beta ratios in frontal channels (AF3, AF4, F3 and F4) during increased arousal
and valence in depressed elderly patients (N=6) (Ramirez et al., 2015). After ten training sessions, patients presented approximately
17% improvement of self-rated BDI scores. We note that both studies did not provide detailed information about other concomitant
treatment (Table 2).
Walker and Lawson (2013) used a different approach to measure lateralization in an uncontrolled single-arm study at a non-
academic institution (a commercial EEG neurofeedback clinic). They enrolled 183 MDD patients that showed no sufficient improvement
after previous psychopharmacological treatment (i.e., were considered treatment resistant). This sample constitutes the largest thus far
collected in the depression neurofeedback literature. However, overall, the report and documentation were very brief, no further
information about patients (e.g., forms of compensation) was reported and a potential conflict of interest was not declared. Patients were
trained to reduce spectral power in theta and increase spectral power in beta frequencies at the right frontopolar channel (FP2) during
six training sessions. The underlying assumption of this training protocol was that modulations of these frequencies in the intended
direction would mimic effects of deep brain stimulation in Brodmann area 25 and entrain inhibitory effects on the amygdala non-
invasively (Walker and Lawson, 2013). However, no source localization analysis was performed. This intervention led to significant
reductions of self-reported symptoms for more than 80% of patients (group average showed a reduction of approximately 44%) (Walker
and Lawson, 2013).
Moving to other areas of the scalp, three studies combined data from frontal channels and temporal and posterior channels in
their protocols. The uncontrolled single-arm study by Paquette et al. (2009) focused on the reduction of high-beta power in fronto-
temporal channels (AF3, AF4, T3 and T4) while inhibiting negative thoughts. After 20 sessions, medicated patients (N=27) presented
an approximately 73% reduction of BDI symptoms. Further, 20 patients did not meet the DSM-IV criteria for MDD anymore. One
month after the end of the treatment, source localization analysis found reduced beta frequencies in emotion-related brain areas including
the orbitofrontal cortex, temporal lobe, amygdala and cingulate cortex (Paquette et al., 2009). Further, Escolano et al. (2014) performed
a non-randomized non-blinded trial in medicated patients, comparing an experimental group (N=40) that underwent alpha power
upregulation training over parieto-occipital channels (performing mental arithmetic) with a control group that received continued
standard care (N=20) in which patients received only their prescribed psychopharmacological medication. Behavioral outcomes
suggested that the intervention group showed increased alpha EEG power and improved cognitive symptoms (working memory)
(Escolano et al., 2014). However, the study did not report any changes of clinical changes despite assessing these at baseline.
In an uncontrolled single-arm study, Cheon et al. (2016) included an experimental group (N=20) of medicated patients that
trained to up-regulate beta power at F3 and down-regulate the alpha/theta ratio in the Pz electrode. After 16 to 24 sessions, patients
presented approximately 70% of reduction in both the Hamilton Depression Rating Scale (HDRS) and the Hamilton Anxiety Rating
Scale (HARS). Later, the same research group adapted the same approach to subject-dependent protocols that were calibrated based on
patients’ symptoms and could include the self-regulation of sensorimotor rhythms or beta band in the electrodes F3, T3, or T4, followed
by down-regulation of the alpha/theta ratio in the Pz electrode (Lee et al., 2019). The choice of the best protocol was based on a previous

14
study evaluating the efficacy of different protocols to reduce specific symptoms (for example, low attention, low self-esteem, high
depression, or high anxiety) across different psychiatric disorders (Cheon et al., 2015). The study was non-randomized and not blinded.
All patients continued pharmacological medication treatment during the course of the study. While the experimental group (N=12)
received neurofeedback training as augmentation, the control group (N=12) received supportive psychotherapy. Results suggested that
patients in the experimental group showed significantly more clinical improvements with about 60% reduction in the primary clinical
outcome after 12 to 24 sessions, while patients in the control group improved only by about 10%. Lastly, we note that some studies were
not included in our primary analyses due to inclusion and exclusion criteria (Table 1); they are briefly summarized in the Supplementary
Material.

15
 Table 2 – Overview of experimental designs of studies using neurofeedback protocols in depressive patients (additional clinical details are provided in Table S1
in the Supplementary Material). ↑ = upregulation; ↓ = downregulation; N = sample size; E = Experimental group; C = Control Group; NR = Not Reported;
Studies with overlapping samples are highlighted in gray.
Abbreviations: BAI = Beck Anxiety Inventory; BDI = Beck Depression Inventory; BPRS = Brief Psychiatric Rating Scale; CGI = Clinical Global Impression;
DSM = Diagnostic and Statistical Manual of Mental Disorders; DSRT = Depression Self-Rated Test; GAS = Global Assessment Scale; HDRS = Hamilton
Depression Rating Scale; HARS = Hamilton Anxiety Rating Scale; MADRS = Montgomery-Asberg Depression Rating Scale; MMPI = Minnesota Multiphasic
Personality Inventory; PHQ = Patient Health Questionnaire-Depression; QIDS-SR16 = Quick Inventory of Depressive Symptoms - Self-Report Version; STAI =
State/Trait Anxiety Scale; Psychopharm. Medication = Psychopharmacological Medication; Clinical/psychometric outcome measures reported as baseline mean
± standard-deviation.

Follow up
(in weeks
Training target in with
Experimental Group Control Condition reference Feedback
(N of allocated (N of allocated Comorbidity Clinical and psychometric Co-occurring Number of to the
subjects / final subjects / final Diagnostic (see details outcome measures (baseline Treatment Random NF primary
Study sample) sample) criteria in Table S1) levels) (NF/controls) ization Blinding sessions endpoint)
EEG
Healthy controls ↑ or ↓
↑ or ↓ slow cortical
slow cortical HDRS-17 (NF: 21.3±4.74) Psychopharm. Continuous
Schneider et potentials in Cz 20 (NF)
potentials in Cz DSM-III-R NR GAS (NF: 40.1±7.95) Medication (8/0) No No No and
al. (1992) (Ninit=8 / 5 (controls)
(Ninit=8 / BPRS (NF: 46.5±6.72) Psychotherapy (NR) visual
Nfinal=8)
Nfinal=8)
↑ beta and ↓ alpha and
Psychopharm.
Hammond theta in Fp1 and F3 20.75 Yes (48 -
No MMPI NR MMPI (NF: 95.75±NR) Medication (Yes-NR) No No Non-specified
(2005) (Ninit=9 / (average) average)
Psychotherapy (0)
Nfinal=8)
↓ beta in AF3, AF4,
T3 and T4 while ↓ Psychopharm. Continuous
Paquette et al. BDI-II (NF: 37.3±9.0)
negative thoughts No DSM-IV Yes Medication (Yes-NR) No No 20 Yes (4) and
(2009) BAI (NF: 18.5±0.3)
(Ninit=30 / Psychotherapy (NR) visual
Nfinal=27)
HDRS-17 (NF: 11.33±7.52;
↑ alpha asymmetry in C: 12.36±7.67)
Psychoeducation Psychopharm. Continuous
Choi et al. F3 and F4 BDI-II (NF: 22.75±12.35; C:
(Ninit=12 / DSM-IV No Medication (0/NR) Yes No 10 Yes (4) and
(2011) (Ninit=12 / 26.18±16.21)
Nfinal=11) Psychotherapy (1/1) audiovisual
Nfinal=12) MMPI-2 (NF: 62.08±12.61;
C: 67.00±16.07)
↓ alpha and ↑ beta in Frequency
Walker and QIDS- Psychopharm.
Fp2 non-specified
Lawson No SR16 No QIDS-SR16 (NF: 11.8±5.0) Medication (0) No No 6 Yes (48)
(Ninit=183 / and
(2013) DSRT Psychotherapy (NR)
Nfinal=183) auditory
↑ alpha asymmetry in
Psychopharm. Continuous
Peeters et al. F3 and F4 26.78
No DSM-IV Yes QIDS-SR16 (NF: 18.4±7.2) Medication (Yes-NR) No No No and
(2014) (Ninit=9 / (average)
Psychotherapy (NR) visual
Nfinal=9)

16
 Standard care
↑ upper alpha band in (continued BDI-II (NF: 23.70±13.51; C:
Psychopharm. Continuous
Escolano et al. parieto-occipital pharmacological 22.25±11.74)
DSM-IV Yes Medication (37/18) No No 8 No and
(2014) channels (Ninit=50 / treatment) PHQ-9 (NF: 13.33±6.84; C:
Psychotherapy (NR) visual
Nfinal=40) (Ninit=24 / 15.65±5.96)
Nfinal=20)
↑ alpha and beta
rations in channels
Continuous
AF3, AF4, F3 and F4 Psychopharm.
Ramirez et al. Non- and
during ↑ arousal and No NR BDI (NF: 15.5±9.90) Medication (NR) No No 10 No
(2015) specified auditory
valence Psychotherapy (0)
(music)
(Ninit=10 /
Nfinal=6)
↑ beta at F3 and ↓ HDRS-23 (NF: 21.38±5.82)
Frequency
alpha/theta ration in HARS (NF: 19.43±8.70) Psychopharm.
Cheon et al. DSM-IV- non-specified
Pz No No BDI-II (NF: 25.25±7.91) Medication (12) No No 16 to 24 No
(2016) TR and
(Ninit=20 / BAI (NF: 19.75±12.76) Psychotherapy (NR)
audiovisual
Nfinal=20) CGI (NF: 3.79±1.30)
Standard care
↑ alpha asymmetry in (continued BDI-II (NF: 30.14±10.25; C:
Psychopharm.
Wang et al. F3 and F4 pharmacological 22.86±13.03)
DSM-V No Medication (6/6) Yes No 6 No Non-specified
(2016) (Ninit=7 / treatment) BAI (NF: 17.86±10.51; C:
Psychotherapy (0/0)
Nfinal=7) (Ninit=7 / 16.00±9.92)
Nfinal=7)
Standard care
Self-regulate SMR or
(continued
beta band in F3, T3, or HDRS-17 (NF:24.33±5.77;
pharmacological
T4 (according to C:23.17±5.42) Frequency
treatment) and placebo Psychopharm.
Lee et al. symptoms), followed DSM-IV- BDI-II (NF: 36.67±14.79; C: non-specified
psychotherapy No Medication (12/12) No No 12 to 24 No
(2019) by ↓ alpha/theta ration TR 25.83±7.99) and
(supportive Psychoterphy (NR)
in Pz CGI-S (NF: 4.75±0.62; C: audiovisual
psychotherapy)
(Ninit=12 / 4.17±0.83)
(Ninit=12 /
Nfinal=12)
Nfinal=12)
↑ alpha asymmetry in
Standard care BDI-II (NFa: 30.25+-8.39;
F3 and F4
(continued NFb: 29.17+-11.47; C:
(Ninit=30 / Psychopharm.
Wang et al. pharmacological 30.44+-9.31
Nfinal=24), Yes Medication (47/23)
(2019) medication) BAI (NFa: 21.33+-12.22;
or ↓ beta in P3 and P4 Psychoterphy (NR)
(Ninit=31 / NFb: 21.52+-9.62; C: 22.04+-
(Ninit=26 /
Nfinal=23) 10.32) Continuous
Nfinal=23)
DSM-IV No No 10 No and
Chen and Lin
audiovisual
(2020)
(sample ↓ beta in P3 and P4 Psychopharm.
See “NFb” values in Wang et
partially (Ninit=26 / No Yes Medication (23)
al. (2019)
overlapped Nfinal=23) Psychotherapy (NR)
with Wang et
al. (2019))
fMRI

17
 ↑ areas involved in Patients performing
positive emotions mental imagery of HDRS-17 (NF: 14.38±NR; C:
Psychopharm. Continuous
Linden et al. during mental imagery positive emotions 13.88±NR)
DSM-IV No Medication (8/8) No No 4 No and
(2012) of positive emotions outside the scanner HDRS-21 (NF: 18.12±NR; C:
Psychotherapy (NR) visual
(Ninit=8 / (Ninit=8 / 17.75±NR)
Nfinal=8) Nfinal=8)
Young et al. HDRS-21 (NF: 19.9±5.15; C:
(2014) ↑ of amygdala during Patients receiving 23.9±5.49)
(samples affective memory feedback from non- MADRS (NF: 27.1±6.69; C:
partially recall related control region Yes 31.4±6.71) No
overlapped (Ninit=14 / (Ninit=7 / HARS (NF: 19.1±5.32; C:
with Yuan et Nfinal=13) Nfinal=6) 23.3±7.74)
al. (2014)) STAI Psychopharm. Yes
Patients receiving Medication (0/0) (double- 1
feedback from non- HDRS-21 (NF: 20.64±4.63; Psychotherapy (NR) blind)
↑ of amygdala during
related control region C: 23.69±4.96; H:
affective memory
Yuan et al. (Ninit=13 / 23.69±4.96) Yes (0.3 to
recall Yes
(2014) Nfinal=13), HARS (NF: 19.93±5.15; C: 4) Continuous
(Ninit=14 / DSM-IV-
and healthy subjects 22.15±7.02; H: 1.31±2.02) No and
Nfinal=14) TR
(Ninit=27 / MADRS visual
Nfinal=27)
Not clear
(new
Zotev et al. HDRS-21 (NF: 20.5±4.0; C:
session in
(2016) ↑ of left amygdala Patients receiving 20.9±3.3)
Not clear (new addition
(samples during affective feedback from non- MADRS (NF: 27.4±6.8; C:
session in addition to to the one
partially memory recall related control region Yes 28.5±3.0) 2 No
the one reported by reported
overlapped (Ninit=14 / (Ninit=13 / HARS (NF: 17.5±4.7; C:
Young et al. (2014)) by
with Yuan et Nfinal=13) Nfinal=11) 19.3±5.2)
Young et
al. (2014)) STAI
al.
(2014))
↓ reactivity of a node Patients receiving
of the salience yoked feedback from Psychopharm. Yes
Hamilton et BDI-II (NF: 33.3±2.3; C: Visual at the
network NF group DSM-IV Yes Medication (6/4) No (double- 1 No
al. (2016) 34.6±4.0) end of the trial
(Ninit=12 / (Ninit=10 / Psychotherapy (NR) blind)
Nfinal=10) Nfinal=10)

Young et al. BDI-II (NF: 27.2±10.7; C:

(2017b) 26.6±13.4)
↑ of amygdala during Patients receiving SHAPS
Yes
affective memory feedback from non- MADRS (NF: 23.5±9.9; C: Psychopharm. Continuous
Young et al. DSM-IV- (double-
recall related control region Yes 23.8±6.7) Medication (0/0) Yes 2 Yes (1) and
(2017a) TR blind)
(Ninit=19 / (Ninit=17 / HDRS-21 (NF: 19.4±7.9; C: Psychotherapy (NR) visual
(samples Nfinal=18) Nfinal=15) 19.1±4.4)
partially HARS (NF: 18.8±7.5; C:
overlapped 18.1±6.3)
with Young et
al. (2017b))

18
 Young et al.
(2018a)
(samples
partially
overlapped
with Young et
al. (2017b))
Patients ↑ areas
↑ of areas involved in HDRS-17 (NF: 19.88±-3.65;
involved in scene
positive emotions C: 19.09±5.09)
processing during Psychopharm. Yes Continuous
Mehler et al. during mental imagery HADS-A (NF:
mental imagery of DSM-IV No Medication (16/16) Yes (single- 5 Yes (6) and
(2018) of positive emotions 12.69±3.84; C: 12.63±4.13)
scenes Psychotherapy (0/0) blind) visual
(Ninit=21 / HADS-D (NF:
(Ninit=22 /
Nfinal=16) 13.06±3.43; C: 12.44±4.35)
Nfinal=16)
↑ correlation between
the right superior
anterior Cognitive reappraisal
BDI-II (29.14±8.66)^
temporal lobe and the techniques outside the Psychopharm. Yes Continuous
Jaeckle et al. MADRS (22.84±6.97)^
posterior subgenual scanner DSM-V Yes Medication (10/10) Yes (single- 3 No and
(2019) QIDS-SR16 (16.79±6.53)^
cortex during affective (Ninit=21 / Psychotherapy (0/0) blind) visual
^no details per group reported
memory recall Nfinal=16)
(Ninit=22 /
Nfinal=19)
↑ of fMRI (left ACC HDRS-21 (NF: 14.4±7.0; C:
Patients receiving
and Amygdala) and 15.1±4.9)
feedback unrelated to
EEG (alpha and beta MADRS (NF: 19.6±10.7; C:
brain activity Psychopharm. Yes
Zotev et al. asymmetry in F3 and DSM-IV- 20.5±5.7) Continuous
(artificially generated NR Medication (0/0) No (single- 1 No
(2019) F4) during affective TR HARS (NF: 13.2±7.5; C: and visual
signals) Psychotherapy (NR) blind)
memory recall 16.1±6.4)
(Ninit=8 /
(Ninit=16 / STAI (NF: 56.9±9.9; C:
Nfinal=8)
Nfinal=16) 59.6±9.6)

19
Table 3 – Overview of main clinical outcomes from studies using neurofeedback protocols in depressive patients. ↑ = increased; ↓ = reduced; * = statistically
significant effect; 1 = differences between the post-experiment measurement in each group and the pre-experiment measurement in the merged sample. Studies
with overlapping samples are highlighted in gray. The primary outcome (if declared) is highlighted in bold.
Abbreviations: BAI = Beck Anxiety Inventory; BDI-II = Beck Depression Inventory - Version 2; CGI = Clinical Global Impression; HDRS = Hamilton
Depression Rating Scale; HARS = Hamilton Anxiety Rating Scale; HADS-A = Hospital Anxiety and Depression Scale (Anxiety Subscale); HADS-D = Hospital
Anxiety and Depression Scale (Depression Subscale); MADRS = Montgomery-Asberg Depression Rating Scale; MMPI = Minnesota Multiphasic Personality
Inventory; QIDS-SR16 = Quick Inventory of Depressive Symptoms - Self-Report Version
Clinical Improvement
Post-NF (compared to baseline) Follow up (compared to baseline) Drop-outs or
Studies
exclusions
Within groups Between groups Within groups Between groups (NF/C)
EEG
Schneider et al. (1992) Not reported Not reported Not applicable Not applicable Not reported
NF:
Hammond (2005) Time x group interaction* Not reported Not reported Unmotivated (1/0)
↓ MMPI-2 (30%)
NF:
Paquette et al. (2009) ↓ BDI-II (72.9%)* Not applicable Not reported Not applicable Tiredness (3)
↓ BAI (58.9%)*

NF:
↓ HDRS-17 (64.0%)*
↓ BDI-II (60.1%)* NF > Controls:
Choi et al. (2011) ↓ BDI-II* Not reported Not reported Logistics (0/1)
Controls: ↓ HDRS-17*
↓ HDRS-17 (10.4%)
↓ BDI-II (18.75%)
NF: NF:
Walker and Lawson (2013) Not applicable Not applicable Not reported
↓QIDS-SR16 (44.1%)* ↓QIDS-SR16 (55.1%)*
NF:
Peeters et al. (2014) Not applicable Not applicable Not applicable Not reported
↓ QIDS-SR16 (29.5%)*
Inability to
perform the
cognitive
Escolano et al. (2014) Not reported Not reported Not applicable Not applicable
assessments (4/1)
Excessive noise
(6/3)
NF:
Ramirez et al. (2015) Not applicable Not applicable Not applicable Illness (4)
↓ BDI (17.2%)
NF: Adverse events of
↓ HDRS-17 (70.9%)* medication (1)
↓ HARS (69.1%)* Tiredness (1)
Cheon et al. (2016) Not applicable Not applicable Not applicable
↓ BDI-II (42.0%)* Logistics (1)
↓ BAI (41.1%)* Lost to follow up
↓ CGI (49.1%)* (2)

20
 NF:
↓ BAI (21.74%) and BDI-II
(18.18%) for responders
↑ BAI (124.27%) and BDI-II
(24.97%)* for non-responders
Wang et al. (2016) No significant effects Not applicable Not applicable Not reported
Controls:
↓ BAI (28.33%) and BDI-II
(27.63% ) for responders
↑ BAI (11.54%) for non-
responders

NF:
↓ HDRS-17 (61.65%)*
↓ BDI-II (53.64%)*
NF>Controls:
↓ CGI (38.53%)*
↓ HDRS-17*
Lee et al. (2019) Not applicable Not applicable Not reported
↓ BDI-II*
Controls:
↓ CGI*
↓ HDRS-17 (10.06%)*
↓ BDI-II (8.36%)
↓ CGI (0.00%)
NFa:
↓ BDI-II (34.45%)*
↓ BAI (38.28%)* NFa>Controls:
Group x session interaction for
Non-specified
NFb: BAI*
reason for
Wang et al. (2019) ↓ BDI-II (38.88%)* Not applicable Not applicable
dropping out after
↓ BAI (43.23%)* NFb>Controls:
allocation (6/3/8)
Group x session interaction for
Controls: BDI-II*
↓ BDI-II (8.74%)
↓ BAI (-0.98%)
Chen and Lin (2020)
See Wang et al.
(sample partially overlapped See Wang et al. (2019) See Wang et al. (2019) See Wang et al. (2019) See Wang et al. (2019)
(2019)
with Wang et al. (2019))
fMRI
NF:
↓ HDRS-17 (28.7%)*
Group x session interaction for
Linden et al. (2012) Not applicable Not applicable Not reported
HDRS17*
Controls:
↑ HDRS-17 (7.2%)
Young et al. (2014)
(samples partially
See Yuan et al. (2014) See Yuan et al. (2014) See Yuan et al. (2014) See Yuan et al. (2014) Tiredness (1/1)
overlapped with Yuan et al.
(2014))

21
 NF: Not clear (new
↓ HDRS-21 (15.6%) control
↓ HARS (18.5%)* participants in
Yuan et al. (2014) Not reported Not reported Not reported addition to the
Controls (MDD): sample reported
↓ HDRS-21 (11.7%) by Young et al.
↓ HARS (18.7%)* (2014))
Zotev et al. (2016)
(samples partially See Yuan et al.
See Yuan et al. (2014) See Yuan et al. (2014) See Yuan et al. (2014) See Yuan et al. (2014)
overlapped with Yuan et al. (2014)
(2014))
Hamilton et al. (2016) Not reported Not reported Not applicable Not applicable No response (2/0)
NF: NF:
↓ MADRS (38.7%)* ↓ MADRS (49.4%)*
↓ BDI-II (32.4%)* ↓ BDI-II (40.8%)*
↓ HDRS-21 (34.0%)* NF >Controls: ↓ HDRS-21 (46.4%)* NF > Controls:
↓ HARS (25.0%)* Group x session interaction for ↓ HARS (34.6%)* ↓ MADRS* Discomfort (1/1)
Young et al. (2017b) MADRS*, ↓ BDI-II* Excessive noise
Controls: BDI-II* Controls: ↓ HDRS-21* (0/1)
↓ MADRS (5.0%) and HDRS-21* ↓ MADRS (8.0%) ↓ HARS
↓ BDI-II (4.9%) ↓ BDI-II (8.6%)
↓ HDRS-21 (10.0%) ↓ HDRS-21 (9.9%)
↓ HARS (7.18%) ↓ HARS (23.2%)*
Young et al. (2017a)
(samples partially See Young et al.
See Young et al. (2017b) See Young et al. (2017b) See Young et al. (2017b) See Young et al. (2017b)
overlapped with Young et al. (2017b)
(2017b))
Young et al. (2018a)
(samples partially See Young et al.
See Young et al. (2017b) See Young et al. (2017b) See Young et al. (2017b) See Young et al. (2017b)
overlapped with Young et al. (2017b)
(2017b))
NF: NF:
↓ HDRS-17 (42.0%)* ↓ HDRS-17 (48.5%)*
↓ HADS-A (14.0%) ↓ HADS-A (30%) Personal reasons
↓ HADS-D (23%) ↓ HADS-D (35%) (4/6)
No significant Group x session
Mehler et al. (2018) No significant effects Discomfort (1/0)
interaction for HDRS-17
Controls: Controls: Lost to follow up
↓ HDRS-17 (43.7%)* ↓ HDRS-17 (60.4)* (3/1)
↓ HADS-A (25%) ↓ HADS-A (39%)
↓ HADS-D (31%) ↓ HADS-D (34%)

22
 NF:
↓ BDI-II (46.2%)*1 Feeling unwell to
↓ MADRS (37.1%)*1 continue (1/2)
↓ QIDS-SR16 (39.5%)*1 Logistics (1/2)
No significant Group x session
Jaeckle et al. (2019) Not applicable Not applicable Adverse effects
interaction for BDI-II
Controls: (insomnia - 1/0)
↓ BDI-II (46.0%)*1 Worsening of
↓ MADRS (31.9%)*1 symptoms (0/1)
↓ QIDS-SR16 (35.2)*1

Zotev et al. (2019) Not reported Not reported Not applicable Not applicable Not reported

23
Table 4 – Overview of other significant outcomes from studies using neurofeedback protocols in depressive patients. ↑ = increased; ↓ = reduced; + = positive; - =
negative. Studies with overlapping samples are highlighted in gray.
Abbreviations: EQ-5D-5L = 5-level version of European Quality of Life Questionnaire 5-Dimensional Classification; BDI-II = Beck Depression Inventory -
Version 2; HDRS = Hamilton Depression Rating Scale; HN-NN = happy/neutral - neutral/neutral faces; IAPS = International Affective Picture System; MADRS
= Montgomery-Asberg Depression Rating Scale; PANAS-NA = Positive Affect Negative Affect Schedule - Negative Affect; POMS = Profile of Mood States;
QIDS-SR16 = Quick Inventory of Depressive Symptoms - Self-Report Version; SN-NN = sad/neutral - neutral/neutral faces; SCP = slow cortical potentials; SDS
= Sheehan Disability Scale; SHAPS = Snaith–Hamilton Pleasure Scale; SRET = self-referent encoding task; STAI = State/Trait Anxiety Scale; TAS = Toronto
Alexithymia Scale; VAS = Visual Analog Scale.
Other significant outcomes
Studies Post-NF (compared to baseline) Follow up (compared to baseline)
Within groups Between groups Within groups Between groups
EEG
NF:
- corr.: SCP control and onset of
NF > Controls:
Schneider et al. (1992) illness Not applicable Not applicable
Control of SCP
+ corr.: SCP control and number of
hospitalizations
Hammond (2005) Not Reported Not applicable Not reported Not applicable
NF:
NF:
↓ frequency of worries, frequency of
↓ beta activity in orbitofrontal cortex,
negative automatic thoughts,
insula, amygdala, temporal pole and
frequency of rumination sadness,
Paquette et al. (2009) Not applicable cingulate cortex Not applicable
dysfunctional attitudes, behavioral
+ corr. (uncorrected): ↓ BDI-II and ↓
inhibition
beta activity in orbitofrontal and
↑ frequency of positive automatic
cingulate cortices
thoughts
NF:
↑ accuracy in the verbal fluency task
↓ reaction time for congruent and NF:
incongruent stimuli in the Stroop task Sustained clinical, physiological, and
Choi et al. (2011) Significant time x group interactions Not reported
↑ of alpha asymmetry neuropsychological improvements
(values not reported)
Controls:
No significant effects
Walker and Lawson
Not reported Not applicable Not reported Not applicable
(2013)
NF:
Peeters et al. (2014) - corr.: ↓ QIDS-SR16 and alpha Not applicable Not applicable Not applicable
asymmetry
NF:
↓ number of errors and reaction time
↑ power in the working memory task
No significant effects in alpha Group x time interaction for the working
Escolano et al. (2014) Not applicable Not applicable
asymmetry memory task

Controls:
No significant effect

24
Ramirez et al. (2015) Not reported Not applicable Not applicable Not applicable
No significant effects in alpha
Cheon et al. (2016) Not applicable Not applicable Not applicable
asymmetry
Wang et al. (2016) No significant effects No significant effects Not applicable Not applicable
NF: NF>Controls:
↑ EQ-5D-5L and ↓ SDS ↑ EQ-5D-5L
Lee et al. (2019) Not applicable Not applicable
Controls: Controls>NF:
↓ EQ-5D-5L and ↑ SDS ↑ SDS
NFb:
↓ P3 high-beta power
Wang et al. (2019) No significant effects Not applicable Not applicable
Controls:
↑ P3 high-beta power
NF:
Chen and Lin (2020)
↓ beta, but not other bands, in P3 and
(sample partially
P4 Not applicable Not applicable Not applicable
overlapped with Wang
+ corr. between ↓ BDI-II and ↓ beta in
et al. (2019))
P3 and P4
fMRI
NF:
↓ POMS
↑ bilateral ventral striatum and left
extra-striate visual cortex activity Controls > NF:
Linden et al. (2012) Not applicable Not applicable
+ corr.: up-regulation and HDRS PANAS-NA

Controls:
↓ POMS
NF:
↓ STAI trait and state anxiety
↓ POMS-depression and anger
NF > Controls:
↓ VAS restlessness, anxiety and
Young et al. (2014) VAS-happiness
irritability
(samples partially Amygdala activity
↑ VAS-happiness Not applicable Not applicable
overlapped with Yuan
↑ left amygdala activity
et al. (2014)) 'Controls > NF:
+ linear trend across all runs
STAI state anxiety
Controls:
↓ VAS-sadness
Controls (healthy) > NF and Controls
NF: NF:
(MDD):
Yuan et al. (2014) - corr.: amygdala-cuneus connectivity + corr.: amygdala-cuneus connectivity Not reported
Amygdala-ACC and amygdala-cuneus
and HDRS and the time to follow-up
connectivity before NF, but not after

25
 NF:
↓ POMS depression, total mood
disturbance
↑ VAS happiness
+ corr.: amygdala activity and self-
reported happiness, and memory-
Zotev et al. (2016)
recall, and VAS-happiness
(samples partially
- correlation: amygdala activity and Not reported Not applicable Not applicable
overlapped with Yuan
POMS-tension, and TAS-total
et al. (2014))
+ corr.: amygdala laterality and TAS-
total
+ corr.: EEG asymmetry and HDRS
and SHAPS-anhedonia
+ corr.: EEG asymmetry and
amygdala laterality
NF:
↓ salience network node response
↓ emotional reactivity to negative
NF > Controls:
IAPS
Hamilton et al. (2016) Reduction in responses to IAPS negative Not applicable Not applicable
↓ in negative SRET
pictures, and in negative SRET
Controls:
No significant effects
NF:
↓ SHAPS
↑ recall of positive specific and overall NF > Controls:
specific memories Recall of specific extended memories, and
↓ recall of categorical positive, overall positive specific memories NF:
categorical, extended positive, ↓ SHAPS*
Young et al. (2017b) extended negative, and overall Controls > NF: No significant effects
extended memories SHAPS Controls:
+ corr.: MADRS and amygdala Recall of categorical extended memories, No significant effects
activity during the final transfer run and positive categorical and extended
memories
Controls:
No significant effects
NF:
NF > Controls:
↓ amygdala activity during n response
Amygdala activity during HN-NN
to sad faces
condition
↑ amygdala activity during n response
Vigilance to positive faces
Young et al. (2017a) to happy faces
(samples partially ↓ reaction time for positive faces
Not reported Not reported Controls > NF:
overlapped with ↓ reaction time for positive words
Amygdala activity during SN-NN
Young et al. (2017b)) ↑ vigilance to positive faces
condition
↓ vigilance to negative faces
Reaction time for positive faces
Reaction time for positive words
Controls:
Vigilance to negative faces
No significant effects

26
 NF > Controls:
Amygdala connectivity with prefrontal
cortical, striatal and subcortical
Young et al. (2018a) regions during memory recall, and limbic
(samples partially regions at rest
Not reported Not reported Not reported
overlapped with
Young et al. (2017b)) Controls > NF:
Amygdala connectivity with right
temporal lobe during positive memory
recall, and bilateral temporal pole at rest
NF :
↑ ROIs across sessions

Controls:
↑ ROIs across sessions
Mehler et al. (2018) No significant effects No significant effects No significant effects
Both groups:
+ corr.: between HDRS-17
improvement (corrected for
confounds) and improvement in self-
efficacy scores
NF:
↓ connectivity between the right
superior anterior
temporal lobe and the posterior
subgenual cortex

Both groups:
Jaeckle et al. (2019) No significant effects Not applicable Not applicable
↓ POMS-depression dejection,
Rosemberg self-esteem scale
↓ self-blame ratings during anger
content
↑ self-esteem ratings
- corr.: between differences in self-
esteem ratings and BDI reduction
NF:
↓ POMS depression, confusion, and
total mood disturbance
↑ VAS happiness
↑ alpha and beta asymmetry, and left
amygdala activity during NF
↑ left amygdala-ACC connectivity NF > Controls:
Zotev et al. (2019) Not applicable Not applicable
+ corr. between alpha asymmetry and left amygdala-ACC connectivity
MADRS-trait depression, and
SHAPS-anhedonia
- corr. between alpha asymmetry and
delta POMS-state depression and
POMS-total mood disturbance

27
 Controls:
No significant effects

28
3.2. fMRI neurofeedback paradigms and clinical effects

fMRI-based neurofeedback protocols commonly employ tasks of emotional self-regulation and train patients to either up- or down-
regulating the BOLD signal in brain regions related to emotion processing, such as the amygdala, dorsal anterior cingulate cortex,
prefrontal cortex, insular cortex, superior temporal gyrus, precentral gyrus, and middle temporal gyrus (Johnston et al., 2010; Linhartová
et al., 2019).
In a first proof-of-concept non-randomized, non-blinded study of fMRI neurofeedback in depression, Linden et al. (2012)
compared two groups of eight medicated MDD patients. While the control group engaged in mere mental imagery training outside the
scanner, the experimental group received four sessions of neurofeedback training, during which they used similar mental strategies to
self-regulate the activity in brain areas responsive to affective visual stimulation. Specifically, to identify responsive ROIs for the
algorithm, all volunteers were initially submitted to affectively charged figures with positive valence. Throughout the sessions, patients
in the neurofeedback group learned to up-regulate the BOLD response of the targeted areas, including the ventrolateral and dorsolateral
portions of the prefrontal cortex, insula, medial temporal lobe, and orbitofrontal cortex.
Further, the neurofeedback group, but not the control group, presented a significantly larger improvement in depressive
symptoms (approximately 28% of improvement and 7% of worsening, respectively) and two patients in the experimental but no patient
in the control group were remitted at the primary endpoint (Linden et al., 2012). In a subsequent larger randomized single-blind
controlled trial (N=16 per group) by Mehler et al. (2018), medicated patients were assigned to one of two neurofeedback training
interventions: whereas the experimental group trained over five sessions to activate limbic areas using positive mental imagery (NFE)
similar to patients in the neurofeedback group in Linden et al. (2012), the active control group trained over five sessions to activate
higher visual areas imagined relaxing scenes (NFS). Training areas in the NFS control group included regions involved in scene
processing, such as the parahippocampal place area and higher visual cortices (Mehler et al., 2018). Although the NFE group was
expected to show superior clinical improvements, no statistically significant group difference was found at the primary endpoint or a
follow up (6 weeks later). However, patients in both groups showed substantial reductions on the HDRS-17 (about 42% and 44% for
the NFE and NFS group, respectively), which lasted and improved slightly further at follow-up six weeks later (about 48% and 59% for
the NFE and NFS group, respectively).
Moreover, about 38% (12/32) of patients were remitted (based on the HDRS-17 score) at the primary endpoint (with 4/16
patients, i.e., 25% in the NFE, and 8/16, i.e., 50% in the NFS group, respectively). Potential reasons that may account for these findings
include that both groups engaged in a potentially beneficial form of mental imagery. Further, post-experimental analyses showed that
both groups presented overlapping active voxels in the anterior insula during the neurofeedback training. Of interest, a correlation
between clinical improvement and a measure of self-efficacy was reported suggesting that the successful training experience may already
provide clinical benefit to patients.
Another set of six studies reported clinical results and exploratory analyses from two independently conducted neurofeedback
experiments (Young et al., 2017b; Young et al., 2014) in which participants trained self-regulation of amygdala activity. In the first

29
study by Young and colleagues (Young et al., 2014), unmedicated MDD patients enrolled in a non-randomized single-blinded, sham-
controlled experiment, unmedicated MDD patients trained to self-regulate the amygdala using positive autobiographical memories
(N=13) in an experiment in which the control group (N=6) received feedback from a brain area (the intraparietal sulcus) that was not
associated with the mental task (Young et al., 2014). After a single session, the first group achieved effective control of the amygdala
responsiveness. Psychometric testing suggested a reduction of anxiety indexes and increased happiness indexes (Young et al., 2014),
but results for clinical effects were not reported albeit the HDRS-21 was assessed at baseline.
In a follow-up study the same research group included this initial data set and tested a few more patients in the patient control
group (new N=13 patients) as well as an additional control group of healthy participants (N=27) (Yuan et al., 2014). Their results
suggested slight decreases on the HDRS-21 in all groups (about 16% of improvement for the experimental group, and 12% for the
patient control group) but no significant group difference. Of interest, post-hoc analyses indicated increased resting-state functional
connectivity between the left amygdala and the left pregenual anterior cingulate cortex (pgACC), and between amygdala and the left
cuneus in both groups following neurofeedback training (Yuan et al., 2014). Finally, upregulation of the left amygdala BOLD activity
during a new (second) session of the same protocol was accompanied by positive average changes in frontal alpha EEG asymmetry,
which significantly correlated with the MDD patients’ trait depression severity (Zotev et al., 2016).
In a subsequent, larger randomized, double-blinded clinical trial, the same research group compared the clinical effects over two
training sessions using a similar training protocol. The authors reported a significant group by session interaction and follow up-analyses
suggested that only the experimental group (N=18) that trained amygdala up-regulation showed improved depressive symptoms (about
39% reduction of the Montgomery-Åsberg Depression Rating Scale - MADRS - at the primary endpoint). Of interest, about 32% (6/19
patients) were remitted (based on the MADRS score) at the primary endpoint (Young et al., 2017b). In contrast, mean scores in the
control group (N=15) remained nearly unchanged (about 5%), and only one patient showed remission. Further, compared to the control
group, the neurofeedback group presented higher hemodynamic and behavioral responses for positive visual stimuli, lower responses
for negative stimuli (Young et al., 2017a). In a follow-up analysis the authors further reported functional connectivity changes between
the amygdala and areas of the frontal and limbic network that correlated with the previously reported clinical improvement (Young et
al., 2018a).
In a subsequent non-randomized single-blind controlled study this neurofeedback paradigm was expanded by Zotev et al. (2019)
to a multimodal, single-blinded, single-session training protocol combining fMRI based self-regulation training of the left amygdala and
left rostral anterior cingulate cortex (rACC), as well as EEG based training of asymmetry in the alpha and beta band. Unmedicated
patients were assigned either to the experimental group (N=16) that received veridical feedback or a control group (N=8) that received
randomly generated feedback signals that were unrelated to their brain activity. The results suggested that the experimental group showed
increased activity in the left amygdala, EEG asymmetries, as well as enhanced functional connectivity between the left amygdala and
the left rACC (Zotev et al., 2019). However, no information about clinical effects was reported in this study.
Hamilton et al. (2016) introduced another paradigm that employed functional connectivity based neurofeedback in which they
investigated the ability of partly medicated MDD patients to down-regulate nodes from the salience network in the presence of negative

30
stimuli. Twenty patients were presented to pictures taken from the IAPS (International Affective Picture System) that were associated
with negative valence to identify nodes involved in processing negative affect. The authors then allocated patients either to an
experimental group that received veridical (N=10), or a control group (N=10) that received a form of sham neurofeedback training
where participants are provided with the replay of visual feedback from the experimental group to control (yoked feedback). When re-
exposed to negative visual stimulation in a post test, only the neurofeedback group but not the control group showed reduced responses
in neural nodes from the salience network. Moreover, there was a trend for lower scores for a self-reported responses to negative images
(Hamilton et al., 2016). However, the study only assessed within group changes but no between group comparison, and clinical effects
were not reported.
More recently, Jaeckle et al. (2019) conducted a randomized, single-blinded trial that consisted of three training sessions. Patients
(majority of them under medication) were either allocated to the experimental group (N=19) that trained up-regulation of functional
connectivity between the right superior anterior temporal lobe and the right subgenual cingulate, or to a control group (N=16) that trained
cognitive reappraisal techniques outside the fMRI scanner. Results suggested that both groups showed significant symptom
improvement in the BDI scale (approximately 46% and 37%, respectively), but no significant difference between groups was found.
Lastly, we note that some studies were not included in our primary analyses due to inclusion and exclusion criteria (Table 1); they are
briefly summarized in the Supplementary Material (Section 4).

3.3. Clinical efficacy for different control condition categories

Both EEG and fMRI neurofeedback studies in MDD thus far published are heterogeneous with regards to some key design aspects of
clinical studies. Albeit several studies employ similar training paradigms (e.g., alpha asymmetry EEG neurofeedback training of frontal
electrodes, or self-regulation fMRI neurofeedback training of limbic areas), they vary substantially with regards to features such as
randomization, blinding and control conditions. For instance, only six studies (2 applying EEG neurofeedback and 4 applying fMRI
neurofeedback protocols) randomized patients to either an experimental or a control arm. Moreover, only four studies (all fMRI) used
double-blinding, while the other fMRI neurofeedback studies (except for the first feasibility study) were single-blinded. Noteworthy,
none of the EEG neurofeedback studies were single- or double-blinded, while all but one fMRI neurofeedback studies were at least
single-blinded (Table 2). Whether neurofeedback studies allow single or double-blinded assessment depends largely on the choice of
the control condition (Sorger et al., 2019): while some designs (e.g. yoked feedback) allow blinding patients (Young et al., 2017b), other
active control conditions that are based on different instructions and veridical feedback do not (Mehler et al., 2018).
Normalized baseline scores of depressive symptoms (see section 2.2) were largely comparable between studies as well as control
conditions (where applicable). With the exception of five studies ((Linden et al., 2012), (Walker and Lawson, 2013), (Ramirez et al.,
2015), (Wang et al., 2016), and (Zotev et al., 2019)), patients were on average moderately to severely depressed and experimental and
control groups were on average largely matched for their depression severity at baseline (group differences were mostly under 10%,

31
with only one study (Wang et al., 2016) showing a difference of 11.55%; see Figures 2A-B). However, most studies did not provide
sufficient clinical information regarding prior treatment experience, treatment resistance, duration of illness, number of episodes and
hospitalizations of patients (see Table S1 in the Supplementary Material). We also note that the EEG-NF field has either employed only
single-arm studies (58%, in particular early studies) or passive control conditions, whereas the fMRI-NF field has exclusively employed
active control conditions (71% inside the scanner and 29% outside the scanner).
Figures 2C-D and 2E-F show the symptom improvement per group and the difference of improvement between groups,
respectively. In general, all groups presented some level of symptom improvement, with the exception of one study in which the control
group presented 7.20% of mean symptom worsening (Linden et al., 2012). Regarding differences across groups, in seven studies the
experimental group showed higher improvement than the control group, while in two studies the effect was in the opposite direction
(Mehler et al., 2018; Wang et al., 2016). Moreover, group differences tended to be larger for studies that used a passive control groups
compared to studies with active control groups, which found relatively small group differences (Figures 2E-F). This exploratory finding
is in line with the notion that non-specific psychosocial effects are additive, and they confirm previous theoretical considerations (Ros
et al., 2020; Sorger et al., 2019; Thibault et al., 2016).
Another factor that may influence differences in clinical findings across studies are the number of training sessions. Given that
the clinical samples listed in Figure 2 followed similar intervention schedules (i.e., 13/15 studies with one to two sessions a week, 1/15
with three sessions a week, and 1/15 not reported) we explored the data for potential dose-effect relationships. In an exploratory plot we
show the relation of symptom improvement in the experimental group vs. the number of sessions (Figure 3A). We found that EEG
neurofeedback studies consisted of average of more training sessions (14.63, compared to 2.43 from fMRI studies), likely due to the
substantially lower operating cost. Further, data suggested a positive linear trend for number of training sessions versus symptom
improvement in the experimental group. We also looked at the relationship between number of training sessions and reported group
differences for controlled studies (Figure 3B), which also suggested a positive trend. Additionally, a positive trend was observed for an
association between the number of training sessions and depressive symptoms at baseline in the experimental group (Figure 3C)
suggesting that more severely depressed patients were included in studies that provided patients with more neurofeedback sessions.
However, a nearly horizontal line was observed for the relation between symptoms at baseline and clinical improvement for experimental
groups (Figure 3D), suggesting no linear association between symptom severity at baseline and the primary endpoint. These data
explorations should be treated with caution, however, given the heterogeneity in study designs with regards to control conditions.

32
33
Figure 2 - (A) Depressive symptoms in percentage scores at baseline, normalized by the individual scale maxima. (B) Sample size weighted average percentages
of depression severity at baseline after grouping studies according to their control condition category. (C) Percentage of within-group improvement in depressive
symptoms at the primary endpoint. (D) Sample size weighted average percentages of within group improvements in depressive symptoms, after grouping studies
according to by their control condition category. (E) Between group differences in improvement. (F) Sample size weighted average percentages of within group
improvements in depressive symptoms after grouping studies according to their control condition category. Studies evaluating the same database were
represented as a single bar. *Wang et al. (2019) is represented by two bars since they tested two different neurofeedback protocols in one single experiment
(please refer to Table 2).

Figure 3 – (A) Trend of symptom improvement in experimental group and number of training sessions; neuroimaging modality indicated. (B) Trend of
differences in symptom improvement between groups and number of training sessions, control condition category indicated. Studies evaluating the same

34
 database were represented as a single point. (C) Trend of symptom severity at baseline and number of sessions. (D) Trend of symptom improvement in the
experimental group and respective symptom severity at baseline. *These studies are represented by two points since they tested two different neurofeedback
protocols in one single experiment (please refer to Table 2).

Average NNTB/NNTHs based on reported remission rates ranged between -5.78 and 4 and were mostly positive (Table 5),
suggesting the experimental groups showed higher efficacy with respect to remission from depression. Noteworthy, only one (unblinded,
non-randomized) study (Lee et al., 2019) could rule out potential superiority of the control condition, which consisted of continued
standard care (psychopharmacological medication), over the experimental condition, which consisted of continued standard care
augmented by EEG neurofeedback training (Table 5). In contrast, the upper bounds of 3 trials were negative and they could hence not
reject the null hypothesis that patients in the control arm showed a better clinical outcome compared to patients in the main treatment
arm (Altman, 1998). One main reason for this finding is likely the relatively small sample sizes of studies that could not exclude potential
superiority of the control group. For instance, although Young et al. (2017b) found a remarkable difference in remission between the
experimental and the control group, the upper bound of the 95% confidence interval was -109.44; this negative values indicates that it
remains possible that about 1 in 109 patients who are allocated to the experimental group will show less improvement compared to the
control group.

Table 5 – Numbers of remitters in experimental and control group and their percentages with respect to total number enrolled patients (dropouts were treated as
non-responders), number needed to treat for one additional patient to benefit (or to be harmed) [NNTB/NNTH] and their respective 95% confidence interval (CI).
Negative NNTB/NNTH or CIs indicate that patients in the control may have shown a better outcome.
Experimental group Control group Lower Upper
Study Remitters / Remission rate Remitters / Remission rate NNTB/NNTH 95% 95%
total sample (%) total sample (%) CI CI
Linden et al. (2012) 2/8 25.00 0/8 0.00 4.00 1.69 -8.31
Young et al. (2017b) 6/19 31.58 1/17 5.88 3.89 2.06 -109.44
Mehler et al. (2018) 4/21 19.05 8/22 36.36 -5.78 10.60 -2.44
Lee et al. (2019) 6/12 50.00 1/12 8.33 2.40 1.49 19.66

3.4. Reported side effects and drop-outs

Side effects are rarely reported for neurofeedback interventions (Table 3, last column). This observation may be explained by the non-
invasive nature of the intervention, but partly also related to reporting practices (see Section 3.5). In general, we note that one limiting
factor for the wide usage of clinical neurofeedback may be physical discomfort experienced before and during each session, respectively.
For example, EEG protocols require a relatively long time for the EEG cap preparation (positioning, conductive gel, calibration) (Nijholt
et al., 2011). It also results in residual gel over the participant’s head after the session. During fMRI protocols, on the other hand, the

35
patient may experience claustrophobia due to the physical restriction imposed by the equipment (Sulzer et al., 2013). These aspects are
particularly relevant to MDD patients because their symptoms can include diminished interest, sleeping problems, psychomotor
agitation, and fatigue or loss of energy (Association, 2013). We documented reported reasons for drop-outs or exclusions (Table 2),
which included lack of motivation (Hammond, 2005), tiredness (Cheon et al., 2016; Paquette et al., 2009; Young et al., 2014), discomfort
(Young et al., 2017a), logistics difficulties (Cheon et al., 2016; Choi et al., 2011) and excessive noise (possibly related with the patient’s
agitation) (Escolano et al., 2014; Young et al., 2017b). However, we note that overall drop-out rates were relatively low and no serious
side effects have been reported.

3.5. Experimental design and reporting quality

As noted above, a first overview of study designs (Table 2) suggests that while most neurofeedback studies published thus far employed
control groups, only a minority conducted blinded assessment or randomized patients. We next assessed the quality of experimental
designs and study reporting more systematically employing the JBI critical appraisal tools (Tufanaru et al., 2017) and CRED-nf checklist
(Ros et al., 2020).

36
 Figure 4 – Summary of CRED-nf scores. (A) the average score per CRED-nf item for category using EEG-based protocols (blue), fMRI-based protocols
(orange), and the overall score across modalities (gray). (B) the average score per study using EEG- (blue) and fMRI-based (orange) protocols. Summary of JBI
scores. (C) the trend of quality improvement measured with CRED-nf scores. (D) the average score per JBI item for studies using EEG-based protocols (blue),
fMRI-based protocols (orange), and the overall score across modalities (gray). (E) the average score per study using EEG- (blue) and fMRI-based (orange)

37
 protocols. (C) the trend of quality improvement measured with JBI scores. Studies reporting results from the same database are reported as an averaged single
bar.

As shown in Figure 4, EEG neurofeedback studies received on average lower scores in all CRED-nf points except for “Outcome
measures” and “Data storage” (the latter was not fulfilled by any study included in this review). Similarly, EEG neurofeedback studies
received on average lower scores for six of nine items of the JBI checklist. However, regarding the items “cause and effect”, “outcome
reliability items”, both imaging methods presented full scores across studies, while for “multiple measurements”, EEG neurofeedback
studies tended to score higher on average. Both the CRED-nf and JBI checklist allowed identifying some major limitations in the field
which we discus below. These will inform our recommendations formulated in Section 4.
Regarding the CRED-nf checklist, we first note that only five studies preregistered their experimental protocol.
Complementarily, concerning the experimental design, one main limitation of EEG neurofeedback studies in depression is the lack of
adequate control groups (only present in 38% of studies). A more general limitation, related to both imaging techniques, is the limited
description of the online brain signal processing and artifact control. Although all included studies at least partly report how data is
extracted and preprocessed (step 3), reporting often remained insufficient.
When reviewing studies for reported outcome measures, we found that while some studies defined "success", or "control"
measures explicitly, many studies did not: according to the CRED-nf scores (see Table S3 in the Supplementary Material), only 51% of
studies reported neurofeedback success based on neural signals (33% EEG and 81% fMRI neurofeedback studies), while 54% plotted
within- or between sessions (38% EEG and 81% fMRI studies).
We also note that, only about 26% of studies declared the primary clinical outcome measure (only 17% EEG and 43% fMRI
neurofeedback studies). The distinction between primary and secondary outcome measures is considered a quality standard in clinical
research: it is central to evaluating the clinical efficacy of an intervention (e.g., to estimate remission rates) and to control for error rates
(in contrast to test results for secondary outcome measures, test results for predeclared primary outcome measures usually do not require
correction for multiple testing). Further, only 19% of studies evaluated psychosocial factors before or after the experiment. However,
some EEG neurofeedback studies did not report if specific self-regulation strategies were provided/suggested to patients (58% EEG and
100% fMRI studies providing this information), and only very few studies reported debriefing results and thus could capture the
strategies used (8% using EEG and 19% using fMRI). Lastly, none of the studies stored the resulting (clinical or physiological) data or
analysis code in publicly available domains.

4. Discussion

In this first systematic review of neurofeedback studies across imaging modalities conducted in depressed patients, we found that both
EEG and fMRI studies report statistically significant and clinically meaningful within group improvements of clinical measures between
6% and 73%. In comparison, between group comparisons showed numerically smaller changes ranging from -7% to 52%. These findings
may be explained by differences in used controlled conditions. It is assumed, however, that overall clinical effects following

38
neurofeedback training can be partly or largely attributed to various non-specific factors: patient’s positive expectancies, the rewarding
experience of positive feedback, but also regression to the mean likely contribute substantially to observed within group improvements.
Neurofeedback training is a complex intervention and involves various degrees of freedom in designing control conditions. These
range from passive control designs (e.g. continued standard care vs. continued standard care and neurofeedback augmentation training);
these are expected to provide the least control of non-specific factors, to active control designs (e.g. continued standard care and
neurofeedback augmentation training with vertical from a control region vs. continued standard care and targeted neurofeedback
augmentation); these are expected to provide the most control for non-specific factors (Ros et al., 2020; Sorger et al., 2019; Thibault et
al., 2016). As recently discussed in-depth by the neurofeedback community (Lubianiker et al., 2019; Sorger et al., 2019), the choice of
optimal control conditions poses a challenge for neurofeedback experiments. Control conditions are important to evaluate non-specific
effects and to compute more informative effect sizes such as NNTB that allow comparisons to other therapeutic approaches. To compare
between-group clinical effects across neurofeedback studies, we therefore grouped these according to their control condition. Results
indeed showed that active control conditions presented smaller group differences in favor of neurofeedback compared to more lenient
passive control conditions (Figure 2E). Of interest, these findings are comparable to those reported for EEG neurofeedback training in
ADHD (Cortese et al., 2016; Group et al., 2020; Van Doren et al., 2019).
One main question when designing an intervention is the dose response relationship and the temporal evolution of treatment
effects. Although there is no clear definition of dosage of neurofeedback training, the number of neurofeedback sessions has been
suggested as a potential proxy (Arns et al., 2009; Vernon et al., 2004). Indeed, first visual data exploration suggested a relationship
between the number of neurofeedback training sessions and overall clinical effects in the treatment group (Figure 3A). However, as
noted earlier, many EEG neurofeedback studies did not feature sufficient control conditions. Thus, this apparent finding may be
confounded by other factors such non-specific neurofeedback, general non-specific, repetition related or natural effects (e.g., regression
to the mean). We hence conducted a second exploratory analysis for controlled studies only, which also suggested a positive relationship
between the number of training sessions and group differences (Figure 3B). One possible explanation for this observation is the learning
process of a neurofeedback task: the feedback is initially based on unconditioned neural variability until the patient can learn, correct,
and optimize the self-regulation strategies (Birbaumer et al., 2013; Ros et al., 2014). Moreover, delayed functional effects are commonly
also observed in cognitive therapy: the improvement of cognitive restructuring strategies happens across sessions, and dose-response
effects have been reported across psychiatric disorders (Robinson et al., 2020). Thus, from a neural as well as a cognitive perspective,
an expected interval until the functional effects can be observed. However, while the presented findings are encouraging, we note that
their interpretation is mainly conjectural given the heterogeneity in control conditions used in neurofeedback studies and the possibility
that also this finding may be confounded/driven by non-specific effects that were not controlled for in rather lenient (e.g., passive)
control conditions of some studies. Taken together, reported findings of clinical effects for neurofeedback training seem substantial and
scale with time; however, therapeutic effects specific to neural targets are likely relatively small, and hence future RCTs will require
larger samples to study neurofeedback-specific effects in depression. Further, longer follow-up periods are desirable; clinical effects
following neurofeedback interventions have been documented to last, and partly further improve for up to several months after the last

39
neurofeedback session (Becerra et al., 2006; Gevensleben et al., 2010; Goldway et al., 2019; Mehler et al., 2018; Rance et al., 2018). In
addition, it remains of interest to investigate to which degree observed effects occur within or between training sessions and whether
there is an interaction thereof.
Moving on to comparing EEG and fMRI, substantial differences in designs were found: Whereas most EEG studies lacked
control conditions and were not blinded, recent fMRI studies increasingly fulfill these standards. In general, we found that fMRI studies
tended to fulfill more study design and reporting quality criteria. One possible explanation for this result may be that most studies were
planned and reported more recently compared to EEG-based protocols. They may also have been able to incorporate criticism raised
against previous EEG neurofeedback studies, benefit from methodological advancements, and broader debates around adequate
statistical aspects (Button et al., 2013; Nieuwenhuis et al., 2011). This trend is exemplified in Figures 4C and 4F.
While most EEG neurofeedback studies can be considered (uncontrolled) phase IIa trials that aim to demonstrate feasibility,
most fMRI neurofeedback studies represent (controlled) phase IIb trials that aim to demonstrate clinical efficacy. However, common to
almost all studies are relatively small sizes, which render these statistically underpowered to detect small or medium effects. From RCTs
conducted on the clinical effects of antidepressant medication, for instance, relatively small effects (Cohen’s d = 0.2 to 0.3) are
documented for treatment vs. placebo controls (Cuijpers et al., 2014; Kirsch et al., 2008) To detect an effect size within this range with
80% probability, studies would need to feature at least about 176 patients per group for a two-arm controlled study. At least for fMRI-
based neurofeedback protocols, such scales are likely only achievable in multi center studies. Some further ideas on this matter are listed
in the recommendations section below. To further illustrate the limited power of existing studies, 3 of 4 studies that also reported
remission rates could not rule out superiority of the control group in an NNTH analysis that we conducted (Table 5).
Evaluating reporting practices, most included studies lacked information about several aspects that are considered essential or
highly desirable such as declaring the primary outcome measure, reporting a sampling plan, reporting feedback controllability or
remission rates. Hence, on average studies in the field still bear considerable risk for bias, which restricts generalizations that can be
drawn from reported findings. Moreover, we note that several published studies included partly overlapping samples, which made it
sometimes difficult to assess their quality in a coherent way. Further, such practice indicated that authors may have employed flexible
sampling stopping rules (without adequate adjustment), which risks increasing type-I error rate (see the Recommendations section below
for some suggestions). Most of these aspects could be addressed by comprehensive study preregistrations, including the declaration of
the primary outcome measure, main hypotheses, intended sample size and planned analyses. Originally introduced in clinical medicine
(DeAngelis et al., 2005), study preregistrations can restrict degrees of freedom and avoid sources for researcher bias, including outcome
switching, inadequately used flexible stopping rules and analytical degrees of freedom (Nosek et al., 2018) as well as publication bias
(Allen and Mehler, 2019).
We also note that many studies, and in particular EEG neurofeedback studies, did not report neurofeedback success measures.
A clear definition on success measures allows assessing the proportion of individuals who show relatively poor neurofeedback control,
a phenomenon that has also been labeled as “illiteracy” (Allison and Neuper, 2010), and which likely pertains 10 to 50% of
neurofeedback users (Alkoby et al., 2018; Allison and Neuper, 2010; Edlinger et al., 2015). Estimating the proportion of non-learners,

40
and ideally identifying predictors for self-regulation success, seem in particular important for neurofeedback studies with depressed
patients who tend to process negative experiences (e.g. no self-regulation success) more negatively (Disner et al., 2017; Peckham et al.,
2010).
Further, documenting experiment factors, such as attention from the staff, comfort in the experiment room, or motivation,
measures of confidence, or frustration, and personal believes might help to understand variations in self-regulation performance (Paret
et al., 2019), but also explain observed clinical effects. Constructs such as self-efficacy that are related to the psychopathology in
depression (Bandura, 1982) may be modifiable through self-regulation training (Linden, 2014; Mehler et al., 2018). Ratings also showed
that none of the included studies has shared their imaging and/or clinical data publicly and that only a few studies were preregistered.
While such reservation may be an expression of data protection concerns, we note that data anonymization tools are widely available
and it should be in the best interest of the community to make use of these and follow recent efforts of the neuroimaging community
clinical medicine in tackling issues around reproducibility and replicability (Poldrack and Gorgolewski, 2014).
Finally, it is crucial to use appropriate and robust methods for data extraction and preprocessing. For instance, most EEG and
fMRI studies do not use state-of-the-art artifact control methods (e.g., electro-oculography and electromyography) when calculating the
feedback signal. Similarly, for fMRI-neurofeedback, control for confounding factors such as online correction of head motion, breathing,
and cardiovascular artefacts are often insufficiently reported, although they may have a major impact on reported findings (Weiss et al.,
2020). This finding is in line with earlier findings for fMRI neurofeedback studies more broadly (Heunis et al., 2020; Thibault et al.,
2018).
Overall, our findings indicate that, CRED-NF and JBI checklist ratings suggest that fMRI neurofeedback studies featured on
average better reporting quality. Yet, we note that the CRED-nf guidelines were published only very recently and hence the authors of
the investigated studies could not use neurofeedback specific guidelines as orientation for design and reporting practices. Comparing
JBI ratings reported here with other fields, the present sample featured an average rating of 6.17, which is similar to those reported in
systematic reviews (that included a similar number of studies) conducted about fMRI neurofeedback training in stroke patients (mean
6.24) and non-clinical/clinical fNIRS neurofeedback (mean 5.55) (Kohl et al., 2020; Wang et al., 2018). Also, with regard to essential,
encouraged and total CRED-NF ratings, we found similar results (with 65% vs. 63%, 13% vs 10% and 47% vs. 45%, respectively)
compared to the fNIRS-NF field (Kohl et al., 2020). Lastly, we note that one main limitation of this review was the relatively small
number of studies that could be included, and which precluded employing other established meta-research techniques such as p-curve
analysis (Simonsohn et al., 2014) or funnel plots to test for small study effects (e.g. due to publication bias). Further, the heterogeneity
in study designs that controlled for non-specific effects to different degrees – which ranged from no control to very conservative active
neurofeedback control conditions – rendered an aggregated effect size across studies rather meaningless. We therefore decided to merely
provide estimates of clinical improvement in percentages averages for studies with similar control conditions.

4.2. Recommendations

41
Despite promising first results with patient groups, current neurofeedback protocols present methodological challenges for real-world
therapeutic applications (Arns et al., 2017; Thibault et al., 2016). Heterogeneity of protocols and inconsistent reporting make replication
and standardization difficult. These aspects are crucial not only for the research community to understand and progress the
neurofeedback technology (Thibault et al., 2017), but also for patients, since a poor setup can cause frustration and lead to discontinued
training (Müller-Putz et al., 2015). Thus, in line with the final aim of this review, we provide here an overview of recommendations that
future researchers should adopt for experiments with depressive patients. A more detailed discussion of these with a particular focus on
points a) - c) can be found in the Supplementary Material (Section 6).

Table 6 – Recommendations for future experiments with depressive patients (some of these points are discussed in more detail in Section 6 of the Supplementary
Material).

a) More comprehensive clinical documentation and phenotyping To ensure reliable clinical results and to allow comparison
between studies, we recommend that future neurofeedback
experiments in depressive patients use formal and standardized
procedures to diagnose and evaluate clinical changes with
clinician-rated scales (e.g., HDRS-21, MADRS) and self-rated
scales (e.g., BDI-II or QIDS-SR16). Besides changes in sum scores
of scales, we encourage reporting changes in individual items to
assess changes in specific symptoms or symptom networks and
cluster different types of responses (Fried and Nesse, 2015; Fried
et al., 2017; Hofmann et al., 2016). Further detailed descriptions
of previous antidepressant treatment and patients’ duration of
illness should be provided to allow to assess the level of chronicity
and treatment resistance of included patients, factors that may
impact clinical outcomes (Kiebs et al., 2019). Further, we note
that etiology of developing MDD is likely quite heterogeneous
across patients (Winokur, 1997) and hence a more comprehensive
clinical and phenotypic characterization may help identifying
patients subgroups who benefit in particular from neurofeedback
training.
b) Choice of appropriate control conditions The use of control conditions is fundamental to determine if any
positive effect is caused by the neurofeedback protocol or by other
reasons. The best control design depends on the research interest,

42
 and a decision tree for control conditions for neurofeedback
applications was recently described by Sorger et al. (2019). In the
context of depressive patients, different control conditions should
be considered.
c) Adequately powered studies Powering studies to be able to detect meaningful effect sizes or
rule these out (Algermissen and Mehler, 2018). For instance,
studies may set minimal clinically important differences (MCID)
reported for depressed patients as their target effect size (Lakens
et al., 2018). Further, alternative sampling strategies such as
sequential Bayes Factor (SBF) sampling may be worthwhile
exploring for clinical neurofeedback studies (Schönbrodt and
Wagenmakers, 2018). Lastly, we recommend that null findings
are followed up with appropriate statistical tests that allow
providing evidence for the absence of an effect (Mehler et al.,
2019).
d) Online and offline quality control of signals Although several studies report the exclusion of subjects due to
excessive artifacts, only few studies intended to perform online
quality control and denoising. This is not a particular problem in
studies applying neurofeedback in MDD populations, but a
current issue in the field (Heunis et al., 2020). Thus, we
recommend that more rigorous approaches should be conducted
during experiments and the reporting of results, for instance with
regards to EOG and EMG noises in EEG-based protocols (Moretti
et al., 2003), or respiration and pulse waves in fMRI-based
experiments (Murphy et al., 2013). Further, it is fundamental to
evaluate and report differences in artifacts between groups (Ros
et al., 2020), since group-biased noisy data can lead to false
conclusions.
e) Standardization of protocols Several clinical neurofeedback studies targeting MDD patients do
not focus on new methodological approaches (for example,
testing signal processing and feedback presentation), but on
potential clinical, cognitive, or neural benefits of targeting one, or
more, brain regions (fMRI), or frequencies (EEG). In this context,

43
 the use of standardized methods to extract information from the
source signal, or to present the feedback would allow direct
comparison between studies. Also, potential comparisons depend
on a clear definition of success/learning, as well as the detailed
report of responders/literates and non-responders/illiterates. In
particular for depressed patients, insufficient self-regulation
success may result in frustration and potentially deteriorate
clinical outcome in individuals. In line with previous consensus
(Ros et al., 2020), we recommend that individual self-regulation
performances should be ideally reported and potential predictors
of self-regulation success explored and researchers should aim to
standardize approaches (Paret et al., 2019).
f) Basic methodological research As pointed out by others (Paret et al., 2019), more basic research
is needed to solve the many open methodological questions and
increase standardization and agreements to finally inform
translational work. This work seems particularly relevant for the
treatment of MDD, which affects the reward system. Lastly, it has
been suggested that neurofeedback may serve as a tool to test
neural models (Nielson et al., 2020) or biomarkers suggested for
MDD. However, there is reason for skepticism and discussions
about the reliability and validity of biomarker research remain
controversial (see Section 6f in the Supplementary Material).
g) Exploring the potential for children and young adults The current review was limited to studies conducted in adults.
However, given the low risk profile of non-invasive
neurofeedback training and the promising clinical findings found
in adults, we recommend that this approach should also be
explored in younger patients. In particular modulating self-
referential beliefs such as self-efficacy may provide substantial
clinical benefits related to anxious (Lewis et al., 2020) but also
depressive symptoms. Noteworthy, first results from an fMRI
neurofeedback study in depressed adolescence showed feasibility
and promising clinical potential (Quevedo et al., 2019) (see also
Section 6g in the Supplementary Material). Feasibility has also

44
 been recently demonstrated in targeting anxiety (Zich et al., 2020)
and depression (Quevedo et al., 2020) in adolescents.
h) Appropriate reporting of methods and results In addition to the proper experimental design, an appropriate
report of methods and results is crucial to advance the
neurofeedback field and propagate reliable results. For example,
an extensive methodological review showed that a substantial
portion of neurofeedback studies do not apply or report adequate
denoising methods in fMRI-based protocols (Heunis et al., 2020)
(complete data base available here: https://rtfmri-
methods.herokuapp.com/). the CRED-nf checklist was created in
a collaborative effort between several dozen laboratories to
support this matter (Ros et al., 2020), including an easy-to-use app
for quick validation (rtfin.org/CREDnf).
i) Study preregistration and open science research practices To make neurofeedback findings transparent and reliable, as well
as to allow further collaboration between research groups, we
strongly recommend that researchers explore and implement open
science research practices where possible (Allen and Mehler,
2019; Nosek et al., 2015) by preregistering their study protocol
and sharing the data that support their final results. Analytical
degrees of freedom remain a controversial topic in neuroimaging
(Botvinik-Nezer et al., 2020; Carp, 2012); real-time experiments
already predeclare a substantial part of their analysis pipeline
when setting parameters for real-time data analysis and it is hence
in particular suited for study preregistration (e.g., Mehler et al.
(2020)) or publishable research protocols (e.g., Cox et al. (2016)).
Regarding data sharing practices, researchers can benefit from
recommendations for reliable analysis pipelines (Nichols et al.,
2017), as well tools to standardize data accessibility and
reproducibility (Gorgolewski et al., 2017) and facilitate data
sharing (Gorgolewski et al., 2016; Poldrack et al., 2013).

45
5. Conclusion

Neurofeedback presents a complex, non-invasive intervention which aims to target cognitive and affective processes affected in patients
with depression through mental imagery-based self-regulation of functionally relevant brain areas or network. As such the approach has
good face validity for MDD. Patients have shown significant clinical improvements as well as cognitive and neural changes following
neurofeedback training with both EEG and fMRI-based protocols. Moreover, given the relatively low risk of side effects due to its non-
invasive nature, we consider neurofeedback in particular worth exploring as an augmentation therapy for patients who have already
received standard care but remain symptomatic. However, our review also found that most studies published thus far still lag current
best practice standards of study design and reporting quality. Some main issues are the lack of study preregistration, the use of mostly
small and/or unbalanced samples as well as the lack of control conditions, randomized treatment allocation or blinding. These issues
render the evaluation of clinical effects difficult and require improvements in future studies. Following a first attempt to quantify the
contribution of different non-specific effects for studies that included a control group, our results suggest that non-specific effects add
up such that more passive control conditions (e.g., continued standard care) yield larger group differences compared to more conservative
active control conditions (e.g., successful neurofeedback self-regulation training from an alternative brain region). We close with a set
of recommendations for future studies, which include suggestions for more comprehensive clinical documentation, considerations
regarding adequate control conditions, a synopsis of some statistical and study design aspects that can help achieving more adequately
powered and hence more informative studies, aspects concerning signal quality and protocol standardization, and lastly pointers to open
science resources.
Acknowledgements
We would like to thank (in alphabetical order) Eunjin Cheon, Paul Hamilton, Bon Hoon Koo, Argyris Stringaris, Kymberly Young,
and Roland Zahn for helpful feedback on an earlier version of this manuscript.

Conflict of interest
SHK and DMAM receive payments for work as independent advisors to a neurofeedback start-up company (Mendi Innovations AB).
LRT and DEJL declare no conflict of interest.

References

Algermissen, J., Mehler, D.M., 2018. May the power be with you: are there highly powered studies in neuroscience, and how can we
get more of them? Journal of neurophysiology 119, 2114-2117.
Alkoby, O., Abu-Rmileh, A., Shriki, O., Todder, D., 2018. Can we predict who will respond to neurofeedback? A review of the
inefficacy problem and existing predictors for successful EEG neurofeedback learning. Neuroscience 378, 155-164.

46
Allen, C., Mehler, D.M., 2019. Open science challenges, benefits and tips in early career and beyond. PLoS biology 17, e3000246.
Allison, B.Z., Neuper, C., 2010. Could anyone use a BCI?, Brain-computer interfaces. Springer, pp. 35-54.
Alonso, J., Lépine, J.-P., 2007. Overview of key data from the European Study of the Epidemiology of Mental Disorders (ESEMeD).
The Journal of clinical psychiatry.
Altman, D.G., 1998. Confidence intervals for the number needed to treat. Bmj 317, 1309-1312.
Arnold, L.E., Lofthouse, N., Hersch, S., Pan, X., Hurt, E., Bates, B., Kassouf, K., Moone, S., Grantier, C., 2013. EEG neurofeedback for
ADHD: double-blind sham-controlled randomized pilot feasibility trial. Journal of attention disorders 17, 410-419.
Arns, M., Batail, J.-M., Bioulac, S., Congedo, M., Daudet, C., Drapier, D., Fovet, T., Jardri, R., Le-Van-Quyen, M., Lotte, F., 2017.
Neurofeedback: One of today's techniques in psychiatry? L'Encéphale 43, 135-145.
Arns, M., De Ridder, S., Strehl, U., Breteler, M., Coenen, A., 2009. Efficacy of neurofeedback treatment in ADHD: the effects on
inattention, impulsivity and hyperactivity: a meta-analysis. Clinical EEG and neuroscience 40, 180-189.
Association, A.P., 2013. Diagnostic and statistical manual of mental disorders (DSM-5®). American Psychiatric Pub.
Baehr, E., Rosenfeld, J.P., Baehr, R., 1997. The clinical use of an alpha asymmetry protocol in the neurofeedback treatment of
depression: Two case studies. Journal of Neurotherapy 2, 10-23.
Bandura, A., 1982. Self-efficacy mechanism in human agency. American psychologist 37, 122.
Bari, A.A., Mikell, C.B., Abosch, A., Ben-Haim, S., Buchanan, R.J., Burton, A.W., Carcieri, S., Cosgrove, G.R., D’Haese, P.-F., Daskalakis,
Z.J., 2018. Charting the road forward in psychiatric neurosurgery: proceedings of the 2016 American Society for Stereotactic and
Functional Neurosurgery workshop on neuromodulation for psychiatric disorders. Journal of Neurology, Neurosurgery & Psychiatry
89, 886-896.
Becerra, J., Fernandez, T., Harmony, T., Caballero, M., Garcia, F., Fernandez-Bouzas, A., Santiago-Rodríguez, E., Prado-Alcalá, R.,
2006. Follow-up study of learning-disabled children treated with neurofeedback or placebo. Clinical EEG and neuroscience 37, 198-
203.
Beck, A.T., 2008. The evolution of the cognitive model of depression and its neurobiological correlates. American Journal of
Psychiatry 165, 969-977.
Bender, R., 2001. Calculating confidence intervals for the number needed to treat. Controlled clinical trials 22, 102-110.
Birbaumer, N., Ruiz, S., Sitaram, R., 2013. Learned regulation of brain metabolism. Trends in cognitive sciences 17, 295-302.
Botvinik-Nezer, R., Holzmeister, F., Camerer, C.F., Dreber, A., Huber, J., Johannesson, M., Kirchler, M., Iwanir, R., Mumford, J.A.,
Adcock, R.A., 2020. Variability in the analysis of a single neuroimaging dataset by many teams. Nature, 1-7.
Brigadoi, S., Cooper, R.J., 2015. How short is short? Optimum source–detector distance for short-separation channels in functional
near-infrared spectroscopy. Neurophotonics 2, 025005.

47
Bromet, E., Andrade, L.H., Hwang, I., Sampson, N.A., Alonso, J., De Girolamo, G., De Graaf, R., Demyttenaere, K., Hu, C., Iwata, N.,
2011. Cross-national epidemiology of DSM-IV major depressive episode. BMC medicine 9, 90.
Button, K.S., Ioannidis, J.P., Mokrysz, C., Nosek, B.A., Flint, J., Robinson, E.S., Munafò, M.R., 2013. Power failure: why small sample
size undermines the reliability of neuroscience. Nature reviews neuroscience 14, 365-376.
Carp, J., 2012. On the plurality of (methodological) worlds: estimating the analytic flexibility of FMRI experiments. Frontiers in
neuroscience 6, 149.
Chen, T.-C., Lin, I.-M., 2020. The learning effects and curves during high beta down-training neurofeedback for patients with major
depressive disorder. Journal of Affective Disorders 266, 235-242.
Cheon, E.-J., Koo, B.-H., Choi, J.-H., 2016. The efficacy of neurofeedback in patients with major depressive disorder: An open labeled
prospective study. Applied psychophysiology and biofeedback 41, 103-110.
Cheon, E.-J., Koo, B.-H., Seo, W.-S., Lee, J.-Y., Choi, J.-H., Song, S.-H., 2015. Effects of neurofeedback on adult patients with
psychiatric disorders in a naturalistic setting. Applied psychophysiology and biofeedback 40, 17-24.
Choi, S.W., Chi, S.E., Chung, S.Y., Kim, J.W., Ahn, C.Y., Kim, H.T., 2011. Is alpha wave neurofeedback effective with randomized clinical
trials in depression? A pilot study. Neuropsychobiology 63, 43-51.
Clark, D.A., Beck, A.T., 2010. Cognitive theory and therapy of anxiety and depression: Convergence with neurobiological findings.
Trends in cognitive sciences 14, 418-424.
Clark, D.A., Beck, A.T., 2011. Cognitive therapy of anxiety disorders: Science and practice. Guilford Press.
Cortese, S., Ferrin, M., Brandeis, D., Holtmann, M., Aggensteiner, P., Daley, D., Santosh, P., Simonoff, E., Stevenson, J., Stringaris, A.,
2016. Neurofeedback for attention-deficit/hyperactivity disorder: meta-analysis of clinical and neuropsychological outcomes from
randomized controlled trials. Journal of the American Academy of Child & Adolescent Psychiatry 55, 444-455.
Cox, W.M., Subramanian, L., Linden, D.E., Lührs, M., McNamara, R., Playle, R., Hood, K., Watson, G., Whittaker, J.R., Sakhuja, R.,
2016. Neurofeedback training for alcohol dependence versus treatment as usual: study protocol for a randomized controlled trial.
Trials 17, 1-10.
Craig, P., Dieppe, P., Macintyre, S., Michie, S., Nazareth, I., Petticrew, M., 2008. Developing and evaluating complex interventions:
the new Medical Research Council guidance. Bmj 337.
Cui, X., Bray, S., Bryant, D.M., Glover, G.H., Reiss, A.L., 2011. A quantitative comparison of NIRS and fMRI across multiple cognitive
tasks. Neuroimage 54, 2808-2821.
Cuijpers, P., Turner, E.H., Koole, S.L., Van Dijke, A., Smit, F., 2014. What is the threshold for a clinically relevant effect? The case of
major depressive disorders. Depression and anxiety 31, 374-378.
Curran, E.A., Stokes, M.J., 2003. Learning to control brain activity: A review of the production and control of EEG components for
driving brain–computer interface (BCI) systems. Brain and cognition 51, 326-336.

48
Cusin, C., Dougherty, D.D., 2012. Somatic therapies for treatment-resistant depression: ECT, TMS, VNS, DBS. Biology of Mood &
Anxiety Disorders 2, 14.
Da Silva, F.L., 2009. EEG: origin and measurement, EEG-fMRI. Springer, pp. 19-38.
De Vos, M., Kroesen, M., Emkes, R., Debener, S., 2014. P300 speller BCI with a mobile EEG system: comparison to a traditional
amplifier. Journal of neural engineering 11, 036008.
DeAngelis, C.D., Drazen, J.M., Frizelle, F.A., Haug, C., Hoey, J., Horton, R., Kotzin, S., Laine, C., Marusic, A., Overbeke, A.J.P., 2005.
Clinical trial registration: a statement from the International Committee of Medical Journal Editors. Archives of dermatology 141, 76-
77.
Delaloye, S., Holtzheimer, P.E., 2014. Deep brain stimulation in the treatment of depression. Dialogues in clinical neuroscience 16,
83.
Deldin, P.J., Chiu, P., 2005. Cognitive restructuring and EEG in major depression. Biological psychology 70, 141-151.
Dimidjian, S., Barrera Jr, M., Martell, C., Munoz, R.F., Lewinsohn, P.M., 2011. The origins and current status of behavioral activation
treatments for depression. Annual review of clinical psychology 7, 1-38.
Disner, S.G., Beevers, C.G., Haigh, E.A., Beck, A.T., 2011. Neural mechanisms of the cognitive model of depression. Nature Reviews
Neuroscience 12, 467-477.
Disner, S.G., Shumake, J.D., Beevers, C.G., 2017. Self-referential schemas and attentional bias predict severity and naturalistic course
of depression symptoms. Cognition and Emotion 31, 632-644.
Earnest, C., 1999. Single case study of EEG asymmetry biofeedback for depression: An independent replication in an adolescent.
Journal of Neurotherapy 3, 28-35.
Edlinger, G., Allison, B.Z., Guger, C., 2015. How many people can use a BCI system?, Clinical Systems Neuroscience. Springer, pp. 33-
66.
Enriquez-Geppert, S., Huster, R.J., Herrmann, C.S., 2017. EEG-neurofeedback as a tool to modulate cognition and behavior: a review
tutorial. Frontiers in human neuroscience 11, 51.
Escolano, C., Navarro-Gil, M., Garcia-Campayo, J., Congedo, M., De Ridder, D., Minguez, J., 2014. A controlled study on the cognitive
effect of alpha neurofeedback training in patients with major depressive disorder. Frontiers in behavioral neuroscience 8, 296.
Fava, M., Davidson, K.G., 1996. Definition and epidemiology of treatment-resistant depression. Psychiatric Clinics of North America
19, 179-200.
Fodor, L.A., Georgescu, R., Cuijpers, P., Szamoskozi, Ş., David, D., Furukawa, T.A., Cristea, I.A., 2020. Efficacy of cognitive bias
modification interventions in anxiety and depressive disorders: a systematic review and network meta-analysis. The Lancet
Psychiatry 7, 506-514.

49
Fournier, J.C., DeRubeis, R.J., Hollon, S.D., Dimidjian, S., Amsterdam, J.D., Shelton, R.C., Fawcett, J., 2010. Antidepressant drug effects
and depression severity: a patient-level meta-analysis. Jama 303, 47-53.
Fovet, T., Jardri, R., Linden, D., 2015. Current issues in the use of fMRI-based neurofeedback to relieve psychiatric symptoms. Current
pharmaceutical design 21, 3384-3394.
Fried, E.I., Nesse, R.M., 2015. Depression is not a consistent syndrome: an investigation of unique symptom patterns in the STAR* D
study. Journal of affective disorders 172, 96-102.
Fried, E.I., van Borkulo, C.D., Cramer, A.O., Boschloo, L., Schoevers, R.A., Borsboom, D., 2017. Mental disorders as networks of
problems: a review of recent insights. Social Psychiatry and Psychiatric Epidemiology 52, 1-10.
Gevensleben, H., Holl, B., Albrecht, B., Schlamp, D., Kratz, O., Studer, P., Rothenberger, A., Moll, G.H., Heinrich, H., 2010.
Neurofeedback training in children with ADHD: 6-month follow-up of a randomised controlled trial. European child & adolescent
psychiatry 19, 715-724.
Goldway, N., Ablin, J., Lubin, O., Zamir, Y., Keynan, J.N., Or-Borichev, A., Cavazza, M., Charles, F., Intrator, N., Brill, S., 2019. Volitional
limbic neuromodulation exerts a beneficial clinical effect on Fibromyalgia. Neuroimage 186, 758-770.
Gorgolewski, K.J., Alfaro-Almagro, F., Auer, T., Bellec, P., Capotă, M., Chakravarty, M.M., Churchill, N.W., Cohen, A.L., Craddock, R.C.,
Devenyi, G.A., 2017. BIDS apps: Improving ease of use, accessibility, and reproducibility of neuroimaging data analysis methods. PLoS
computational biology 13, e1005209.
Gorgolewski, K.J., Varoquaux, G., Rivera, G., Schwartz, Y., Sochat, V.V., Ghosh, S.S., Maumet, C., Nichols, T.E., Poline, J.-B., Yarkoni, T.,
2016. NeuroVault. org: A repository for sharing unthresholded statistical maps, parcellations, and atlases of the human brain.
Neuroimage 124, 1242-1244.
Groenewold, N.A., Opmeer, E.M., de Jonge, P., Aleman, A., Costafreda, S.G., 2013. Emotional valence modulates brain functional
abnormalities in depression: evidence from a meta-analysis of fMRI studies. Neuroscience & Biobehavioral Reviews 37, 152-163.
Group, T.N.C., Arnold, L.E., Arns, M., Barterian, J., Bergman, R., Black, S., Conners, C.K., Connor, S., Dasgupta, S., deBeus, R., 2020.
Double-Blind Placebo-Controlled Randomized Clinical Trial of Neurofeedback for Attention-Deficit/Hyperactivity Disorder With 13
Month Follow-up. Journal of the American Academy of Child & Adolescent Psychiatry.
Gruzelier, J.H., 2014. EEG-neurofeedback for optimising performance. I: a review of cognitive and affective outcome in healthy
participants. Neuroscience & Biobehavioral Reviews 44, 124-141.
Hamilton, J.P., Farmer, M., Fogelman, P., Gotlib, I.H., 2015. Depressive rumination, the default-mode network, and the dark matter
of clinical neuroscience. Biological psychiatry 78, 224-230.
Hamilton, J.P., Glover, G.H., Bagarinao, E., Chang, C., Mackey, S., Sacchet, M.D., Gotlib, I.H., 2016. Effects of salience-network-node
neurofeedback training on affective biases in major depressive disorder. Psychiatry Research: Neuroimaging 249, 91-96.
Hammond, D.C., 2005. Neurofeedback treatment of depression and anxiety. Journal of Adult Development 12, 131-137.

50
Hampson, M., Scheinost, D., Qiu, M., Bhawnani, J., Lacadie, C.M., Leckman, J.F., Constable, R.T., Papademetris, X., 2011. Biofeedback
of real-time functional magnetic resonance imaging data from the supplementary motor area reduces functional connectivity to
subcortical regions. Brain connectivity 1, 91-98.
Harmon-Jones, E., Gable, P.A., Peterson, C.K., 2010. The role of asymmetric frontal cortical activity in emotion-related phenomena: A
review and update. Biological psychology 84, 451-462.
Heunis, S., Hellrung, L., Van der Meer, J., Bergert, S., Sladky, R., Pamplona, G., Skouras, S., 2019. rtQC: An open-source toolbox for
real-time fMRI quality control.
Heunis, S., Lamerichs, R., Zinger, S., Caballero-Gaudes, C., Jansen, J.F., Aldenkamp, B., Breeuwer, M., 2020. Quality and denoising in
real-time functional magnetic resonance imaging neurofeedback: A methods review. Human Brain Mapping.
Hinterberger, T., Kübler, A., Kaiser, J., Neumann, N., Birbaumer, N., 2003. A brain–computer interface (BCI) for the locked-in:
comparison of different EEG classifications for the thought translation device. Clinical Neurophysiology 114, 416-425.
Hoertel, N., Franco, S., Wall, M., Oquendo, M., Kerridge, B., Limosin, F., Blanco, C., 2015. Mental disorders and risk of suicide
attempt: a national prospective study. Molecular psychiatry 20, 718.
Hofmann, S.G., Curtiss, J., McNally, R.J., 2016. A complex network perspective on clinical science. Perspectives on Psychological
Science 11, 597-605.
Hoshi, Y., 2003. Functional near-infrared optical imaging: Utility and limitations in human brain mapping. Psychophysiology 40, 511-
520.
Huppert, T.J., Hoge, R.D., Diamond, S.G., Franceschini, M.A., Boas, D.A., 2006. A temporal comparison of BOLD, ASL, and NIRS
hemodynamic responses to motor stimuli in adult humans. Neuroimage 29, 368-382.
Jaeckle, T., Williams, S.C., Barker, G.J., Basilio, R., Carr, E., Goldsmith, K., Colasanti, A., Giampietro, V., Cleare, A., Young, A.H., 2019.
Self-blaming emotions in major depression: a randomised pilot trial comparing fMRI neurofeedback training with self-guided
psychological strategies (NeuroMooD). medRxiv, 19004309.
Jensen, J.S., Bielefeldt, A.Ø., Hróbjartsson, A., 2017. Active placebo control groups of pharmacological interventions were rarely used
but merited serious consideration: a methodological overview. Journal of clinical epidemiology 87, 35-46.
Johnston, S.J., Boehm, S.G., Healy, D., Goebel, R., Linden, D.E., 2010. Neurofeedback: A promising tool for the self-regulation of
emotion networks. Neuroimage 49, 1066-1072.
Kaiser, R.H., Andrews-Hanna, J.R., Wager, T.D., Pizzagalli, D.A., 2015. Large-scale network dysfunction in major depressive disorder: a
meta-analysis of resting-state functional connectivity. JAMA psychiatry 72, 603-611.
Kiebs, M., Hurlemann, R., Mutz, J., 2019. Repetitive transcranial magnetic stimulation in non-treatment-resistant depression. The
British Journal of Psychiatry 215, 445-446.

51
Kim, S., Birbaumer, N., 2014. Real-time functional MRI neurofeedback: a tool for psychiatry. Current opinion in psychiatry 27, 332-
336.
Kirsch, I., Deacon, B.J., Huedo-Medina, T.B., Scoboria, A., Moore, T.J., Johnson, B.T., 2008. Initial severity and antidepressant
benefits: a meta-analysis of data submitted to the Food and Drug Administration. PLoS medicine 5, e45.
Kohl, S.H., Mehler, D.M.A., Lührs, M., Thibault, R.T., Konrad, K., Sorger, B., 2020. The potential of functional near-infrared
spectroscopy-based neurofeedback – a systematic review and recommendations for best practice. Frontiers in Neuroscience 14,
594.
Koush, Y., Rosa, M.J., Robineau, F., Heinen, K., Rieger, S.W., Weiskopf, N., Vuilleumier, P., Van De Ville, D., Scharnowski, F., 2013.
Connectivity-based neurofeedback: dynamic causal modeling for real-time fMRI. Neuroimage 81, 422-430.
Krucoff, M.O., Rahimpour, S., Slutzky, M.W., Edgerton, V.R., Turner, D.A., 2016. Enhancing nervous system recovery through
neurobiologics, neural interface training, and neurorehabilitation. Frontiers in neuroscience 10, 584.
Krusienski, D.J., Sellers, E.W., Cabestaing, F., Bayoudh, S., McFarland, D.J., Vaughan, T.M., Wolpaw, J.R., 2006. A comparison of
classification techniques for the P300 Speller. Journal of neural engineering 3, 299.
Kupfer, D.J., Frank, E., Phillips, M.L., 2012. Major depressive disorder: new clinical, neurobiological, and treatment perspectives. The
Lancet 379, 1045-1055.
Lakens, D., Scheel, A.M., Isager, P.M., 2018. Equivalence testing for psychological research: A tutorial. Advances in Methods and
Practices in Psychological Science 1, 259-269.
Lee, Y.-J., Lee, G.-W., Seo, W.-S., Koo, B.-H., Kim, H.-G., Cheon, E.-J., 2019. Neurofeedback Treatment on Depressive Symptoms and
Functional Recovery in Treatment-Resistant Patients with Major Depressive Disorder: an Open-Label Pilot Study. Journal of Korean
medical science 34.
Lewinsohn, P.M., Rosenbaum, M., 1987. Recall of parental behavior by acute depressives, remitted depressives, and nondepressives.
Journal of Personality and Social Psychology 52, 611.
Lewis, K.M., Matsumoto, C., Cardinale, E., Jones, E.L., Gold, A.L., Stringaris, A., Leibenluft, E., Pine, D.S., Brotman, M.A., 2020. Self-
Efficacy As a Target for Neuroscience Research on Moderators of Treatment Outcomes in Pediatric Anxiety. Journal of Child and
Adolescent Psychopharmacology 30, 205-214.
Linden, D.E., 2014. Neurofeedback and networks of depression. Dialogues in clinical neuroscience 16, 103.
Linden, D.E., Habes, I., Johnston, S.J., Linden, S., Tatineni, R., Subramanian, L., Sorger, B., Healy, D., Goebel, R., 2012. Real-time self-
regulation of emotion networks in patients with depression. PloS one 7, e38115.
Linden, D.E., Turner, D.L., 2016. Real-time functional magnetic resonance imaging neurofeedback in motor neurorehabilitation.
Current opinion in neurology 29, 412.

52
Linhartová, P., Látalová, A., Kóša, B., Kašpárek, T., Schmahl, C., Paret, C., 2019. fMRI neurofeedback in emotion regulation: A
literature review. NeuroImage 193, 75-92.
Lotte, F., Congedo, M., Lécuyer, A., Lamarche, F., Arnaldi, B., 2007. A review of classification algorithms for EEG-based brain–
computer interfaces. Journal of neural engineering 4, R1.
Lubianiker, N., Goldway, N., Fruchtman-Steinbok, T., Paret, C., Keynan, J.N., Singer, N., Cohen, A., Kadosh, K.C., Linden, D.E., Hendler,
T., 2019. Process-based framework for precise neuromodulation. Nature human behaviour 3, 436-445.
MacDuffie, K.E., MacInnes, J., Dickerson, K.C., Eddington, K.M., Strauman, T.J., Adcock, R.A., 2018. Single session real-time fMRI
neurofeedback has a lasting impact on cognitive behavioral therapy strategies. NeuroImage: Clinical 19, 868-875.
Mano, M., Lécuyer, A., Bannier, E., Perronnet, L., Noorzadeh, S., Barillot, C., 2017. How to build a hybrid neurofeedback platform
combining EEG and fMRI. Frontiers in neuroscience 11, 140.
McFarland, D.J., McCane, L.M., Wolpaw, J.R., 1998. EEG-based communication and control: short-term role of feedback. IEEE
Transactions on Rehabilitation Engineering 6, 7-11.
Megumi, F., Yamashita, A., Kawato, M., Imamizu, H., 2015. Functional MRI neurofeedback training on connectivity between two
regions induces long-lasting changes in intrinsic functional network. Frontiers in human neuroscience 9, 160.
Mehler, D., Williams, A.N., Whittaker, J., Krause, F., Lührs, M., Kunas, S., Wise, R., Shetty, H., Turner, D.L., Linden, D.E., 2020. Graded
fMRI neurofeedback training of motor imagery in middle cerebral artery stroke patients-a preregistered proof-of-concept study.
Frontiers in Human Neuroscience 14, 226.
Mehler, D.M., Edelsbrunner, P.A., Matić, K., 2019. Appreciating the significance of nonsignificant findings in psychology. Journal of
European Psychology Students 10.
Mehler, D.M., Sokunbi, M.O., Habes, I., Barawi, K., Subramanian, L., Range, M., Evans, J., Hood, K., Lührs, M., Keedwell, P., 2018.
Targeting the affective brain—a randomized controlled trial of real-time fMRI neurofeedback in patients with depression.
Neuropsychopharmacology 43, 2578.
Mehler, D.M.A., Kording, K.P., 2018. The lure of causal statements: Rampant mis-inference of causality in estimated connectivity.
arXiv preprint arXiv:1812.03363.
Micoulaud-Franchi, J.-A., Batail, J.-M., Fovet, T., Philip, P., Cermolacce, M., Jaumard-Hakoun, A., Vialatte, F., 2019. Towards a
Pragmatic Approach to a Psychophysiological Unit of Analysis for Mental and Brain Disorders: An EEG-Copeia for Neurofeedback.
Applied psychophysiology and biofeedback, 1-22.
Micoulaud-Franchi, J.-A., Fovet, T., 2018. A framework for disentangling the hyperbolic truth of neurofeedback: Comment on
Thibault and Raz (2017).
Mirifar, A., Beckmann, J., Ehrlenspiel, F., 2017. Neurofeedback as supplementary training for optimizing athletes’ performance: A
systematic review with implications for future research. Neuroscience & Biobehavioral Reviews 75, 419-432.

53
Moher, D., Liberati, A., Tetzlaff, J., Altman, D.G., The, P.G., 2009. Preferred Reporting Items for Systematic Reviews and Meta-
Analyses: The PRISMA Statement. PLOS Medicine 6, e1000097.
Moncrieff, J., Wessely, S., Hardy, R., 2004. Active placebos versus antidepressants for depression. Cochrane database of systematic
reviews.
Moretti, D.V., Babiloni, F., Carducci, F., Cincotti, F., Remondini, E., Rossini, P., Salinari, S., Babiloni, C., 2003. Computerized processing
of EEG–EOG–EMG artifacts for multi-centric studies in EEG oscillations and event-related potentials. International Journal of
Psychophysiology 47, 199-216.
Mulders, P.C., van Eijndhoven, P.F., Schene, A.H., Beckmann, C.F., Tendolkar, I., 2015. Resting-state functional connectivity in major
depressive disorder: a review. Neuroscience & Biobehavioral Reviews 56, 330-344.
Müller-Putz, G., Leeb, R., Tangermann, M., Höhne, J., Kübler, A., Cincotti, F., Mattia, D., Rupp, R., Müller, K.-R., Millán, J.d.R., 2015.
Towards noninvasive hybrid brain–computer interfaces: framework, practice, clinical application, and beyond. Proceedings of the
IEEE 103, 926-943.
Murphy, K., Birn, R.M., Bandettini, P.A., 2013. Resting-state fMRI confounds and cleanup. Neuroimage 80, 349-359.
Mutz, J., Vipulananthan, V., Carter, B., Hurlemann, R., Fu, C.H., Young, A.H., 2019. Comparative efficacy and acceptability of non-
surgical brain stimulation for the acute treatment of major depressive episodes in adults: systematic review and network meta-
analysis. bmj 364, l1079.
Newcombe, R.G., 1998. Interval estimation for the difference between independent proportions: comparison of eleven methods.
Statistics in medicine 17, 873-890.
Nichols, T.E., Das, S., Eickhoff, S.B., Evans, A.C., Glatard, T., Hanke, M., Kriegeskorte, N., Milham, M.P., Poldrack, R.A., Poline, J.-B.,
2017. Best practices in data analysis and sharing in neuroimaging using MRI. Nature neuroscience 20, 299-303.
Nielson, D.M., Keren, H., O'callaghan, G., Jackson, S.M., Douka, I., Zheng, C.Y., Belil, P.V.-R., Pornpattananangkul, N., Camp, C.C.,
Gorham, L.S., 2020. Great Expectations: A Critical Review of and Recommendations for the study of Reward Processing as a Cause
and Predictor of Depression. BioRxiv.
Nieuwenhuis, S., Forstmann, B.U., Wagenmakers, E.-J., 2011. Erroneous analyses of interactions in neuroscience: a problem of
significance. Nature neuroscience 14, 1105-1107.
Nijholt, A., Allison, B.Z., Jacob, R.J., 2011. Brain-computer interaction: can multimodality help?, Proceedings of the 13th international
conference on multimodal interfaces. ACM, pp. 35-40.
Nosek, B.A., Alter, G., Banks, G.C., Borsboom, D., Bowman, S.D., Breckler, S.J., Buck, S., Chambers, C.D., Chin, G., Christensen, G.,
2015. Promoting an open research culture. Science 348, 1422-1425.
Nosek, B.A., Ebersole, C.R., DeHaven, A.C., Mellor, D.T., 2018. The preregistration revolution. Proceedings of the National Academy
of Sciences 115, 2600-2606.

54
Olfson, M., Blanco, C., Wall, M., Liu, S.-M., Saha, T.D., Pickering, R.P., Grant, B.F., 2017. National trends in suicide attempts among
adults in the United States. JAMA psychiatry 74, 1095-1103.
Otte, C., Gold, S.M., Penninx, B.W., Pariante, C.M., Etkin, A., Fava, M., Mohr, D.C., Schatzberg, A.F., 2016. Major depressive disorder.
Nature Reviews Disease Primers 2, 16065.
Palmiero, M., Piccardi, L., 2017. Frontal EEG asymmetry of mood: A mini-review. Frontiers in Behavioral Neuroscience 11, 224.
Paquette, V., Beauregard, M., Beaulieu-Prévost, D., 2009. Effect of a psychoneurotherapy on brain electromagnetic tomography in
individuals with major depressive disorder. Psychiatry Research: Neuroimaging 174, 231-239.
Paret, C., Goldway, N., Zich, C., Keynan, J.N., Hendler, T., Linden, D., Kadosh, K.C., 2019. Current progress in real-time functional
magnetic resonance-based neurofeedback: Methodological challenges and achievements. NeuroImage, 116107.
Patten, S.B., 2009. Accumulation of major depressive episodes over time in a prospective study indicates that retrospectively
assessed lifetime prevalence estimates are too low. BMC psychiatry 9, 19.
Peckham, A.D., McHugh, R.K., Otto, M.W., 2010. A meta-analysis of the magnitude of biased attention in depression. Depression and
anxiety 27, 1135-1142.
Peeters, F., Oehlen, M., Ronner, J., van Os, J., Lousberg, R., 2014. Neurofeedback as a treatment for major depressive disorder–a
pilot study. PloS one 9, e91837.
Perronnet, L., Lécuyer, A., Mano, M., Bannier, E., Lotte, F., Clerc, M., Barillot, C., 2017. Unimodal versus bimodal EEG-fMRI
neurofeedback of a motor imagery task. Frontiers in human neuroscience 11, 193.
Poldrack, R.A., Barch, D.M., Mitchell, J., Wager, T., Wagner, A.D., Devlin, J.T., Cumba, C., Koyejo, O., Milham, M., 2013. Toward open
sharing of task-based fMRI data: the OpenfMRI project. Frontiers in neuroinformatics 7, 12.
Poldrack, R.A., Gorgolewski, K.J., 2014. Making big data open: data sharing in neuroimaging. Nature neuroscience 17, 1510.
Quevedo, K., Liu, G., Teoh, J.Y., Ghosh, S., Zeffiro, T., Ahrweiler, N., Zhang, N., Wedan, R., Oh, S., Guercio, G., 2019. Neurofeedback
and neuroplasticity of visual self-processing in depressed and healthy adolescents: A preliminary study. Developmental cognitive
neuroscience 40, 100707.
Quevedo, K., Yuan Teoh, J., Engstrom, M., Wedan, R., Santana-Gonzalez, C., Zewde, B., Porter, D., Cohen Kadosh, K., 2020. Amygdala
Circuitry During Neurofeedback Training and Symptoms’ Change in Adolescents With Varying Depression. Frontiers in behavioral
neuroscience 14, 110.
Ramirez, R., Palencia-Lefler, M., Giraldo, S., Vamvakousis, Z., 2015. Musical neurofeedback for treating depression in elderly people.
Frontiers in neuroscience 9, 354.
Ramot, M., Kimmich, S., Gonzalez-Castillo, J., Roopchansingh, V., Popal, H., White, E., Gotts, S.J., Martin, A., 2017. Direct modulation
of aberrant brain network connectivity through real-time NeuroFeedback. elife 6, e28974.

55
Rance, M., Walsh, C., Sukhodolsky, D.G., Pittman, B., Qiu, M., Kichuk, S.A., Wasylink, S., Koller, W.N., Bloch, M., Gruner, P., 2018.
Time course of clinical change following neurofeedback. Neuroimage 181, 807-813.
Ries, A.J., Touryan, J., Vettel, J., McDowell, K., Hairston, W.D., 2014. A comparison of electroencephalography signals acquired from
conventional and mobile systems. Journal of Neuroscience and Neuroengineering 3, 10-20.
Robinson, L., Kellett, S., Delgadillo, J., 2020. Dose-response patterns in low and high intensity cognitive behavioral therapy for
common mental health problems. Depression and Anxiety 37, 285-294.
Ros, T., Enriquez-Geppert, S., Zotev, V., Young, K., Wood, G., Whitfield-Gabrieli, S., Wan, F., Vialatte, F., Van De Ville, D., Todder, D.,
2020. Consensus on the reporting and experimental design of clinical and cognitive-behavioural neurofeedback studies (CRED-nf
checklist). Brain 143, 1674–1685.
Ros, T., J Baars, B., Lanius, R.A., Vuilleumier, P., 2014. Tuning pathological brain oscillations with neurofeedback: a systems
neuroscience framework. Frontiers in human neuroscience 8, 1008.
Rosenfeld, J.P., Baehr, E., Baehr, R., Gotlib, I.H., Ranganath, C., 1996. Preliminary evidence that daily changes in frontal alpha
asymmetry correlate with changes in affect in therapy sessions. International Journal of Psychophysiology 23, 137-141.
Rossi, S., Hallett, M., Rossini, P.M., Pascual-Leone, A., Group, S.o.T.C., 2009. Safety, ethical considerations, and application guidelines
for the use of transcranial magnetic stimulation in clinical practice and research. Clinical neurophysiology 120, 2008-2039.
Rush, A.J., Trivedi, M.H., Wisniewski, S.R., Nierenberg, A.A., Stewart, J.W., Warden, D., Niederehe, G., Thase, M.E., Lavori, P.W.,
Lebowitz, B.D., 2006. Acute and longer-term outcomes in depressed outpatients requiring one or several treatment steps: a STAR* D
report. American Journal of Psychiatry 163, 1905-1917.
Sacchet, M.D., Gotlib, I.H., 2016. Neurofeedback training for major depressive disorder: recent developments and future directions.
Taylor & Francis.
Sackeim, H.A., Prudic, J., Fuller, R., Keilp, J., Lavori, P.W., Olfson, M., 2007. The cognitive effects of electroconvulsive therapy in
community settings. Neuropsychopharmacology 32, 244-254.
Sato, H., Kawahara, J.-i., 2011. Selective bias in retrospective self-reports of negative mood states. Anxiety, Stress & Coping 24, 359-
367.
Scheinost, D., Stoica, T., Saksa, J., Papademetris, X., Constable, R., Pittenger, C., Hampson, M., 2013. Orbitofrontal cortex
neurofeedback produces lasting changes in contamination anxiety and resting-state connectivity. Translational psychiatry 3, e250.
Schneider, F., Heimann, H., Mattes, R., Lutzenberger, W., Birbaumer, N., 1992. Self-regulation of slow cortical potentials in
psychiatric patients: Depression. Biofeedback and self-regulation 17, 203-214.
Schönbrodt, F.D., Wagenmakers, E.-J., 2018. Bayes factor design analysis: Planning for compelling evidence. Psychonomic bulletin &
review 25, 128-142.

56
Sharp, T., 2012. Molecular and cellular mechanisms of antidepressant action, Behavioral Neurobiology of Depression and Its
Treatment. Springer, pp. 309-325.
Simonsohn, U., Nelson, L.D., Simmons, J.P., 2014. p-curve and effect size: Correcting for publication bias using only significant
results. Perspectives on Psychological Science 9, 666-681.
Sitaram, R., Ros, T., Stoeckel, L., Haller, S., Scharnowski, F., Lewis-Peacock, J., Weiskopf, N., Blefari, M.L., Rana, M., Oblak, E., 2017.
Closed-loop brain training: the science of neurofeedback. Nature Reviews Neuroscience 18, 86.
Skottnik, L., Linden, D.E.J., 2019. Mental Imagery and Brain Regulation—New Links Between Psychotherapy and Neuroscience.
Frontiers in Psychiatry 10.
Sorger, B., Scharnowski, F., Linden, D.E., Hampson, M., Young, K.D., 2019. Control freaks: Towards optimal selection of control
conditions for fMRI neurofeedback studies. Neuroimage 186, 256-265.
Strangman, G., Boas, D.A., Sutton, J.P., 2002. Non-invasive neuroimaging using near-infrared light. Biological psychiatry 52, 679-693.
Sulzer, J., Haller, S., Scharnowski, F., Weiskopf, N., Birbaumer, N., Blefari, M.L., Bruehl, A.B., Cohen, L.G., DeCharms, R.C., Gassert, R.,
2013. Real-time fMRI neurofeedback: progress and challenges. Neuroimage 76, 386-399.
Thibault, R.T., Lifshitz, M., Raz, A., 2016. The self-regulating brain and neurofeedback: experimental science and clinical promise.
cortex 74, 247-261.
Thibault, R.T., Lifshitz, M., Raz, A., 2017. Neurofeedback or neuroplacebo? Brain 140, 862-864.
Thibault, R.T., MacPherson, A., Lifshitz, M., Roth, R.R., Raz, A., 2018. Neurofeedback with fMRI: A critical systematic review.
Neuroimage 172, 786-807.
Thibault, R.T., Raz, A., 2016. Neurofeedback: The power of psychosocial therapeutics. The Lancet Psychiatry 3, e18.
Thibodeau, R., Jorgensen, R.S., Kim, S., 2006. Depression, anxiety, and resting frontal EEG asymmetry: a meta-analytic review.
Journal of abnormal psychology 115, 715.
Tufanaru, C., Munn, Z., Aromataris, E., Campbell, J., Hopp, L., 2017. Chapter 3: Systematic reviews of effectiveness. Joanna Briggs
Institute Reviewer's Manual. The Joanna Briggs Institute.
Van Doren, J., Arns, M., Heinrich, H., Vollebregt, M.A., Strehl, U., Loo, S.K., 2019. Sustained effects of neurofeedback in ADHD: a
systematic review and meta-analysis. European child & adolescent psychiatry 28, 293-305.
Vernon, D., Frick, A., Gruzelier, J., 2004. Neurofeedback as a treatment for ADHD: a methodological review with implications for
future research. Journal of Neurotherapy 8, 53-82.
Villringer, A., Planck, J., Hock, C., Schleinkofer, L., Dirnagl, U., 1993. Near infrared spectroscopy (NIRS): a new tool to study
hemodynamic changes during activation of brain function in human adults. Neuroscience letters 154, 101-104.
Walker, J.E., Lawson, R., 2013. FP02 beta training for drug-resistant depression—A new protocol that usually reduces depression and
keeps it reduced. Journal of Neurotherapy 17, 198-200.

57
Wang, S.-Y., Lin, I.-M., Fan, S.-Y., Tsai, Y.-C., Yen, C.-F., Yeh, Y.-C., Huang, M.-F., Lee, Y., Chiu, N.-M., Hung, C.-F., 2019. The effects of
alpha asymmetry and high-beta down-training neurofeedback for patients with the major depressive disorder and anxiety
symptoms. Journal of affective disorders 257, 287-296.
Wang, S.-Y., Lin, I.-M., Peper, E., Chen, Y.-T., Tang, T.-C., Yeh, Y.-C., Tsai, Y.-C., Chu, C.-C., 2016. The efficacy of neurofeedback among
patients with major depressive disorder: Preliminary study. NeuroRegulation 3, 127-127.
Wang, T., Mantini, D., Gillebert, C.R., 2018. The potential of real-time fMRI neurofeedback for stroke rehabilitation: A systematic
review. cortex 107, 148-165.
Watanabe, T., Sasaki, Y., Shibata, K., Kawato, M., 2017. Advances in fMRI real-time neurofeedback. Trends in cognitive sciences 21,
997-1010.
Weiskopf, N., 2012. Real-time fMRI and its application to neurofeedback. Neuroimage 62, 682-692.
Weiss, F., Zamoscik, V., Schmidt, S.N., Halli, P., Kirsch, P., Gerchen, M.F., 2020. Just a very expensive breathing training? Risk of
respiratory artefacts in functional connectivity-based real-time fMRI neurofeedback. Neuroimage 210, 116580.
Winokur, G., 1997. All roads lead to depression: clinically homogeneous, etiologically heterogeneous. Journal of Affective Disorders
45, 97-108.
Wood, G., Kober, S.E., 2018. EEG neurofeedback is under strong control of psychosocial factors. Applied psychophysiology and
biofeedback 43, 293-300.
Young, K.D., Misaki, M., Harmer, C.J., Victor, T., Zotev, V., Phillips, R., Siegle, G.J., Drevets, W.C., Bodurka, J., 2017a. Real-time
functional magnetic resonance imaging amygdala neurofeedback changes positive information processing in major depressive
disorder. Biological psychiatry 82, 578-586.
Young, K.D., Siegle, G.J., Misaki, M., Zotev, V., Phillips, R., Drevets, W.C., Bodurka, J., 2018a. Altered task-based and resting-state
amygdala functional connectivity following real-time fMRI amygdala neurofeedback training in major depressive disorder.
NeuroImage: Clinical 17, 691-703.
Young, K.D., Siegle, G.J., Zotev, V., Phillips, R., Misaki, M., Yuan, H., Drevets, W.C., Bodurka, J., 2017b. Randomized clinical trial of
real-time fMRI amygdala neurofeedback for major depressive disorder: effects on symptoms and autobiographical memory recall.
American Journal of Psychiatry 174, 748-755.
Young, K.D., Zotev, V., Phillips, R., Misaki, M., Drevets, W.C., Bodurka, J., 2018b. Amygdala real-time functional magnetic resonance
imaging neurofeedback for major depressive disorder: A review. Psychiatry and clinical neurosciences 72, 466-481.
Young, K.D., Zotev, V., Phillips, R., Misaki, M., Yuan, H., Drevets, W.C., Bodurka, J., 2014. Real-time FMRI neurofeedback training of
amygdala activity in patients with major depressive disorder. PloS one 9, e88785.

58
Yuan, H., Young, K.D., Phillips, R., Zotev, V., Misaki, M., Bodurka, J., 2014. Resting-state functional connectivity modulation and
sustained changes after real-time functional magnetic resonance imaging neurofeedback training in depression. Brain connectivity 4,
690-701.
Zich, C., Johnstone, N., Lührs, M., Lisk, S., Haller, S.P., Lipp, A., Lau, J.Y., Kadosh, K.C., 2020. Modulatory effects of dynamic fMRI-
based neurofeedback on emotion regulation networks in adolescent females. NeuroImage 220, 117053.
Zotev, V., Krueger, F., Phillips, R., Alvarez, R.P., Simmons, W.K., Bellgowan, P., Drevets, W.C., Bodurka, J., 2011. Self-regulation of
amygdala activation using real-time fMRI neurofeedback. PloS one 6, e24522.
Zotev, V., Mayeli, A., Misaki, M., Bodurka, J., 2019. Emotion self-regulation training in major depressive disorder using simultaneous
real-time fMRI and EEG neurofeedback. arXiv preprint arXiv:1909.05764.
Zotev, V., Phillips, R., Young, K.D., Drevets, W.C., Bodurka, J., 2013. Prefrontal control of the amygdala during real-time fMRI
neurofeedback training of emotion regulation. PloS one 8, e79184.
Zotev, V., Yuan, H., Misaki, M., Phillips, R., Young, K.D., Feldner, M.T., Bodurka, J., 2016. Correlation between amygdala BOLD activity
and frontal EEG asymmetry during real-time fMRI neurofeedback training in patients with depression. NeuroImage: Clinical 11, 224-
238.

59