Neurofeedback

Neurofeedback

Neurofeedback is a specific form of biofeedback, which feeds back information about brain activity to allow for training of voluntary regulation of brain activity (Weiskopf et al., 2004b). At the beginning of the new millennium human neurofeedback was primarily based on recordings using electroencephalography (EEG). The use of a rather small number of electrodes (e.g., Birbaumer et al., 1999; Egner and Gruzelier, 2003; Hinterberger et al., 2004a) allowed for only unreliable localization of active brain areas and very limited access to deep subcortical brain areas. Even with modern multi-channel EEG systems the localization of the electric source is an intrinsically ill-posed problem (Baillet et al., 2001). This was likely the reason why the majority of studies focused on feedback of poorly localized frequency bands of the EEG such as slow cortical potentials (SCP; e.g., Birbaumer et al., 1999; Hinterberger et al., 2003; Kotchoubey et al., 2001), theta or alpha bands (e.g., Egner and Gruzelier, 2003; Leins et al., 2007). This type of EEG feedback was successfully used clinically, for example in epilepsy (Kotchoubey et al., 1997, 2001; Strehl et al., 2006), or as a BCI, enabling locked-in patients to communicate with the outside world (Birbaumer et al., 1999). It also proved to be a powerful tool for studying the relationship between brain activity and behavior. For example, it was shown that lateralized SCP regulation induced lateralized reaction time differences (Rockstroh et al., 1990).

1 Introduction

Neurofeedback is a form of operant conditioning where through immediate feedback the participant learns to self-regulate their neurophysiology. EEG-neurofeedback, arising as an offshoot of biofeedback, has a growing evidence base and now neurofeedback has found acceptance with other brain imaging modalities including fMRI feedback over the past decade (see for review DeCharms, 2008; Ruiz et al., 2013 and for a promising recent clinical implication see Linden et al., 2012), as well as near infrared spectrometry (NIRS; Mihara et al., 2012; Kober et al., 2013) and transcranial doppler sonography (Duschek et al., 2011). Additionally fMRI has been used to monitor the outcome of EEG-feedback (Levesque et al., 2006) with some basic neuroscience reports of EEG-neurofeedback now in conjunction with TMS and fMRI (Ros et al., 2010, 2013). EEG-neurofeedback applications are increasing exponentially both in the optimal performance field (see for review Gruzelier, 2014a) as well as in the clinic (Scott et al., 2005; Sterman, 2000; Gevensleben et al., 2009; Cortoos et al., 2009; Hartmann et al., 2013; Unterrainer et al., 2013), with attention deficit hyperactivity disorder (ADHD) the most popular clinical application (Monastra et al., 2005; Arns et al., 2009; Lofthouse et al., 2012). Developments in understanding the EEG in contemporary affective and cognitive neuroscience have led to innovative protocols including upper-alpha, gamma, frontal midline theta and so on, which have supplemented those drawn from the clinical domain of neurofeedback applications such as sensory-motor rhythm (SMR) and alpha/theta (A/T) training.

Research in the performing arts (Egner and Gruzelier, 2003; Gruzelier and Egner, 2004), along with ancillary neurophysiological studies with musicians (Egner and Gruzelier, 2001, 2004a,b), have played a part in stimulating this renewal of interest in EEG-neurofeedback (Stewart, 2002; Tilstone, 2003), one reason being that the order of magnitude of the enhancement in elite conservatoire music performance was of professional and pedagogic significance. Gains in music performance were associated in two studies with the elevation of theta (4–7 Hz) over alpha (8–12 Hz), termed the alpha/theta (A/T) protocol, and did not occur with faster-wave sensory-motor rhythm (SMR, 12–14 Hz) and beta1 (15–18 Hz) training which had, however, produced gains in the musicians’ attention (Egner and Gruzelier, 2001, 2004a,b). Musicality/Creativity was the main performance category enhanced by the A/T training, a protocol originally developed to enhance creativity but without evidence in support (Green and Green, 1977). The result with musicians has inspired a series of performing arts studies (Gruzelier, 2012), with the upshot that this protocol has provided the most replicable results to date, not only regarding artistic performance but within the optimal performance field in general. Gains following A/T training have been found with elite and novice musicians and dancers, seven in all, whereas SMR ratio training has had a beneficial effect mainly on novice artistic performance in five instances, putatively through an impact on lower-level abilities which become automatic in elite performance (Fitts and Posner, 1967).

Creativity, especially creative performance in the arts, will be the focus of this second part of a review on the efficacy and validation of EEG-neurofeedback for optimising function in healthy individuals. Part I of the review (Gruzelier, 2014a) dealt largely with laboratory assessments of cognition and affect. A third article, part III, will consider theoretical and methodological issues as a guide for future studies (Gruzelier, 2014a). A secondary focus of Part II stems from the perspective that assessment of artistic performance provides a window on creativity that has seldom entered the scientific arena and is one which provides ecological validity for the measurement of creativity, the lack of which is of cardinal concern to contemporary researchers on creativity, now briefly considered.

1.2 Why would alpha/theta neurofeedback enhance artistic performance?

Both alpha and theta activity have historically been implicated in the creative process.

1.2.1 Alpha activity and creativity

Whereas the full EEG spectrum from delta to gamma bands has been examined in approaches to understanding creativity (Dietrich and Kanso, 2010), it is alpha activity that has been both a persistent empirical focus (e.g. Martindale and Hines, 1975; Martindale et al., 1984; Jausovec, 2000; Fink and Neubauer, 2006; Grabner et al., 2007) and a theoretical focus as embedded in low arousal and diffuse attention theories of creativity (Mendelsohn, 1976; Martindale, 1999). Further Bazanova and colleagues are undertaking the most comprehensive analysis of features of alpha activity including maximum peak frequency, range width, degree of event-related desynchronisation, and characteristics of alpha spindles (see Bazanova and Vernon, in this issue), which disclose particular alpha band width correlates in relation to creativity assessed with Torrance test measures of fluency, originality and flexibility (Bazanova and Aftanas, 2008).

Alpha formed the basis of an early attempt to explore relations between achieving operant control of eyes-closed alpha power (7–13 Hz, O2-P4) and creativity. Martindale and Armstrong (1974) divided thirty students into high and low creative groups on the basis of the Remote Associates Test (Mednick and Mednick, 1967) and the Unusual Uses Test (Guilford et al., 1978) instructing them to keep a tone on so as to activate a mental state for as long as possible in a single session, and requiring alpha enhancement, which was followed by alpha suppression. The creative groups differed in the dynamics of learning. The high creative group displayed an immediate acquisition of alpha enhancement to 125% of baseline which was sustained without further improvement, and followed this by substantive suppression. The low creative subjects gradually reached the enhancement level of the high creative subjects, and then achieved less than half the suppression of the high creative group. The dynamics of operant control were interpreted as evidence of both a greater facility for focusing attention and flexibility in shifting cognitive strategies in creative individuals putatively underpinning creative thinking.

Alpha was also implicated in a more recent cognitive intervention study (N = 15) in which Fink et al. (2006) trained participants in divergent thinking techniques over two weeks (see also Benedek et al., 2006). When comparing a battery of divergent thinking tests before and after training, more ideas were rated as original following training when compared with a non-intervention control group, and contemporaneously there was more frontal alpha following training when compared with the control group.

1.2.2 Theta activity, hypnogogia and creativity

Historical interest in the EEG theta rhythm arose through its potential as an index of hypnogogia (Schachter, 1976), a state about which there has been a wealth of cultural historical documentation that the hypnogogic, reverie or twilight state between waking and sleeping and the converse has induced creative insights (Koestler, 1964, The Act of Creation). Famously the chemist Kekule, who in 1896 claimed to discover the benzene ring through an hypnogogic image of a snake biting its tail, came to become an advocate of hypnogogia ‘let us learn to dream gentlemen’. Cocteau conceived the entire scenarios for plays upon awaking. Edison adopted a technique of holding a ball in his hand to maintain the borderline state. While Mozart wrote:

“When I am, as it were, completely myself, entirely alone, and of good cheer-say, travelling in a carriage, or walking after a good meal, or during the night when I cannot sleep; it is on such occasions that my ideas flow best and most abundantly. Whence and how they come, I do not know; nor can I force them. Those ideas that please me I retain in memory and am accustomed, as I have been told, to hum them to myself. If I continue in this way, it soon occurs to me how I may turn this or that morsel to account, so as to make a good dish of it…. All this fires my soul, and, provided that I am not disturbed, my subject enlarges itself, becomes methodised and defined, and the whole, though it be long, stands almost complete and finished in my mind, so that I can survey it, like a fine picture or a beautiful statue, at a glance. Nor do I hear in my imagination the parts successively, but I hear them, as it were, all at once (gleich alles zusammen). What a delight this is cannot tell! All this inventing, this producing, takes place in a pleasing lively dream… (Holmes, 1912).

1.2.3 The alpha/theta neurofeedback protocol

Informed by evidence that with the progression towards hypnogogia and stage 1 sleep the theta amplitude predominates over alpha (Niedermeyer, 1999), the alpha/theta protocol was evolved to achieve hypnogogia in order to promote creativity and well-being (for review Gruzelier, 2009).

Elmer Green and colleagues first set about reinforcing theta and alpha activity through auditory feedback in a deeply relaxed eyes-closed state in order to facilitate creativity (Green et al., 1971; Green and Green, 1977). Hypnogogic phenomenology as described by Schachter (1976) was successfully achieved, especially when a predominance of theta over alpha was obtained. While month-long practice led to improved well-being and psychological integration with seemingly lasting psychotherapeutic benefit, there were no anecdotal benefits for creative insights.

Subsequently the potential of training alpha and theta for therapy was taken up in controlled studies as a primary part of a mixed modality package with army veterans having diagnoses of alcoholism in conjunction with anxiety/depression (Peniston and Kulkosky, 1989, 1990; Saxby and Peniston, 1995) and also with posttraumatic stress disorder (PTSD; Peniston and Kulkosky, 1991; Peniston et al., 1993). They inferred: “It is as though the patient was capable of integrating past traumatic experiences by coping with previously unresolved conflicts represented in the essential anxiety-free images and memories generated during the theta state of consciousness.” This led to formulation of the contemporary A/T protocol and its application with patients and healthy individuals for optimal performance. In drug addiction benefits of neurofeedback with A/T training have been subsequently reported in a trial with stimulant misusers in residential care (Scott et al., 2005), while well-being assessed with the Profile of Mood States (McNair et al., 1992) has been enhanced in withdrawn students in a controlled study (Raymond et al., 2005a). However, the A/T protocol has been especially efficacious in enhancing creative performance. Before reviewing this evidence A/T methodological studies will be considered.

1.2.4 The nature of alpha/theta learning

The trainee is taught to raise posterior theta (4–8 Hz) over alpha (8–12 Hz) amplitude with eyes closed while not falling asleep and with pleasant auditory reinforcement. This is unlike the more conventional neurofeedback training procedure which involves visual feedback on a screen. Typically on eye closure and relaxation the EEG displays high amplitude rhythmic alpha activity and with further deactivation alpha slowly subsides along with theta activity (see Fig. 1), until theta gradually becomes predominant and increases typically in conjunction with an increase in delta activity, although theta may increase independently of delta (Gruzelier et al., 2013a). The point when theta activity supersedes alpha activity, the stability of which is subject to individual differences, is called the theta-alpha “crossover”, which is commonly associated with alterations in consciousness leading to the onset of early sleep-stage I (e.g. Broughton and Hasan, 1995; Niedermeyer, 1999; De Gennaro et al., 2001). Successful progression is defined by an increase in the theta/alpha (t/a) ratio both within and across sessions. However, because of the brevity of sessions which optimal performance sessions in healthy individuals of necessity usually dictates, the everyday fluctuations in arousal state as participants present for training and which underpin their readiness to enter stage 1 sleep can make day to day across-session t/a ratio progression elusive. Typically the more reliable evidence is obtained from within- rather than across-session changes.

Fig. 1. Mean within-session theta and alpha amplitude (mv) showing theta crossover at 15 min; dancers tested in groups of six (Gruzelier et al., 2013b).

A series of methodological studies has been conducted on A/T training. With the aim of examining whether hypnogogic visualisations were promoted by A/T feedback Moore et al. (2000) contrasted 40-min sessions of training either posterior O2 alpha (8–12 Hz)/theta (4–8 Hz), alpha-only, or EMG feedback, all preceded by thermal biofeedback. Visualisations were found with all interventions and all produced A/T cross-over. Interpretation was compromised by poor compliance from the abstinent substance misuse outpatients (N = 35). This precluded meaningful group comparisons leading the authors to examine groups on the basis of pooled sessions rather than subjects, though the data disclosed more than twice the number of sessions in the A/T group, while the calculation of mean t/a ratios did not allow for the independent dynamics of alpha and theta wave production to be considered.

The first evidence of operant control of A/T training arose from an analysis of temporal dynamics where students were randomly assigned to Pz alpha (8–12 Hz)/theta (4–7 Hz) training or to noncontingent sham training consisting of the playback of another subject's session (Egner et al., 2002). Within two weeks five, 15-min sessions preceded by baseline were conducted, and the Thayer Activation/Deactivation Checklist (Thayer, 1967) was administered at the end of each session. In the A/T group there was a linear increase in t/a ratios, an increase not found in the control group. Across-session mean ratios were significantly higher in the contingent group on two/five sessions, underscoring the occurrence of across-session variability. The groups did not differ in their reduced Thayer activation, indicating firstly that the sham procedure was as relaxing as the experimental procedure, with no evidence of possible frustration because of noncontingency through the false feedback. Secondly it was inferred that the t/a short-term relaxation was not captured by the phenomenology of the broad descriptive activation/deactivation assessment, supported by the absence of significant correlation between the t/a ratios and the scales, as will be elucidated by the performing arts studies.

Subsequently temporal dynamics were further explored (Egner and Gruzelier, 2004b) by comparing frontal (Fz) with parietal (Pz) training in view of the broadly different theta correlates from posterior arousal and fronto-limbic theta systems. Also because of evidence that the longer-term outcome two weeks later in the resting EEG was at frontal sites following posterior A/T training, taking the form of reduced frontal beta1 and 2 (Egner and Gruzelier, 2004b). Furthermore there has been widespread current interest in the psychological significance of anterior theta measured from the frontal midline, though in the waking not the hypnogogic state (Inouye et al., 1994; Grunwald et al., 2001; Jensen and Tesche, 2002; Kubota et al., 2001; Missonnier and Deiber, 2006). The dynamics and learning with Pz and Fz training were found to differ. At the conventional posterior site reliable operant control was obtained and dynamics were in line with both deactivation and the signature of the wakefulness-to-sleep transition (De Gennaro et al., 2001), namely a lesser decrease in theta than alpha activity underpinning the increase in the t/a ratio. At the frontal midline site there was an absence of operant control while increments in theta were relatively larger than increments in alpha unlike the waking to sleep transition. The results implicated different generators in the dynamics of A/T training at Pz and Fz. Accordingly A/T training has centred on posterior theta with its associations with lowered arousal.

1.2.5 A/t and cognitive creativity

In the first controlled attempt to examine the outcome of the A/T protocol on cognitive creativity Boynton (2001) examined the impact of eight, twenty-min, weekly sessions of training with participants (N = 62) in groups of two to six requiring them to either increase Pz theta (4–8 Hz) over alpha (8–12 Hz) amplitude or to relax with eyes-closed. Both were supplemented by pre-training lectures, post-session discussions and background music. Cognitive creativity was measured with the Torrance (1974) and Guilford et al. (1978) tests and well-being with the Friedman (1994) scale. There was improvement in creativity and well-being without preferential group changes. No EEG data were reported.

Doppelmayr and Weber (2011) with a primary interest in visuospatial rotation examined the impact of thirty sessions of SMR (C3,4) ratio training compared with theta/beta1 (4.5–7.5/15–21 Hz) training and a control consisting of random 1 Hz bins in the range 6–35 Hz. A battery of cognitive measures included the Verbal Creativity Test (Schoppe, 1975) which provides a creativity index calculated from nine subtests, six of which are related to verbal tasks such as inventing new names, and the other three are titled Unusual Applications, Utopic Situations, and Inventing Nicknames. While SMR training benefitted mental rotation, none of the groups showed improvement in creativity.

An increase in cognitive creativity was found in a study with young contemporary dancers who were randomised to one of four groups, either A/T or heart rate variability (HRV) training, a dance theory comparison group or a non-intervention control group (Gruzelier et al., 2013b). The results on dance performance are outlined in Section 2.3. Following A/T training there was an increase in expressive creativity on the Guilford test (1978) when compared with the comparison groups, see Fig. 2. No changes were found with the Insight test (Dow and Mayer, 2004).

Fig. 2. Mean Guilford divergent thinking change scores for A/T, HRV, choreology and non-intervention control groups (N = 45) (Gruzelier et al., 2013b).

Abstract

Neurofeedback is an increasingly recognized therapeutic option in various neuropsychiatric disorders to treat dysfunctions in cognitive control as well as disorder-specific symptoms. In this review we propose that neurofeedback may also reflect a valuable therapeutic option to treat executive control functions in Gilles-de-la-Tourette syndrome (GTS). Deficits in executive control functions when ADHD symptoms appear in GTS likely reflect pathophysiological processes in cortico-thalamic-striatal circuits and may also underlie the motor symptoms in GTS. Such executive control deficits evident in comorbid GTS/ADHD depend on neurophysiological processes well-known to be modifiable by neurofeedback. However, so far efforts to use neurofeedback to treat cognitive dysfunctions are scarce. We outline why neurofeedback should be considered a promising treatment option, what forms of neurofeedback may prove to be most effective and how neurofeedback may be implemented in existing intervention strategies to treat comorbid GTS/ADHD and associated dysfunctions in cognitive control. As cognitive control deficits in GTS mostly appear in comorbid GTS/ADHD, neurofeedback may be most useful in this frequent combination of disorders.

Abstract

Neurofeedback is a promising approach for non-invasive modulation of human brain activity with applications for treatment of mental disorders and enhancement of brain performance. Neurofeedback techniques are commonly based on either electroencephalography (EEG) or real-time functional magnetic resonance imaging (rtfMRI). Advances in simultaneous EEG–fMRI have made it possible to combine the two approaches. Here we report the first implementation of simultaneous multimodal rtfMRI and EEG neurofeedback (rtfMRI–EEG-nf). It is based on a novel system for real-time integration of simultaneous rtfMRI and EEG data streams. We applied the rtfMRI–EEG-nf to training of emotional self-regulation in healthy subjects performing a positive emotion induction task based on retrieval of happy autobiographical memories. The participants were able to simultaneously regulate their BOLD fMRI activation in the left amygdala and frontal EEG power asymmetry in the high-beta band using the rtfMRI − EEG-nf. Our proof-of-concept results demonstrate the feasibility of simultaneous self-regulation of both hemodynamic (rtfMRI) and electrophysiological (EEG) activities of the human brain. They suggest potential applications of rtfMRI–EEG-nf in the development of novel cognitive neuroscience research paradigms and enhanced cognitive therapeutic approaches for major neuropsychiatric disorders, particularly depression.

2.2.3 Neurofeedback

Neurofeedback is a self-regulation technique in which real-time parameters of brain activity are presented to the subject through visual, auditory, or tactile modality, while the subject is supposed to voluntarily or involuntarily alter these parameters to reach a desired state of brain functioning (Fig. 7.6) [86]. It allows people to self-regulate their cognitive and emotional responses and retrain brain function through the continuous access to contingent data between their actions and their brain signals.

Fig. 7.6. Neurofeedback diagram illustrating the process by which EEG signals are acquired, amplified, processed, and shown back to the subject.

(From: Cadabam's Hospitals.)

One of the registration techniques that can be used for neurofeedback is EEG, which is used to monitor the electrophysiological brain waves, while a computer-based program is used to analyze and show back to the subject their brainwave activity through sound or visual elements. Those elements help subjects self-regulate their cognitive and emotional state, aiming at modulating these brain signals. This process is aimed at enabling subjects to improve their cognitive state and brain function.

Neurofeedback has been explored as a therapy for several psychiatric and neurocognitive disorders. However, the demonstration of robust clinical effects remains a major hurdle. Although numerous studies have demonstrated that neurofeedback may alleviate symptoms of different neurological and mental health conditions [87], others have had mixed results and have been affected by differences in study design, difficulties in distinguishing responders from nonresponders and the scarcity of homogenous patient populations [88]. Psychological factors, including differential influence of feedback, reward, and experimental instructions and locus of control, also need to be further investigated to better understand the heterogeneous clinical effects of neurofeedback [88].

There are many different types of neurofeedback. The most frequently used form of neurofeedback is frequency/power neurofeedback. This technique has been found to be particularly useful for ADHD, anxiety, and insomnia [89]. Another commonly used type of neurofeedback is low-energy neurofeedback system (LENS), which delivers a weak electromagnetic signal to carry feedback to the person receiving it [90]. This form of neurofeedback is commonly used to treat traumatic brain injury, ADHD, insomnia, fibromyalgia, restless legs syndrome, anxiety, depression, and anger.

Despite the existing research of neurofeedback in neurological and psychiatric disorders, understanding its potential sex differences remains in infancy stages. An early study by Kovacevic et al. examined collective neurofeedback in an immersive art environment and found that females showed more power in beta and gamma ranges [91]. Another study by Wood and Kober investigated the effect of sex of participant and experimenter on the outcomes of neurofeedback training [92]. Two key findings emerged: (1) Female participants trained by female experimenters were not able to upregulate the sensorimotor rhythm of EEG (12–15 Hz, referred to as SMR) during a single session of neurofeedback training, whereas both male and female participants were able to upregulate SMR when instructed by male experimenters; and (2) a strong positive correlation between training outcomes and locus of control in dealing with technology was observed only in the female participants trained by female experimenters. These findings suggest that sex differences and locus of control in dealing with technology differences may implicate neurofeedback training outcomes.

Moreno-Garcia et al. examined the efficacy of three well-known alternative therapies for ADHD (neurofeedback, pharmacological treatment, and behavioral therapy), each administered alone with no overlapping treatment [93]. Using multilevel models, it was found that regardless of the treatment received, sex helped explain between variation in EEG measures and changes. Furthermore, a metaanalysis of nonpharmacological treatments for ADHD showed the effectiveness of neurofeedback, especially in girls [94]. However, data differ according to used effectiveness measures. These findings suggest that personalized applications of neurofeedback should be further explored, and that the efficacy of neurofeedback treatment in relation to both sexes calls for more investigation.

As the above studies suggest, sex-related differences and psychosocial factors related to gender may impact neurofeedback training outcomes. These differences include females showing more power in beta and gamma ranges of neurofeedback, differences between sexes in locus of control and outcomes of neurofeedback training, and sex being a predictive factor for the outcomes of neurofeedback in ADHD [91–94]. Therefore, there is a clear need for these factors to be considered and documented in future neurofeedback studies.

6.5.2 Neurofeedback

Neurofeedback is defined as learning to control brain events by giving sensory (conscious) feedback, contingent on some brain event. In the animal literature it is often described as ‘operant conditioning’, which is really an experimental procedure rather than an explanatory framework. Neurofeedback is one kind of biofeedback training, which also includes training of peripheral events like increasing the warmth of a finger. Simply by holding a thermometer showing small changes in finger temperature one can learn to increase peripheral warmth, for example, which seems to help people relax and lower blood pressure.

Biofeedback studies of animals, using operant conditioning, have shown positive results over several decades. Animals with implanted electrodes can be trained to increase the activity of single neurons, and of larger populations of neurons. When justified for medical reasons similar training effects can be shown in humans. It is important just to stop and consider what a remarkable result that is. In a brain with billions of neurons, simple sensory feedback can allow training of voluntary control over otherwise entirely involuntary neuronal firing and other brain events.

However, neurofeedback training for specific brain oscillations, for example, may not result in enhancement of the targeted event. Surprisingly, even ‘alpha neurofeeback training’, for example, does not necessarily result in proven increments in alpha wave activity. This is a disconnect in the literature, which has mostly focused on testing practical outcomes. Oddly enough, the ‘feedback’ aspect of neurofeedback is sometimes lost.

This becomes less of a paradox if we consider that brain rhythms are under precise homeostatic control (John, 2002). If neurofeedback works, it is not just flipping a light switch in the brain; rather, we are jumping up and down on a large rubber boat floating on a lake. Our jumping causes real changes, but the brain still stays afloat – which is a very good thing. The brain is complex, dynamic, and multistable. That is how living systems can survive over many kinds altered environments.

1 Introduction

Neurofeedback refers to a self-regulation technique that provides the individual with feedback about specific brain activity in connection with a related behavior. The underlying assumption at the core of this practice posits that through this type of feedback one can entrain, change, and regulate neural activity. This trend appeals to both researchers and practitioners, who wish to understand the neurobiological mechanisms as well as the therapeutic potential this approach may offer. Beyond electroencephalography (EEG), the advent of modern real-time brain imaging technology elucidates the time-course and location of brain activity and seems to open the road to new prospects, including the modulation of seemingly volitionless neural functions (Fig. 1). And yet, imaging-based neurofeedback has hardly transitioned from the cognitive neuroscience lab into the clinical trenches. In this paper we highlight the relative merits and current shortcomings of neurofeedback in the context of contemporary imaging technologies. We discuss how new modalities of brain imaging may provide a future trajectory to consider meaningful research resulting in potential inclusion in the clinical armamentarium.

Fig. 1. A conceptual diagram depicting rtfMRI-nf of the left primary motor cortex.

An evolutionary derivative of biofeedback, in the 1960s neurofeedback emerged to employ neural feedback via EEG (Kamiya, 2011). To this day, specialty clinics and private institutions continue to offer variations on EEG neurofeedback (EEG-nf) for an array of disorders and impairments, although this intervention has been largely dismissed as placebo-driven (see next section). Beyond EEG-nf, the advent of new technologies for imaging the living human brain has vastly expanded the scope of neurofeedback, which today includes more novel methods such as functional magnetic resonance imaging (fMRI), functional near infrared spectroscopy (fNIRS), and magnetoencephalography (MEG) (TIMELINE and Table 1). Thus, current-day neurofeedback draws on diverse imaging methods to help drive volitional control over electromagnetic and hemodynamic alterations in brain activity (Cannon, 2015; Hammond, 2011; Thibault, Lifshitz, Birbaumer, & Raz, 2015). Within each imaging modality, moreover, researchers have developed distinct neurofeedback protocols that target different brain signals and their concomitant physiological processes (Hammond, 2011; Sulzer, Haller, et al., 2013). Whereas proponents of neurofeedback sometimes lump together these diverse protocols, research findings support some imaging techniques more than others. In this paper we explore the potential merits and shortcomings of modern neurofeedback techniques and contextualize their place in the current technological landscape.

Timeline. A sketch of EEG-nf since its inception in the late 1950s. This visual representation captures the different organizations, journals, and events that shaped and fostered this technique. For over 40 years after the first neurofeedback experiment, EEG-nf dominated pertinent discussions. In 2003, however, real-time fMRI experiments sparked a new generation of research. References: 1. (Wyrwicka & Sterman, 1968), 2. (Sterman & Friar, 1972), 3. (Lubar & Shouse, 1976), 4. (Weiskopf et al., 2003), 5. (Lal et al., 2005), 6. (deCharms et al., 2005) 7. (Sitaram, Zhang, et al., 2007), 8. (Buch et al., 2008), 9. (Mihara et al., 2012), 10. (Sulzer, Haller, et al., 2013), 11. (Zotev, Phillips, Young, et al., 2013; Zotev, Phillips, Yuan, et al., 2013), 12. (Arnold et al., 2013), 13. (Ogrim & Hestad, 2013), 14. (Vollebregt et al., 2014), 15. (Koush et al., 2013), 16. (Florin et al., 2014), 17. (Okazaki et al., 2015), 18. (Marx et al., 2015), 19. (DeBettencourt, Cohen, Lee, Norman, & Turk-Browne, 2015).

Table 1. Advantages, shortcomings, and applications of neurofeedback imaging modalities.

Empty Cell	EEG	MEG	fMRI	fNIRS
Underlying Signal	Electrical activity from pyramidal cells perpendicular to the scalp (mainly gyri)	Magnetic fields produced by pyramidal cells perpendicular and tangential to the cortical surface	Blood oxygenation level dependent contrast (which indirectly relates with neuronal activity)	Volume of oxygenated and/or deoxygenated blood (which indirectly relates with neuronal activity)
Typical Feedback Signal Source	One central electrode or a multi-electrode cap	Sensors over sensorimotor cortex	Single brain regions, 3mm × 3mm voxels	Several sensors over sensorimotor cortex
Feedback delay	< 50 ms	< 50 ms	∼1.5 s (plus 4-6 s hemodynamic delay)	∼0.5 s (plus 4-6 s hemodynamic delay)
Resolution
temporal	Milliseconds	Milliseconds	Seconds	Seconds
spatial	Centimeters	∼10mm	Millimeters	Centimeters
depth	Superficial	Depth constrains interpolation accuracy	Deep (any region)	Superficial (<4 cm)
Portable	Yes	No	No	Yes
Cost (USD)
Initial set-up	500-50,000	2,000,000	500,000-2,000,000	50,000-300,000
Running costs	No extra fees	∼500/hour	∼500/hour	No extra fees
Relevant Literature	Plentiful	Emerging	Moderate	Emerging
Main Applications	Pediatric ADHD, epilepsy, various psychological disorders	Brain computer interfaces (Experimental)	Psychological conditions (chronic pain, depression, schizophrenia, etc.) (Experimental)	Brain computer interfaces, stroke rehabilitation (Experimental)

Introduction

Neurofeedback is a general methodological approach that uses various neuroimaging techniques to acquire real-time measures of brain activity and enable volitional self-regulation of brain function. The development of real-time functional magnetic resonance imaging (rtfMRI) (Cox et al., 1995), in which fMRI data processing and display keep up with MRI image acquisition, has made it possible to implement rtfMRI neurofeedback (e.g. deCharms, 2008; Sulzer et al., 2013; Weiskopf et al., 2004). rtfMRI neurofeedback (rtfMRI-nf) allows a subject inside an MRI scanner to watch and self-regulate blood-oxygenation-level-dependent (BOLD) fMRI activity in target region(s) of his/her own brain in what is experienced as real time. Studies performed over the past decade demonstrated the feasibility of rtfMRI-nf-based self-regulation of various localized brain regions, including the dorsal anterior cingulate cortex (Weiskopf et al., 2003), rostral anterior cingulate cortex (deCharms et al., 2005), auditory cortex (Yoo et al., 2006), anterior insular cortex (Caria et al., 2007; Ruiz et al., 2013), inferior frontal gyrus (Rota et al., 2009), supplementary motor area (Subramanian et al., 2011), subgenual anterior cingulate cortex (Hamilton et al., 2011), amygdala (Zotev et al., 2011), orbitofrontal cortex (Hampson et al., 2012), primary motor cortex (Berman et al., 2012), and others. Implementations of rtfMRI-nf for regulation of extended networks of brain areas defined using either functional localizers (e.g. Johnston et al., 2010; Linden et al., 2012) or support vector classification (LaConte, 2011; Sitaram et al., 2011) have also been reported.

In contrast to rtfMRI, which has temporal resolution equal to fMRI repetition time TR (order of a few seconds), electroencephalography (EEG) has millisecond temporal resolution and can record electrophysiological brain activity as it evolves in actual real time. EEG neurofeedback (EEG-nf) allows a subject to control certain characteristics of his/her own electrical brain activity as measured by EEG electrodes connected to the scalp. EEG-nf has a longer history and more reported applications to various patient populations than rtfMRI-nf. Some examples include: the sensorimotor rhythm (SMR) EEG-nf for treatment of epilepsy and seizure disorders (e.g. Sterman, 2000; Sterman and Friar, 1972); the SMR-theta and beta–theta EEG-nf for treatment of attention-deficit/hyperactivity disorder (e.g. Gevensleben et al., 2009; Levesque et al., 2006; Lubar and Lubar, 1984); the alpha–theta EEG-nf for treatment of substance use disorders (e.g. Peniston and Kulkosky, 1989; Sokhadze et al., 2008); the alpha–theta EEG-nf for deep relaxation (e.g. Egner et al., 2002) and creative performance enhancement (e.g. Gruzelier, 2009); the upper-alpha EEG-nf for cognitive enhancement (Hanslmayr et al., 2005; Zoefel et al., 2011); the frontal asymmetry EEG-nf for emotion regulation (Allen et al., 2001); and the high-beta EEG-nf for treatment of major depressive disorder (MDD) (Paquette et al., 2009).

The development and advances in simultaneous EEG–fMRI technique (e.g. Mulert and Lemieux, 2010), in which a subject wears an EEG cap inside an MRI scanner and EEG recordings are performed concurrently with fMRI data acquisition, have opened up new possibilities for neurofeedback research. Simultaneous EEG–fMRI provides the following important opportunities in the context of brain neuromodulation. First, electrophysiological correlates of rtfMRI-nf can be explored using EEG data recorded simultaneously with rtfMRI-nf training. Second, performance of EEG-nf can be validated based on fMRI data acquired simultaneously with EEG-nf training. Third, rtfMRI-nf can be dynamically modified using the simultaneously measured EEG activity. Finally, simultaneous multimodal rtfMRI–EEG neurofeedback can be provided to a subject to enable simultaneous self-regulation of both hemodynamic (rtfMRI) and electrophysiological (EEG) brain activities.

Here we report the first implementation of simultaneous multimodal rtfMRI–EEG neurofeedback (rtfMRI–EEG-nf) and its proof-of-concept application in training of emotional self-regulation. Our implementation of rtfMRI–EEG-nf is based on a novel, first-of-its-kind real-time integration of rtfMRI and EEG data streams for the purpose of brain neuromodulation.

During the experiment, healthy volunteers performed a positive emotion induction task by evoking happy autobiographical memories while simultaneously trying to regulate and raise two neurofeedback bars (rtfMRI-nf and EEG-nf) on the screen. The rtfMRI-nf was based on BOLD activation in a left amygdala region-of-interest (ROI), similar to our previous study of emotional self-regulation that used only rtfMRI-nf (Zotev et al., 2011). The EEG-nf, provided simultaneously with the rtfMRI-nf, was based on frontal hemispheric (left–right) EEG power asymmetry in the high-beta (beta3, 21–30 Hz) EEG frequency band.

Frontal EEG asymmetry is an important and widely used EEG characteristic of emotion and emotional reactivity (e.g. Davidson, 1992). It has been interpreted within the framework of the approach − withdrawal hypothesis (e.g. Davidson, 1992; Tomarken and Keener, 1998), which suggests that activation of the left frontal brain regions is associated with approach (i.e. higher responsivity to rewarding and positive stimuli), while activation of the right frontal regions is associated with withdrawal (i.e. tendency to avoid novel and potentially threatening stimuli). Brain activation is typically quantified by a reduction in alpha EEG power. The approach–withdrawal hypothesis applies to both emotional trait properties and emotional state changes in response to stimuli (e.g. Coan and Allen, 2004; Davidson et al., 1990; Sutton and Davidson, 1997). Numerous EEG studies have indicated that depression and anxiety are associated with reduced relative activation of the left frontal regions and increased relative activation of the right frontal regions (e.g. Thibodeau et al., 2006; Tomarken and Keener, 1998). Thus, frontal EEG power asymmetry is a natural target measure for EEG-nf aimed at training of emotional self-regulation, particularly in MDD patients.

Two studies have previously employed EEG-nf paradigms involving frontal EEG asymmetry. Allen et al. (2001) used EEG-nf based on the frontal EEG asymmetry in the alpha band for a group of healthy participants. They observed systematic changes in the asymmetry as the training progressed and associated changes in self-reported emotional responses. Paquette et al. (2009) applied EEG-nf based on EEG power in the high-beta band measured at two frontal and two temporal sites and used it in combination with psychotherapy sessions for a group of MDD patients. They reported a significant reduction in MDD symptoms associated with a significant decrease in high-beta EEG activity within the right frontal and limbic regions. This work followed up on the results of an earlier study (Pizzagalli et al., 2002) that demonstrated that MDD patients exhibited significantly higher resting EEG activity in the right frontal brain regions than healthy controls specifically in the high-beta band. The psychoneurotherapy (Paquette et al., 2009) led to significant changes in the high-beta EEG power asymmetry between the corresponding brain regions on the left and on the right.

In the present work, we implemented the EEG-nf based on the frontal EEG asymmetry in the high-beta band (21–30 Hz) rather than in the alpha band (8–13 Hz) because EEG–fMRI artifacts, caused by cardioballistic (CB) head motions as well as random head movements, are substantially reduced in this case. Also, electrophysiological activity in the high-beta band is relevant to depression, as mentioned above. The rtfMRI–EEG-nf was used in the present study for simultaneous upregulation of BOLD fMRI activation in the left amygdala ROI and frontal EEG power asymmetry in the high-beta band during the positive emotion induction task.

Neurofeedback in disease

Neurofeedback, a subset of biofeedback in which the subject is presented with feedback derived from their own brain signals in real time, is an established method with which to reconfigure brain networks. In neurofeedback, a subject is presented with features typically derived from a specific region of interest, and is asked to modulate the amplitude of those features (Fig. 3). For example, a specific frequency band of an EEG recording could be extracted from the electrode overlying a particular brain region, and the power of that signal could be displayed to the subject as a level on a thermometer or bar graph (Fig. 3C). Typically, the most frequently used neurofeedback modality has been EEG, however other modalities such as fMRI are recently beginning to be investigated due to their increased spatial resolution [9,42].

Leveraging neurofeedback via fMRI to treat neuropsychiatric diseases has a rich recent history and touches a wide variety of clinical symptoms [9,43,44]. For example, by learning to voluntarily suppress activity in dorsolateral prefrontal cortex (implicated in engagement) and the right insula (implicated in regulation), participants can successfully suppress a phobia of spiders [45]. Furthermore, learned voluntary control in order to increase regional activity in the supplementary motor area and the parahippocampal cortex (implicated in memory), lead to a decrease in motor reaction times and a decrease in memory for specific words, respectively [46].

More clinically, neurofeedback has been used to teach war veterans suffering from post-traumatic stress disorder (PTSD) to regulate activity in their amygdala (implicated in emotion) to produce a meaningful decrease in clinical symptomatology [47]. Additionally, a set of patients with post-herpetic neuralgia (chronic nerve pain) were taught via neurofeedback to regulate the regional activity of the rostral anterior cingulate cortex (thought to be involved in pain perception), to successfully decrease their level of perceived pain [48]. While many of these techniques have produced meaningful results, the current state of therapeutic neurofeedback mediated neuromodulation has focused on regional activity while remaining relatively agnostic to inter-regional connectivity. This focus places the field at a disadvantage because many neuropsychiatric diseases are thought to be disorders of connectivity [49–51]. Furthermore, those studies that do target inter-regional connectivity ([10,52–54] for a broad overview) do not establish mechanistically how the network changed to produce the observed results.

Historically, the decision of what exactly to feedback to the patient has largely been informed by results from behavioral neuroscience. For example, to treat depression, allow the subject to regulate the level of activity of areas associated with positive emotions [55]. To treat obesity, allow the subject to control the level of activity of areas associated with gustatory functions [56]. While this reasoning is intuitive, it lacks a mechanistic understanding of how neurofeedback is affecting the brain as a whole, challenging the optimization of delivery [57]. Without a mechanistic understanding of how the network needs to be altered to obtain a particular result, it remains challenging to knowingly limit the effects of the neurofeedback to the symptomatology being targeted, thereby minimizing undesired effects. Faced with the need to develop such an understanding, it is useful to turn to recent advances in other fields like network neuroscience and control theory, for a holistic, mechanistic, and dynamical view of the brain that can inform the optimization of these clinical interventions.

Introduction

Neurofeedback (NF) – also known as EEG biofeedback – is a type of biofeedback based on features of brain activity. The real-time feedback allows for particular patterns of brain activity to be ‘rewarded’ with some type of stimulation – visual, auditory or even tactile – and thereby to enable the user to learn how to modify certain aspects of his/her brain activity. This learning activity may include the ability to adjust the amplitude, frequency, coherence and other neurophysiologic metrics of brain activity. NF protocols have been shown to be effective treatments for many patients with a wide range of disorders, including: epilepsy (e.g. Kotchoubey et al., 2001; Walker and Kozlowski, 2005; Sterman and Egner, 2006; Tan et al., 2009; Strehl et al., 2014), attention deficit hyperactivity disorder (ADHD) (e.g. Thompson and Thompson, 2005; Leins et al., 2007; Arns et al., 2009; Sonuga-Barke et al., 2013), stroke (Doppelmayr et al., 2007; Rayegani et al., 2014), autistic spectrum disorder (ASD) (e.g. Kouijzer et al., 2009;Thompson and Thompson, 2009), emotional disorders (e.g. Raymond et al., 2005a,b; Othmer and Othmer, 2009; Reiter et al., 2016), and tinnitus (e.g. Schenk et al., 2003; Hartmann et al., 2014). In addition, NF has been tested as a means to enhance the cognitive abilities of healthy individuals (e.g. Hoedlmoser et al., 2008; Zoefel et al., 2011; Wang and Hsieh, 2013; Gruzelier, 2014a,b).

Although NF has been found to be useful in the rehabilitation and treatment of different disorders, a significant proportion of participants does not benefit from the NF treatment and fails to achieve the requisite control over brain metrics, even after multiple training sessions (e.g., Hanslmayr et al., 2005; Doehnert et al., 2008; Zoefel et al., 2011; Enriquez-Geppert et al., 2014). In this paper, we review both the problem of NF inefficacy and the existing predictors for successful NF learning. We focus here on EEG NF and not on fMRI NF: these two techniques may share a similar inefficacy problem, but they are based on fundamentally different technologies and involve different procedures. The first section of the review presents the ‘inefficacy problem’ and its implications for NF and brain–computer interface (BCI) applications. In the second section, we summarize the literature on different predictors for success in modifying brain activity using NF or BCI applications. In the third section, we suggest new directions for research that are based on personalized NF. In the fourth and final section, we broaden the objectives of this review by presenting recommendations that we believe should be considered in future studies.