Predictors of Successful Modulation of Emotion-Related Networks in a Study of Real-Time f-MRI Neurofeedback with

Adolescent Females

By

Graham Staunton

Submitted for the Degree of

Doctor of Psychology(Clinical Psychology)

School of PsychologyFaculty of Health and Medical Sciences

University of SurreySeptember 2019

© Graham Paul Staunton 2019

Abstract of Empirical Paper

Neurofeedback using real-time functional magnetic resonance imaging is a technique which

supports participants to modify their own brain activity for therapeutic benefit. It has gained

attention for its potential as an alternative intervention for a range of mental health issues.

However, there is considerable variance in participant success. Understanding the factors

involved in this variance could assist with the prediction of outcomes and personalised

support. Here, the results of a literature review on this topic and a subsequent secondary data

analysis for a therapeutically significant population are presented. Specifically, in Zich and

colleagues (under review), adolescent female participants (n = 28) learned to modulate

healthy patterns of emotional regulation in a study which collected pre-intervention data on

variables to do with mood, IQ, development, initial functional connectivity (IFC) and control

beliefs. The current analysis aimed to predict outcomes based on these measures. Differences

between participants who improved (learners) and declined (non-learners) in performance

were also explored. Principal components analysis reduced the measures into the factors

‘Control’ (mood and control beliefs) and ‘Maturity’, (age, puberty and IQ). Multiple

regression analysis with these factors and IFC did not significantly predict performance or

change. However, IFC significantly predicted performance (β =-415.75, p = .004, CI = -

415.75, -92.37). At a group level, non-learners initially performed above the mean but could

not sustain this success. Learners had significantly higher thought control ability and verbal

comprehension scores than non-learners. These results and other findings from the literature

require confirmation with larger samples. It was concluded that predicting outcomes for

individuals may be challenging as individual factors are likely to have complex roles, though

initial connectivity readings may predict performance effectively. Implications include the

potential for personalised support to improve outcomes for non-regulators. Individual factors

should be routinely considered to develop the efficacy of neurofeedback.

Acknowledgements

I would like to express my gratitude to my research supervisor, Dr Kathrin Cohen Kadosh, for her guidance and support for me in producing this work. I would also like to thank the

other researchers who were involved in carrying out the neurofeedback training on which my analysis is based: Catharina Zich, Simone Haller, Michael Lührs, Stephen Lisk and Jennifer Lau. I am very grateful to Dr Peter Williams and Dr Peter Hilpert for their expertise on the statistical analysis. Finally, I would like to thank all the course staff who taught us over the three years and the many members of Cohort 45, family and friends who have helped me

through this process.

List of Publications

Parts of this thesis have been published in:

Staunton, G. & Cohen Kadosh, K (2019). A systematic review of the psychological factors that influence neurofeedback learning outcomes. NeuroImage, 185, 545-555.

ContentsPage

Part 1 – Research – Empirical Paper 6

List of Appendices to the Empirical Paper 60

Appendices to the Empirical Paper 61

Part 2 – Research – Literature Review 80

Part 3 – Clinical Experience 131

Part 4 - Assessments 134

Part One: Research

PREDICTORS OF SUCCESSFUL MODULATION OF EMOTION-

RELATED NETWORKS IN A STUDY OF REAL-TIME FMRI

NEUROFEEDBACK WITH ADOLESCENT FEMALES

Word Count: 9966

Statement of Journal Choice: This work is intended for publishing by Neuroimage. Statement

of Journal Choice: Neuroimage had an impact factor within the top 6.5% of journals in 2017.

A substantial quantity of fMRI neurofeedback studies are published in this journal.

School of PsychologyFaculty of Health and Medical Sciences

University of SurreyApril 2019

Abstract

Neurofeedback using real-time functional magnetic resonance imaging is a technique which

supports participants to modify their own brain activity for therapeutic benefit. It has gained

attention for its potential as an alternative intervention for a range of mental health issues.

However, there is considerable variance in participant success. Understanding the factors

involved in this variance could assist with the prediction of outcomes and personalised

support. Here, the results of a literature review on this topic and a subsequent secondary data

analysis for a therapeutically significant population are presented. Specifically, in Zich and

colleagues (under review), adolescent female participants (n = 28) learned to modulate

healthy patterns of emotional regulation in a study which collected pre-intervention data on

variables to do with mood, IQ, development, initial functional connectivity (IFC) and control

beliefs. The current analysis aimed to predict outcomes based on these measures. Differences

between participants who improved (learners) and declined (non-learners) in performance

were also explored. Principal components analysis reduced the measures into the factors

‘Control’ (mood and control beliefs) and ‘Maturity’, (age, puberty and IQ). Multiple

regression analysis with these factors and IFC did not significantly predict performance or

change. However, IFC significantly predicted performance (β =-415.75, p = .004, CI = -

415.75, -92.37). At a group level, non-learners initially performed above the mean but could

not sustain this success. Learners had significantly higher thought control ability and verbal

comprehension scores than non-learners. These results and other findings from the literature

require confirmation with larger samples. It was concluded that predicting outcomes for

individuals may be challenging as individual factors are likely to have complex roles, though

initial connectivity readings may predict performance effectively. Implications include the

potential for personalised support to improve outcomes for non-regulators. Individual factors

should be routinely considered to develop the efficacy of neurofeedback.

Introduction

Neurofeedback has developed in more recent decades from a strong tradition of biofeedback

approaches. The approach aims to train participants in the self-modulation of targeted neural

networks based on the feedback of real-time signals from their own brain (Johnston et al.,

2010, Scharnowski & Weiskopf, 2015; Sitaram et al., 2016). For example, activity in a region

of interest may be directly linked to a digital display of a thermometer. Increasing activity in

the region of interest increases the displayed temperature of the thermometer, showing the

participant that they have achieved (or not) the desired result. In this way neurofeedback can

support people to develop and control beneficial patterns of brain activity in its clinical

applications (e.g. Marzbani, Marateb & Mansourian, 2016). As clinical disorders have a

bioneurological basis, neurofeedback has significant therapeutic potential for a range of

neurological and psychological conditions as an alternative to talking therapies and

pharmacological interventions (Niv, 2013). However, not everyone is able to benefit from

neurofeedback and the reasons for this are not fully understood. It is the aim of the current

study to consider the participant variables that influence how people respond to

neurofeedback. This allows the prediction of neurofeedback outcomes, the development of

ways to improve the efficacy of neurofeedback and also helps us to understand who is most

likely to benefit from such a technique. Here, an overview of different types of

neurofeedback, currently understood learning mechanisms and the therapeutic potential of the

technique is provided. This is followed by a brief summary of a systematic literature review

on the psychological factors linked to neurofeedback outcomes (Staunton & Cohen Kadosh,

2019) and an overview of research into other types of predictor variables. A critique of this

literature is given, followed by the current theoretical understanding of the mechanisms that

may be involved in neurofeedback. Finally, to add to this literature, a secondary data analysis

is reported which explores variables which contribute to neurofeedback training outcomes in

a recent training protocol for emotional regulation with adolescent females (Zich et al., under

review); a population of particular relevance to interventions for emotional processing.

Types of Neurofeedback

Of the various types of neurofeedback used in recent decades, real-time functional magnetic

resonance imaging (fMRI) represents a particularly promising approach. This is due to the

high spatial resolution (~2mm) and whole-brain coverage of fMRI, which allows the analysis

of patterns of functional activity in deep sub-cortical structures (Sorger, Reithler, Dahmen, &

Goebel, 2012). The method of fMRI works by detecting the concentration of oxygenated and

deoxygenated haemoglobin in the neural vasculature, which is reflective of the metabolic

demands of the underlying neural activation (Attwell & Ladecola, 2002). This relatively slow

response of blood oxygenation to neuronal activity is the haemodynamic response function

(HRF) and peaks at about 5 seconds after stimulus onset (Sitaram et al., 2016).

The most commonly used alternative to fMRI is real-time electroencephalography (EEG); the

advantage of which is that it requires fewer resources and has higher temporal resolution

(~1ms) than fMRI. However, non-invasive EEG has lower spatial resolution (~1-5cm;

Sitaram et al., 2016) and lacks the whole-brain coverage of fMRI. In real-time EEG, brain

signals are transformed into frequencies (e.g., delta (0–4 Hz), theta (4–7 Hz), alpha (8–12

Hz), beta (12–30 Hz) and gamma (>30 Hz) bands) and then key features of the signals are

extracted (Nicolas-Alonso & Gomez-Gil, 2012). These features can be analysed to interpret

short-term and long-term response patterns in the signal called event-related potentials

(Nicolas-Alonso & Gomez-Gil, 2012). EEG is often used with a methodology closely related

to neurofeedback called brain-computer interface (BCI). BCI setups intend for participants to

use brain activity in direct communication or control with an external device such as

prosthetic assistive technology for people who are physically disabled (Hamedi, Salleh &

Noor, 2016). EEG BCI setups are most commonly based on changes in EEG signal

oscillations called sensorimotor rhythms (SMR; Wierzgaaa et al., 2018) although some use an

event related potential involved in decision making called P300 (Sellers, Arbel & Donchin,

2012).

Mechanisms of Neurofeedback

In order to understand which variables may contribute to outcomes of neurofeedback, the

mechanisms behind self-modulation through feedback need to be considered. Many of the

mechanisms discussed in the literature originated in early biofeedback research; such as in

studies where participants learn to influence their heart rate.

Learning from feedback has been conceptualised as a form of skill-learning (Lang, 1975) as

in motor processes in a similar way to learning for example to play a sport or a musical

instrument. DeCharms (2008) suggests that brain activity is usually implicit and automatic,

but that neurofeedback allows explicit control over brain regions which can then be applied to

everyday life. This theory competes with behaviourist theory; namely that participants learn

though the process of operant conditioning (Skinner, 1945). This theory suggests that positive

feedback will reinforce a desired behaviour due to the positive associations with behaviour

and consequence. In the case of neurofeedback, the desired behaviour is the modulation of

brain activity. This then translates into theories of Hebbian connections forming between

neural networks which ‘fire together’ and subsequently ‘wire together’ through long-term

potentiation (Hebb, 1941). There is evidence of this mechanism being applied successfully in

neurofeedback. For example, Chiew and colleagues (2012) attempted to set-up operant

conditioning with the use of ‘points’ as a reward for reaching a threshold level of

performance in their fMRI neurofeedback study on self-modulating motor area activity. This

threshold was continuously adjusted so that better performance increased the threshold and

poor performance decreased the threshold. However, they found that the difference in the

number of rewards given to both good and poor right-handed performers was present from

the first run, before the learning effect had time to take place. They concluded that success in

neurofeedback may involve attentional abilities as well as operant learning processes.

Therefore, the theory of operant conditioning may struggle to explain performance in

addition to learning. The theory also lacks detail to explain the process by which different

methods for modulating brain activity are cognitively selected by individuals.

A theory which offers explanation to these selection processes is Dual-process theory

(Lacroix, 1981). It posits that there are two separate processes in biofeedback learning. These

are ‘efferent’ processes where the individual attempts to manipulate the stimulus (e.g.

thermometer) and ‘afferent’ processes where the individual allows the stimulus to help

discriminate which responses are relevant. The theory suggests that it is fundamentally the

efferent processes which lead to learning in most circumstances. This might explain more

recent findings that those that actively experiment perform better than those that are more

hesitant and reflective (Jeunet et al., 2015). Initially, acquisition comes from efferent

processes of identifying useful strategies from a range of strategies that are already available.

However, if there is no useful strategy available then afferent processes may be used instead.

The strategies that are available and their selection will depend on the individual, their history

and the context of the experiment (e.g. what strategies are provided to the individual). The

strategies associated with positive feedback are attended to by the individual more often. This

array of associated strategies can then be continually refined over time; perhaps even into

smaller, more specific parts of a strategy.

It is unclear from the evidence whether these less than recent theories generalise in a

straightforward way from biofeedback to neurofeedback of all types. Evidence-based theory

behind mechanisms involved in neurofeedback performance and learning therefore seems to

be an area of need in the literature.

The Therapeutic Potential of Neurofeedback and its application in Emotional Processing

The field of neurofeedback has tremendous potential for its contributions to neurology,

psychiatry and clinical psychology. The advances of fMRI and other modern imaging

techniques mean that mental health disorders can be better traced to their neurological

underpinnings by highlighting key neural networks (Linden et al., 2012). This can aid in the

development of ‘endophenotypes’, or neurological biomarkers for diagnosis such as in

depression (e.g. Lenox, 2008) or schizophrenia (e.g. Barnett & Fletcher, 2008). It also means

that deficits in these systems can be better understood and much needed mechanism-based

therapeutic interventions can be developed (Linden, 2006).

Even at this early stage in its development, neurofeedback has already been utilised to modify

neural mechanisms in clinical populations. Positive effects of fMRI neurofeedback training

have been demonstrated in schizophrenia (e.g. Ruiz et al., 2013) depression (e.g. Linden et

al., 2012), obesity (e.g. Spetter et al., 2017), chronic pain (e.g. deCharms et al., 2005)

Parkinson’s (e.g. Subramanian et al., 2011) and post-traumatic stress disorder (e.g. Zotev et

al., 2018). The neural changes from fMRI-based neurofeedback have been shown to be

detectable for up to 14 months post training (Megumi et al., 2015, deCharms et al.2005;

Robineau et al., 2017) and can lead to associated behavioural changes (Caria, Sitaram, &

Birbaumer, 2012). One key area in which neurofeedback has shown great potential for

preventing and intervening with mental health problems is in processing and self-regulation

of emotion.

Emotional regulation can be defined as the cognitive processes influencing the intensity and

duration of emotions (Gratz & Roemer, 2004). A metanalysis by Aldao, Nolen-Hoeksema &

Schweizer (2010) found deficits in emotional regulation across many mental health disorders

which in turn lead to maladaptive strategies (such as avoidance or rumination) to cope in the

short-term. However, in the long-term this leads to further dysregulation and greater distress,

which is one of the reasons emotional regulation is often targeted in therapeutic interventions

(Gratz & Roemer, 2004).

On a neurological level, emotional regulation involves two key regions; the Amygdala and

the Dorso-lateral prefrontal cortex (DLPFC; Paret et al., 2016). The pattern of functional

connectivity (i.e. the correlated activation of two neural substrates in haemodynamic

modalities) between these areas is considered to be key for healthy emotional regulation

(Paret et al., 2016). This pattern develops from childhood to adolescence. Specifically, it has

been found that with children under the age of ten, there is a positive relationship in the

activity between the DLPFC and amygdala which then shifts in adolescence (Gee et al.,

2013). Healthy mature brains exhibit a negative relationship between the areas where DLPFC

activity down regulates the amygdala and vice versa (Gee et al., 2013). Development of this

pattern is critical for positive mental health and adjustment in adolescents (e.g. Chervonsky &

Hunt, 2018). As some studies suggest 50% of mental health problems develop by age 14

(Kessler et al., 2005) neurofeedback may offer a valuable early intervention to support these

healthy patterns.

Other interventions available from health services have been shown to be effective for mental

health issues, but they also have considerable limitations. Pharmacological and psychological

interventions are in the broadest sense found to be effective in roughly half of those treated

(Laynard, Clark, Knapp & Mayraz, 2007). Pharmacological interventions are often hindered

by issues with non-compliance (Bambauer, et al., 2007; Goethe et al., 2007). Psychological

therapy can often be a time-consuming long-term intervention and there may be barriers to

engagement. For example, some consider themselves (or are consider by others) as

incompatible with intense emotional dialogue; such as men, older adults, people of certain

cultures, people with impaired cognition and those with social anxieties (e.g. Di Bona et al.,

2014, Royal College of Psychiatrists, 2004). Understanding of the neurological mechanisms

of these interventions is still based on preliminary evidence but they may be influencing the

healthy functional connectivity patterns that neurofeedback can target directly (Barsaglini et

al., 2014). Neurofeedback therefore offers a useful alternative which has the potential to be

relatively time efficient, long lasting and directly mechanism based. There is growing

evidence that neurofeedback can be used to both up-regulate the DLPFC and also to down

regulate the amygdala in adults (Johnston et al., 2010; Paret et al., 2014; Sarkheil et al., 2015;

Zotev et al., 2011; Zotev, Phillips, Young, Drevets & Bodurka, 2013; Zotev, Yuan, Misaki &

Bodurke, 2014; Zotev et al., 2018) and also children and adolescents (Cohen Kadosh et al.,

2016).

The Inefficiency Problem

Despite the early overall successes of neurofeedback, there are some participants who

struggle to respond to neurofeedback and are often referred to in the literature as non-

responders, non-learners, non-performers or non-regulators and represent 30-50% of the

population (Alkoby, Abu-Rmileh, Shriki, & Todder, 2017). This is not unique to

neurofeedback as the phenomenon of ‘BCI Illiteracy’ in which BCI is unsuccessful for 10-

30% of participants (e.g. Vidaurre & Blankertz, 2010) is also acknowledged in the literature.

The issue more generally in neurofeedback has been referred to as the “Inefficiency problem”

(Alkoby et al., 2017). In order to better understand the effects of neurofeedback and to

increase its efficiency as an option for intervention, the individual differences and contextual

factors that lead to variations in success need to be addressed.

Predicting the variance from the results of neurofeedback is no easy task. There are a

multitude of potentially influential variables that range from mood and motivational factors,

strategies and styles of learning (e.g. active vs reflective) or neurophysiological factors such

as resting brain region activity levels (Diaz Hernandez, Rieger, & Koenig, 2016). Variables

can be further classified as being inter or intrapersonal in that they may differ between

participants or within the same participant over time respectively (Grosse-Wentrup &

Schölkopf, 2013). Variables may be fluctuating states or stable traits (Jeunet, N’Kaoua, &

Lotte, 2016) and explanatory or correlational variables (Grosse-Wentrup & Schölkopf, 2013).

The distinction of these variables is necessary due to the complexity of their contribution.

Previous evidence has found that participant performance can vary greatly within the space of

minutes (e.g. Grosse-Wentrup & Schölkopf, 2013) along with psychological states (e.g.

Nijboer Birbaumer & Kubler, 2010).

In addition to the complexity of predictors, there is also variation in how researchers are

defining successful responses to neurofeedback. This is partly due to the variety of aims and

methodologies in this literature; which may offer some cross-methodological validity, but

also bring into question whether non-responders in one study are representative of those in

another. One important distinction is the difference between performance and learning; which

may be separate skills. For example, some studies may value the generation of as strong an

alpha signal as possible at any stage in EEG neurofeedback (e.g. Nan et al., 2012), while

some may emphasise learning and improvement of a skill in question within and between

sessions (e.g. Diaz Hernandez et al., 2016). Despite these challenges, there is an emerging

foundation of knowledge implicating certain groups of variables as relevant which include

psychological variables, learning dynamics and physiological variables.

Psychological Variables that Contribute to Neurofeedback Performance and Learning: A

Review of the Literature

For psychological variables, a systematic review of the literature was conducted to find

variables that have been implicated in neurofeedback and BCI literature (Staunton & Cohen

Kadosh, 2019). Psychological variables are of particular interest as there may be more

potential to intervene with psychological issues such as state anxiety as opposed to

physiological factors such as grey matter volume or age (Grosse-Wentrup & Schölkopf,

2013). This review utilised the PRISMA methodology (Figure 1; Moher, Liberati, Tetzlaff,

Altman, & PRISMA group, 2009).

Of 270 papers investigated, 20 studies were found which directly explored psychological

variables that contribute to the outcomes of neurofeedback. BCI literature was included as it

is closely related to neurofeedback in methodology. The variety of methodologies and

measurement techniques found in the literature at this stage led to difficulties in using

quantitative analyses effectively (Thibault et al., 2018) and so a qualitative review was

undertaken. Research was found to implicate attention, motivation, mood and a variety of

other personality or task specific factors to influence outcomes from neurofeedback.

Attention: Attention was influential in all studies which measured this as a predictor of

neurofeedback outcomes (Chiew, Laconte & Graham, 2012; Daum et al., 1993; Gruzelier et

al., 1993; Hammer et al., 2012). This includes both attention span (Daum et al., 1993) and

also simple attentional abilities as measured with performance tests from the Wechsler Scale

20 full text articles included in qualitative synthesis

33 fMRI full text articles excluded as no

inclusion/mention of psychological predictors

54 full text articles assessed for eligibility

217 records excluded270 abstracts screened

270 records after duplicates removed

10 records identified through other sources

271 records identified through database searching

Figure 1: Flowchart of the systematic review following the procedure set-out in (Moher, Liberati, Tetzlaff,

Altman, & PRISMA group, 2009)

of Intelligence (Wechsler, 1981) or the Cognitrone (Schuhfried, 2007). For example,

Hammer and colleagues (2012) found ability to concentrate as measured by the Cognitrone

accounted for 19% of the variance in correct responses on a kinaesthetic imagery task in an

SMR BCI set-up. The effect of attention is also inferred from the effect of fatigue on

performance (Gruzelier et al., 1993) and from the activation of task-related neurological

networks in one fMRI study (Chiew et al., 2012).

Motivation: Motivation was found to be influential in five out of eight studies (Diaz Hernadez

et al., 2016; Kleih et al., 2010; Leeb et al., 2007; Nijboer et al., 2008, Nijboer Birbaumer &

Kubler, 2010) which explored it as a variable. Specifically, higher motivation is linked with

better outcomes. This includes different elements of motivation as measured by the QCM-

BCI, adapted from the Questionnaire for Current Motivation (QCM, Rheinberg, Vollmeyer,

Burns, 2001) such as interest in the task, confidence in mastering the task and the challenge

perceived in the task (Nijboer et al., 2008; Nijboer, Birbaumer & Kubler, 2010; Kleih et al.,

2010). Fear of incompetence, another dimension of the scale indicative of reduced

motivation, has been found to be predictive of poorer performance (Nijboer, Birbaumer &

Kubler, 2010) except in one auditory feedback condition where it was associated with better

performance (Nijboer, Birbaumer & Kubler, 2010). An example of the influence of task

interest was explored in Leeb and colleagues (2007). They found that BCI performance

significantly improved in more interesting virtual reality neurofeedback compared to less

stimulating smiley-face cued feedback training; at least for those that were initially

underperforming. Another study by Diaz Hernandez and colleagues (2016) found that

motivational incongruence, i.e. the difference between a person’s personal goals and what

they are actually able to achieve, is predictive of poorer performance and learning when

influencing class-D microstates in an EEG study (Diaz Hernandez, Rieger & Koenig, 2016).

Some studies have found no link with motivation and outcome; such as Kleih and Kubler

(2013) in their BCI P300 setup. However, Kleih and Kubler considered that the more

motivated group may have been fatigued following a presentation about neurofeedback,

because they were more engaged. This is evidenced by their poorer results on a subsequent

performance measure of attention. This highlights some issues in manipulating motivation

which have been found in other studies which have attempted to use financial incentives to

manipulate extrinsic motivation (e.g. Kleih et al., 2010). Motivation generally is reported to

be high by people taking part in neurofeedback studies, possibly due to how participants are

recruited (often through self-selection) and the because of the treatment potential of certain

conditions. This often means that motivation is at ceiling for most participants. This was also

the conclusion of Hammer et al., (2012) who found no link between SMR influence and

motivation in their study with 83 participants in higher education. Finally, Enriquez-Geppert

and colleagues (2014) found no link between motivational factors and modulation of fm-theta

in their EEG BCI setup. However, their measures of motivation were single Likert scales and

less robust than the well validated QCM-BCI.

Mood: Factors relating to mood were influential in several BCI studies (Gruzelier et al. ,1999;

Hardman et al., 1997; Jeunet et al., 2015; Nijboer et al., 2008) with higher ratings of anxiety

and depression leading to less beneficial outcomes. For example, Hardman and colleagues

(1997) found in an EEG neurofeedback study with participants with schizophrenia that those

that scored higher on ‘withdrawal’ to do with social anxiety were less able to modulate

negativity between their frontal lobes. However, measures of mood vary and raise questions

with regard to convergent construct validity. For example, some studies apply quality of life

scales (Nijboar et al., 2008), others use symptom scales (Gruzelier et al., 1999; Hardman et

al., 1997) or measures of ‘tension’ (Jeunet et al., 2015). There are also some findings that

suggest mood is unrelated to outcomes. Nijboer, Birbaumer and Kubler (2010) found mood

unrelated to both P300 and SMR BCI outcomes. In the study mentioned earlier to do with

motivational incongruence, Diaz Hernandez and collegues (2016) found no influence of state

anxiety measured through the State-Trait Anxiety Inventory (Spielberger, 2010). Marxen and

colleagues (2016) found no influence of anxiety measured through the STAI and depression

measured through the Beck Depression Inventory (Beck et al., 1996) in their fMRI

neurofeedback study involving modulating activity in the amygdala. However, there have

also been some interesting findings relating to susceptibility to feeling emotion in response to

the emotion of others. Susceptibility to anger was found to link to impaired emotional

regulation in two fMRI-based neurofeedback studies (Marxen et al., 2016; Zotev et al., 2011).

Moreover, under specific circumstances, the ability to label emotions may be important in

emotional regulation learning (Marxen et al., 2016).

Personality and other factors: there were limited findings for personality being an influential

factor. In one study, agreeableness has been found to negatively relate to emotional

regulation learning with fMRI-based neurofeedback (Marxen et al. 2016). Additionally, in the

BCI study mentioned earlier exploring motivation in a P300 setup, higher empathy was found

to lead to worse performance (Kleih & Kubler, 2013). However, other studies looking at

personality measured by the Five-Factor Personality Inventory (Borkenau and Ostendorf,

1993) as a predictor of neurofeedback outcomes had non-significant findings (Diaz

Hernandez et al., 2016; Hammer et al., 2012). There has been limited research on other

variables. For example, BCI research has found contradictory findings for locus of control

over technology; in that it can predict both better (Burde & Blankertz, 2006) and worse

(Witte et al., 2013) performance. Other variables may be task specific. For example, self-

rated coping strategies for pain can predict success in reducing activity in pain sensitive

regions (Emmert et al., 2017). Also, abstractness (linked to creativity) and mental rotation

abilities were linked to SMR BCI success in Jeunet and colleagues (2015) perhaps because

this setup requires imagining movements, which may be linked with spatial awareness.

Finally, a single study into the self-regulation of slow cortical potentials links higher IQ with

better neurofeedback outcomes in children with ADHD (Zuberer et al, 2018).

In summary, the evidence base is limited and somewhat disparate, which makes it difficult to

draw concrete conclusions about different variables. However, there is accumulating

evidence for the importance of attention and this also has some face validity as attention is

required for learning and performance in many tasks. Mood and motivation may be related

but further research is required to explain why half of the studies found no influence of these

factors. There is also some preliminary evidence for certain personality factors and task

specific variables that may warrant further investigation.

Other types of Variables that Contribute to Neurofeedback Performance and Learning

Apart from psychological factors there are a range of other potentially influential variables of

interest including learning dynamics, neurophysiological factors and demographic factors.

Learning Dynamics: The way participants learn appears to have some interesting effects on

success in neurofeedback and BCI. Several studies have examined whether providing

strategies for participants is preferable to allowing participants to find strategies for

themselves. There have been mixed findings with some finding that providing no strategy led

to a better outcome (e.g. Hardman et al., 1997; Kober et al., 2013; Sepúlveda et al., 2016),

some finding that suggesting a strategy is necessary for learning (e.g. Neuper et al., 2005;

Scharnowski et al., 2015) and some finding a relative equivalence of both methodologies

(e.g. Hardman et al., 1997). This is likely to depend on how the strategy fits with both the

task and the participant. For example, Nan et al., (2012) found positive imagery was the best

overall strategy for increasing alpha signals in an EEG study (to aid short-term memory) but

effective strategies varied considerably between participants. It is also noteworthy that the

strategy used may affect which idiosyncratic variables affect performance. For example, the

ability to label emotions was found to predict amygdala regulation in an fMRI neurofeedback

study by Zotev and colleagues (2011) where participants were instructed to modify amygdala

activity by recalling positive memories. In contrast, a related study by Marxen and colleagues

(2016) found no link between these variables when they offered no strategy to participants.

One interpretation of this discrepancy is that identifying emotions may be particularly

important for locating memories which produce positive affect; whereas the participants

given no strategy may be free to find other ways to reduce amygdala activity which are more

aligned to their strengths.

Other than the complex area of strategies used for learning, there has also been a small

amount of attention given to learning style. Jeunet et al. (2015) found that active learners

outperform reflective learners in SMR BCI setups. This suggests that it can be more

advantageous to actively experiment and try lots of strategies in a scanner rather than be more

cautious and analyse what is happening. In the same study, it was also an advantage if

learners were “self-reliant”; a trait reflecting the ability to learn autonomously. This is an

interesting finding which also highlights the fact that BCI and neurofeedback may be difficult

for people who are more reliant on help from others. Neurofeedback studies rarely use any

form of ‘coaching’ from researchers and so participants must manage their responses to

feedback, their morale and their attention and therefore their own learning by themselves.

Neurophysiological Factors: Neurophysiological predictors have mainly gained attention in

BCI literature (Alkoby et al., 2017) but are evident across modalities including fMRI

neurofeedback (e.g. Scheinost et al., 2014). Such predictors are often based on initial readings

or performance which can predict later performance (e.g. Kotchoubey et al., 1999; Neumann

& Birbaumer, 2003, Weber et al., 2010, Scheinost et al., 2014) or features of scanning such

as resting state readings of brain activity or signals. Features of white matter connections or

regional volume are also implicated (Blankertz et al., 2010, Wan, Nan, Vai & Rosa, 2014,

Halder et al., 2013, Enriquez-Geppert et al., 2013). These features may have a role in

screening for suitable candidates, though they may be less convenient than other measures

which do not require a scanner to access; such as those measured by questionnaire.

Demographic Factors: Only a handful of studies appear to have explored the influence of

demographic factors on outcome. Recently, age has been found to be associated with the self-

regulation of slow-cortical potentials in children with ADHD (Zuberer et al, 2018) although a

randomised controlled trial with a similar setup found no association for this population

(Janssen et al., 2016). One interesting study on the effect of gender in SMR regulation found

that there was no learning effect when both the participant and the researcher were female in

learning trials (Wood & Kober, 2018). The researchers considered that this could be related

to expectations for gender roles interfering with attention. These findings of influential

demographic factors have not yet been found to generalise to fMRI literature. One might

expect age to be important in younger generations who may still be developing executive

functions (Brocki & Bohlin, 2004) as frontal regions to do with executive processing are

implicated in neurofeedback learning (Emmert et al., 2016). However, a study on fRMI

feedback for emotional regulation in children and adolescents found no link with

demographic variables and outcome (Cohen Kadosh et al., 2016).

Critique of the Literature

There is currently a good evidence base to suggest that people can learn to control their brain

activity through neurofeedback. However, the literature faces some challenges that need to be

better addressed. One example of this is the use of appropriate control groups, where many

studies do not include a control condition that matches the conditions of the treatment group

(Sorger, et al., 2019, Thibault et al., 2018). Additionally, not all fMRI studies control for

noise variables such as respiration and muscle activity, which can have a significant influence

on the BOLD signal (Thibault et al., 2018). It is also of particular relevance to the

inefficiency problem that significant results in neurofeedback are sometimes overrepresented.

Specifically, there are examples of significant results reported which are only found across

certain trials, and significant improvements for certain successful participants are presented

while non-significant results from non-responders are not (Thibault et al., 2018). This may

also mean that results may not be significant in all trials and those that are classed as

responders may in fact not respond in some trials, while the reverse may be true for non-

responders. While the significant results are important, the non-significant results can be of

particular interest to help with the understanding of what causes variability in success. For

example, many studies which have unsuccessful participants may benefit from follow-ups to

consider particular difficulties with individuals and trial interventions such as by providing

alternative strategies.

It is also of particular relevance to consider what is meaningful change as supported by

neurofeedback. For example, it is not clear how much of a reduction in amygdala activity is

required for there to be a noticeable effect in subjective feelings of emotional distress. This is

why it is of particular importance that studies utilise valid and reliable measures of the issues

they intend to treat. Many studies use simple measures such as single item Likert scales for

mood rather than more robust measures (Staunton & Cohen Kadosh, 2019). Due to the

variability of neurofeedback outcomes both between different participants and within the

same participants over time, it is of importance to monitor variables between trials. This

could be done with simple tracking measures in combination with the use of pre and post

measures.

The Current study: A Secondary Data Analysis of Zich et al., (under review)

As the inefficiency problem in neurofeedback has limited research and predictors of success

have yet to be explored in key therapeutic populations, it was the aim of this study to add to

the literature. The current paper is secondary data analysis of a study by Zich and colleagues

(under review) which had a clear example of the inefficiency problem and also included a

variety of psychological measures in pre and post analysis. This therefore represents an

opportunity to explore factors which relate to inefficiency in neurofeedback outcomes. The

authors aimed to train participants to modulate the neurological activity of areas to do with

emotional regulation in adolescent females. This group is chosen due to therapeutic reasons

described earlier relating to adolescence being a key stage in which to intervene with

emotional regulation difficulties. Additionally, male and female hormone and brain

development differs longitudinally and so each sex must be examined separately at this age

group to control for these variables (Herting et al., 2018). The training involved four sessions

of feeding back the control of negative functional connectivity, as in healthy mature brains,

between the Amgydala and the DLPFC. This was displayed with the use of an on-screen

digital thermometer, with higher temperatures associated with greater negativity between the

areas. Zich and colleagues were able to find a significant effect of the neurofeedback training

which increased the healthier and more mature pattern of a negative correlation between the

two areas over and above a non-neurofeedback protocol.

Practice related change, rather than performance, is particularly important in this study as the

clinical benefit of the methodology used would be as an intervention for those with or at risk

of emotional regulation difficulties. However, in Zich et al., (under review) more than half of

the participants were in fact less successful over time (who will be referred to as non-

learners) along with the half that improved (whom will be referred to as learners). This is

another example of the inefficiency problem; with some people seeming to be unable to use

the neurofeedback training to benefit. Taking the behavioural measures of the study along

with any other relevant information, the current paper aims to find variables that contribute to

explaining the variance in performance and practice related change. It was also aimed to

explore the individual differences between participants who improve over time compared to

those whose performance deteriorated.

Hypotheses:

1. a) It was hypothesised that factors implicated by the literature related to mood and

task specific control beliefs and strategies would significantly predict practice related

change and performance.

b) No effect was expected from factors relating to age, IQ and stage of puberty due to

a lack of an evidence base.

c) It was hypothesised that initial readings of functional connectivity, a

neurophysiological factor implicated by the literature, would significantly predict

performance but not practice related change, as found in previous studies.

2. a) It was hypothesised that non-learners would be unable to regulate as indicated by

the inefficiency problem described in the literature and so would underperform

throughout training compared to learners.

b) It was hypothesised there would be significantly greater scores for measures of

thought and emotional control beliefs and strategies prior to intervention in learners

compared to non-learners.

c) It was hypothesised there would be significantly lower scores of anxiety and

depression prior to intervention in learners compared to non-learners.

d) No significant differences were expected in age, IQ, initial functional connectivity

and stage of puberty between learners and non-learners.

The benefit of this study will be to improve understanding of both performance and change

over time in neurofeedback with a key therapeutic population. This in turn allows the

prediction of neurofeedback outcomes, the development of ways to improve the efficacy of

neurofeedback and also helps us to understand who is most likely to benefit from such a

technique.

Method

The current paper is a secondary data analysis. An overview of the methodology of the

original paper (Zich et al., under review) is included below for background information.

Participants

Zich et al., (under review) recruited forty female adolescent participants from local schools in

the Oxfordshire/Gloucestershire area. Of these, 28 were selected for the current study (Table

1) as they were in the experimental condition of neurofeedback where feedback on negative

functional connectivity was provided, as opposed to a smaller group (N = 12) where feedback

for positive connectivity was provided. No participants had reported difficulties with vision

or histories of neurological and psychiatric disorders (determined via self-report). Informed

written consent was obtained from the primary caregiver and informed written assent was

obtained from the adolescent. Participants received an Amazon voucher (£20) for their

participation. The study was approved by the Central Oxfordshire Ethics Committee (MSD-

IDREC-C2-2015-023) along with all researchers involved in accordance with the Declaration

of Helsinki (Appendix A).

Table 1. Demographic Information from Participants

Variable OutcomeAge Mean age = 15.21 years; SD = 1.23 years; Range 13-17 years

Race/Ethnicity 60% White British, 9% Mixed, 6% Asian, 25% UnstatedHandedness 69% Right Handed, 23% Left Handed, 8% Unstated

IQ Mean IQ = 106.6, SD = 7.85, Range 92-122

fMRI data acquisition

A 3 T Siemens MAGNETOM Prisma MRI scanner (Siemens AG, Erlangen, Germany) with

a standard 32-channel head matrix coil was utilised in Zich et al., (under review). A high-

resolution structural scan was obtained, which was followed by functional imaging during the

localizer task and neurofeedback sessions. Zich et al. (under review) used a social scenes

task (Haller et al. 2016) as the localizer (Appendix B).

Neurofeedback task

Each participant undertook four identical runs lasting 4.8 min each. These runs all began with

a fixation cross which was presented for 20 volumes (18.66 s). Then seven neurofeedback

and seven non-neurofeedback blocks were presented in alternating order, with each lasting 20

volumes. The order of presentation was counterbalanced and randomised across participants.

Stimulus presentation was controlled with BrainStim 1.1.0.1 (open source stimulation

software, Maastricht University, http://svengijsen.github.io/BrainStim/). After the fixation

cross, a ten-segment thermometer was presented at the centre of the screen on a dark grey

background. During non-neurofeedback mini-blocks, the thermometer was stationary at

segment six (i.e. 60%), whereas during the weighted neurofeedback mini-blocks, the

thermometer featured a green frame and would update with each volume. The temperature

reflected negativity between amygdala and DL-PFC which was calculated in real-time over a

moving window of 20 volumes. It was defined as the partial correlation between PFC and

amygdala activity, against corticospinal tract activity. All participants were asked to increase

the temperature on the thermometer by revisiting positive thoughts and emotions (Appendix

B).

Predictor Variables

To our knowledge, there has not been previous literature to explore predictor variables for

females of this age group. Therefore, a wide variety of variables from the original study was

considered (Table 2). Psychological variables were explored using self-report questionnaires

(Appendix C) other than IQ which was obtained through assessment. Other demographic

factors such as age and a self-report measure of stage of puberty were established in the pre-

interview. All the measures used have been found to be valid. Reliability statistics are

presented in Table 2 where available. One neurophysiological factor, initial functional

connectivity (IFC), was established prior to the neurofeedback runs. Learning strategies were

explored in the post interview and completed with both a checklist of different strategies and

a qualitative statement about what they found helpful. A 1-4 Likert scale of how difficult they

perceived the task was included. Post measurement of psychological variables were not

explored in this study as we were concerned with predicting practice related change and

performance.

Table 2. Description and Measurement of Variables included in Analysis

Variable (abbreviation)

Method of Measurement Description of Measurement Internal Consistency (α)

Test-Retest Reliability

(r)

Age Demographic and Health questionnaire

Self-administered, includes demographic and relevant

medical history

NA NA

Stage of Puberty Puberty Development Scale (PDS; Carskadon &

Acebo, 1993)

Self-administered scale indicating stage of puberty.

Correlates strongly with paediatrician ratings of

development

.70 NA

Initial Functional Connectivity

(IFC)

Use of fMRI setup prior to neurofeedback task and following Localiser task.

Partial correlation (rp) between PFC and amygdala activity,

against CST activity

NA NA

Though Control Questionnaire

(TCQ)

Thought control questionnaire (TCQ)

(Wells & Davies, 1994)

30-item measure to assess effectiveness of strategies used

for the control of unpleasant/unwanted thoughts

.64 to .79 .83

Thought Control Ability (TCA)

Thought control ability questionnaire (TCAQ, Luciano, Algarabel, Tomas, & Martinez,

2005)

25-item measure of individual differences in perceived ability to

control unwanted & intrusive thoughts

.92 .88

IQ Wechsler Abbreviated Intelligence Scale (WASI-

II)(Wechsler, 1999)

Global intelligence/IQ measure. Includes a Verbal comprehension

measure and a Perceptual Reasoning measure

NA NA

Emotional regulation and

cognitive coping

Cognitive Emotion Regulation Questionnaire Short (CERQ)(Garnefski,

Kraaij, & Spinhoven, 2001)

18-item measure of cognitive coping strategies. Includes dimensions such as positive

focus or reappraisal

.67 to .81 NA

Mood Moods and feelings questionnaire (MFQ)

(Angold, 1995)

13 Item measure of depression. Correlates well with psychiatric

referrals

.85 NA

State/trait anxiety State/Trait Anxiety Inventory for Children (STAIC) (Spielberger,

Gorsuch & Luchene,1976)

40-item measure of state & trait anxiety; 20 items for each

dimension

.86 to.94 .36 to .75

Social Anxiety Social Anxiety Scale for adolescents (SAS: La

Greca & Lopez, 1998).

22 item measure of social anxiety. Correlates strongly with other measures of social anxiety

.76 to .91 NA

Strategy Used (Qualitative

only)

Debriefing interview questionnaire

Checklist for four different strategies: Thinking about the past, thinking about the future, thinking about someone, trying

to stop being unhappy. Also includes a qualitative description

NA NA

Dependent Variables

Practice related change was established by calculating learning slopes for each participant

over the four neurofeedback runs based on performance in each session. Performance was

conceived in terms of the consistency to which participants were able to generate negative

connectivity between the amygdala and DL-PFC. This was obtained as a frequency of the

number of negative correlations in the moving window of functional connectivity analysis

during runs, with a maximum of 291 per run. Finally, participants were divided into those

with a positive value for learning slope that improved (learners) and those with a negative

value for learning slope whose performance deteriorated over time (non-learners). This

enabled the exploration of group level differences between those that achieved positive

practice related change in response to neurofeedback and those that did not. The mean

learning slope for learners was 6.06 (SD = 5.43) and for non-learners it was -9.78 (SD =

7.04).

Data Analysis

First, a series of correlations was conducted to explore associations between all the variables

in the study. This revealed associations between predictor variables (see Results). A principal

components analysis was then applied to generate common factors amongst these predictor

variables. This analysis (see results) revealed two factors in particular. One included

measures to do with perceived control over thoughts and emotions which was labelled

“Control”. The other involved factors to do with physical development and IQ which was

labelled “Maturity”. IFC was unrelated to the other variables and so was removed from the

analysis. The two factors and IFC were then applied in a linear multiple regression model to

predict variance in performance according to the number of negative correlations achieved

and practice related change according to learning slope.

In addition to these analyses, it was explored as to whether there were group-level differences

between participants with a positive learning slope (defined as learners) and participants with

a negative learning slope (defined as non-learners). Initially, analyses of variance were

conducted to confirm that learners and non-learners significantly improved and deteriorated

in performance respectively. Then a series of independent samples t-tests were conducted

examining differences in performance over time for each group and differences in measures

of the independent (predictor) variables for learners and non-learners.

Results

Assumptions Testing

The tests for normality examining standardised skewness and the Shapiro-Wilks test

indicated that the data were statistically normal with the exception of age and scores on the

MFQ and PDS (see Appendix D). Three outliers were identified and removed for analyses

involving the PDS scale (Appendix D). There were a total of four missing cases for TCA-Q,

two missing cases for MFQ and three missing cases for state anxiety which were removed list

wise for relevant analyses. For principal components analysis, Kaiser-Meyer-Olkin was

acceptable (.696) and Bartlett’s Test of Sphericity was significant (p = .000). The factors

produced had a normal distribution (Appendix D). For multiple regression, the enter method

was used for variables. Normal/scatterplots and Cook’s D were acceptable (see Appendix E)

and adjusted R square values were interpreted due to the relatively small sample size. For

Analyses of Variance (ANOVA), tests for sphericity were non-significant (see Appendix E).

For T-tests, tests for homogeneity of variance were non-significant with the exception of the

number of negative correlations in session one and PDS. In these cases, a Mann-Whitney U

test was used as an alternative. Hedge’s g was used instead of Cohen’s d to calculate effect

size for the T-tests due to unequal sample sizes.

Descriptive Statistics

Table 3 shows means, standard deviations and ranges for all the variables involved in the

analysis.

Table 3: Sample Means, standard deviations and ranges for Variables used in Analysis

Variable Mean (SD) Range

Age 15.21 (1.23) 13, 17

Stage of Puberty (PDS) 18.14 (2.52) 12, 22

Initial Functional Connectivity (IFC) .297 (.238) -.108, .705

Though Control Questionnaire (TCQ) 61.32 (8.12) 50, 80

Thought Control Ability (TCA) 78.67 (14.98) 52, 107

IQ (WASI-II) 106.19 (8.30) 91.75, 121.75

Emotional regulation/cognitive coping (CERQ) 60.43 (10.81) 39, 78

Mood and Feelings Questionnaire (MFQ) 5.27 (4.62) 0, 18

State Anxiety (STAI) 33.44 (9.07) 20, 51

Trait anxiety (STAI) 43.96 (11.64) 25, 64

Social Anxiety (SAS) 49.32 (14.62) 29, 85

Performance (number of negative correlations) 274.89 (100.98) 111, 465

Practice Related Change (learning slope) -2.42 (10.17) -24.30, 14.60

IFC averaged at a positive value for the sample suggesting the development of adult

connectivity patterns was still in process for this group. The sample means mostly fell well

within a standard deviation from means reported in previous literature (Table 2). However,

the sample was above average for women in scores for the thought control questionnaire

(from Wells & Davies, 1994: M = 48.29, SD = 6.21) and had slightly higher than average IQ.

Correlations

There were many significant correlations between predictor variables (see Appendix F)

suggesting they may be tapping into similar elements. In addition, a mild positive correlation

was found between learning slope and scores on the verbal comprehension test (r = .373, n =

28, p = .050, CI = .000, .655). IFC significantly correlated with performance (r = -.560, n =

28, p = .002, CI = -.771, -.236). Those with greater initial negativity between regions

achieved a greater number of negative correlations between the regions through

neurofeedback (Figure 2). No other correlations were found between predictor variables and

either performance and practice-related change.

Principal Components Analysis

A principal components analysis was conducted using the rotation method of Oblimin with

Kaiser Normalization to see whether predictor variables were tapping into similar constructs.

A scree plot and eigenvalues suggested two potential factors (see Appendix G). However,

after interpretation it appeared that the third factor was tapping into similar elements from

factors one and two. Subsequentially, a two-factor analysis was forced. The variable IFC did

not fit into major factors and so was removed from the analysis. The two-factor model

accounted for 64.68% of the variance in the predictor variables. The first factor consisted of

measures to do with mood, and self-ratings of control and strategies used to manage thoughts

and emotions. This was labelled ‘Control’ and appears to relate to the perceived efficacy of

participants for psychological self-management. The second factor consisted of age, the

puberty development scale and IQ. This appeared to tap into elements of physical and

cognitive development and was labelled ‘Maturity’.

Hypothesis 1: Predictors of Practice Related Change and Performance

Multiple regression analysis was used to test if the factors ‘Control’, ‘Maturity’ and IFC

significantly predicted participants' learning slope over the four blocks. The results of the

regression indicated that the model predicted 6.6% of the variance, and the model was not

significant (Adjusted R2=-.066, F (3, 17) = .587, p =.632). In addition, a multiple regression

analysis was used to test if the factors ‘Control’, ‘Maturity’ and IFC significantly predicted

performance over the four blocks. The results of the regression indicated that the model

explained 43.1% of the variance and was significant (Adjusted R2=.431, F (3, 17) = 4.11, p

=.020). The results of the regression indicated that initial functional connectivity was the

only significant predictor of performance (β =-254.06, p = .004, CI = -415.75, -92.37).

Figure 2: Correlation of Initial Functional Connectivity with Overall Performance

Establishment of Learner and Non-Learner groups

Participants were organised into groups of learners and non-learners. Participants with a

positive learning slope were labelled ‘learners’ and those with a negative learning slope were

labelled ‘non-learners’. To confirm the main effect of time on the performance of learners, a

one-way repeated measure ANOVA was conducted. There was a significant effect of time on

performance (Wilks’ Lambda = .392, F (3, 10) = 5.18, p = .02, η2 = .608 observed

power .781). To confirm the main effect of time on the performance of non-learners, a one-

way repeated measure ANOVA was conducted. There was a significant effect of time on

performance (Wilks’ Lambda = 319, F (3, 12) = 8.55, p = .003, η2 = .681 observed

power .961). Follow up pairwise comparisons indicated that there was a significant difference

(p = .024) between blocks one and four for learners. There was a significant improvement in

performance over time for learners. For non-learners there was a significant difference (p

<.05) between session one and three, one and four and sessions two and four. The

performance of non-learners deteriorated over time.

Hypothesis 2: Comparison of Performance of Learners and Non-Learners

A series of independent samples T-tests and Mann-Whitney U tests were carried out to

explore differences across the four sessions between learners and non-learners.

A Mann-Whitney U test found that learners (Mdn = 61) and non-learners (Mdn = 80) did not

differ significantly in performance in the first run of four (U = 70.0, p = .205, r = .24).

However, an independent T-test suggested that learners (M = 75.77, SD = 23.39)

significantly outperformed non-learners (M = 53.13 SD = 32.22) by the fourth run (t, 26 =

2.097, p = .046, g = 0.794, CI = .45, 44.82). Learners climbed above the overall mean (M=

63.64, SD = 25.24) over the four sessions where as non-learners dropped beneath it (Figure

3). However, taking the whole four sessions into account, there was no significant difference

between learners and non-learners in performance (t, 26 = .082, p = .935).

Figure 3: Bar Chart Showing the Change in Number of Negative Correlations Achieved over the Four Blocks

for Learners and Non-Learners

Overall Mean per Session

Hypothesis 2: Comparison of Predictor Variables between Learners and Non-Learners

A series of independent samples T-tests and Mann-Whitney U tests were carried out to

explore differences in predictor variables between those with a positive trend in performance

(learners) compared to those that showed a negative trend (non-learners).

There was a significant difference in thought control ability (t, 22 = 2.625, p = .015, g = 1.07,

CI = 3.93, 24.8), with learners having greater perceived ability (M = 86.45, SD = 15.41) than

non-learners (M = 72.08, SD = 11.40). There was also a significant difference between

groups in IQ-Verbal comprehension (t, 26 = 2.089, p = .047, g = .78, CI = -.013, 10.33).

Learners had higher verbal comprehension ability (M = 58.23 SD = 5.78) than non-learners

(M = 53.07, SD = 7.30). This difference accounted for learners having higher overall IQ (M =

110 SD = 8.63) than non-learners (M = 103 SD = 6.20) by 7 points; a difference which was

statistically significant (t, 26 = 2.786, p = .010, g = .94, CI = 1.22, 12.78).

It may be of note that state anxiety was near the significance threshold. Learners had lower

state anxiety (M = 29.83, SD = 8.67) than non-learners (M = 36.77, SD = 8.39) though this

was not significant (t = -2.032, 23, p = .054, g = .81, CI = .303, 13.58). No other significant

differences between predictor variables were found between groups.

Discussion

While fMRI neurofeedback is a promising technique for intervening with a range of clinical

disorders, not everyone who receives the intervention is able to benefit. Here the question

was explored as to which factors can predict practice-related change and performance in

neurofeedback, and whether there are differences between those that benefit from

neurofeedback and those that do not. A systematic literature review implicated attention,

mood and motivational factors as potential psychological predictors. Further research has

explored learning dynamics as well as neurophysiological and demographic factors. The

current report aimed to add to this literature with a secondary data analysis of an fMRI

neurofeedback study which found significant improvement of emotional regulation in some

participants and no improvements or reduced performance over time in others (Zich et al.,

under review). This question is posed with a therapeutically important client group who are

at a crucial stage of development for healthy brain connectivity patterns. It was found that

factors to do with self-perceived ‘Control’ and ‘Maturity’ did neither predict performance nor

practice related change over the four blocks of training to a significant degree. An initial

reading of functional connectivity was found to be a significant predictor of performance but

not practice related change. At a group level, it was found that learners began with

performance below the mean and improved to a level above the mean, whereas non-learners

started with performance greater than the mean but were below the mean by the final session.

No differences were found between learners and non-learners in overall performance.

Significant individual differences were found between those that improve over time and those

that decrease in performance over time. Namely, both self-reported thought-control ability

and measures of verbal comprehension were found to be higher in those that improve over

time. It is possible that state anxiety is also different between the groups; though this was

above the threshold for significance in this study.

Predictors of Performance and Practice Related Change

The factor ‘Control’ is inclusive of measures of mood and anxiety, and to an extent, control

beliefs. Therefore, previous findings that mood (Gruzelier, Hardman, Wild & Zaman, 1999;

Hardman et al. 1997; Nijboer et al., 2010; Marxen et al., 2016; Zotev et al., 2007) and task

specific control strategies and beliefs (Burde and Blankertz, 2006; Emmert et al., 2017, Witte

et al., 2007) can predict neurofeedback outcomes are not supported here. This contradicts

hypothesis 1a and may indicate that real-world abilities to manage and control emotions and

thoughts may be separate to the abilities required for the control of processes of

neurofeedback. It also suggests that the influence of psychological variables may be more

complex as found by some studies (e.g. Nijboer et al., 2008) where psychological factors

affect different participants in different ways and under different methodologies. This may

mean that using such variables as a predictor of success is challenging.

Hypothesis 1b that variables to do with the factor ‘Maturity’ would not predict outcomes was

supported. Only one study implicated age and IQ as influential; a study which involved the

self-regulation of slow cortical potentials (Zuberer et al, 2018) and while a model including

‘Maturity’ predicted performance, neither of these variables contributed significantly.

Therefore, previous findings that demographic variables such as age are unrelated to

performance in emotional regulation (Cohen Kadosh et al., 2016) were supported. Finally, the

hypothesised (hypothesis 1c) ability of IFC to predict performance was supported and in line

with previous evidence (e.g. Kotchoubey et al., 1999; Neumann & Birbaumer, 2003, Weber

et al., 2010, Scheinost et al., 2014) which suggests that initial readings provide a prediction of

performance for the rest of training. However, this did not predict practice related change and

so initial performance may be more important in neurofeedback studies where performance is

emphasised (e.g. EEG P100 amplitude studies).

Differences between Learners and Non-Learners

The finding that non-learners began from a higher performance than learners was

contradictory to hypothesis 2a. This suggests that other factors such as fatigue may have

increased through training for these individuals. Such a finding would be in line with findings

from other studies that suggest maintaining attention through the effects of fatigue is crucial

to success in neurofeedback (e.g. Gruzlier et al., 1993). The lack of difference between

groups in overall performance suggests that non-learners are just as able to achieve the

desired pattern of functional connectivity. This does not fit with the narrative that non-

learners are simply unable to succeed in neurofeedback (e.g. Alkoby et al., 2017; Enriquez-

Geppert, 2014,) and may instead suggest more complex issues. For example, non-learners

may begin with better performance and so have less to gain from extended neurofeedback.

They may then become bored, less motivated and more likely to fatigue. However, there were

also some differences in participant variables between groups.

At a group level, ratings of the ability to control thoughts was greater in learners in partial

support of hypothesis 2b. However, no other ratings of control and regulation strategies

differed between groups. The results partially support the findings of (Burde &, 2006) that

control beliefs are assistive in neurofeedback, but contradict the opposing finding of Witte

and colleagues (2013). Findings that self-ratings of coping strategies can predict success in

fMRI neurofeedback (Emmert et al., 2017) along with measures of emotional regulation in

active feedback emotional regulation trials (Marxen et al., 2016) were not supported in the

present study. It is particularly surprising that thought control strategies (TCQ) should not

differ between groups whereas perceived thought control ability (TCA) does. This might

suggest that the confidence one has in their own abilities to control their own thoughts is a

better measure of potential outcomes in the scanner than the reports of actual strategies they

apply day to day. It is conceivable for example that some participants may have good

knowledge of strategies that help but have no confidence in their ability to use them

successfully.

State anxiety was near the significance threshold for being higher in non-learners, but overall

hypothesis 2c was not supported. The original study by Zich and colleagues (under review)

found a correlation with state anxiety and change related to the presentation of

neurofeedback; suggesting some involvement of this variable in outcomes. These mixed

findings are to an extent reflective of the literature which has found varying effects of anxiety

on neurofeedback outcomes (Gruzelier et al. ,1999; Hardman et al., 1997; Jeunet et al., 2015;

Nijboer et al., 2008; Nijboer, Birbaumer & Kubler, 2010). Note also that findings from other

studies that low mood can have a negative impact on neurofeedback success (Gruzelier et al.,

1999; Nijboer et al., 2008) were not supported in this study.

The finding that scores on the verbal comprehension element of IQ were greater in learners of

emotional regulation appears to be novel and contradictory to hypothesis 2d. However, no

differences for age, IFC and puberty development were found which was in line with the

hypothesis. Note also that there was a positive correlation between verbal comprehension and

learning slope; although on the borderline of significance. To a certain extent, measures of IQ

tap into attentional abilities. However, no difference was found between groups in the

perceptual reasoning element of IQ which should also benefit from attentional abilities. It is

possible that verbal comprehension helps with successful strategies such as generating

reappraisal thoughts, or labelling emotions and associated memories. Previous research has

found the ability to label emotions is useful in fMRI neurofeedback emotional regulation

where recall of positive memories is a suggested strategy (Marxen et al., 2016). This is

relevant in the current study because all but one participant cited that they used recall of

positive memories as a strategy at some stage. It may also be that participants with lower

verbal comprehension ability find experimental instructions abstract; though this seems less

likely as both learners and non-learners did not differ in performance. The result may also be

a feature unique to this population because those in the earliest stages of adolescence may be

at a time of development where IQ related factors such as the development of executive

functions, including complex attentional skills are still in process (Brocki & Bohlin, 2004).

Based on the literature review and the findings of the current study, a potential model of

learning and performance is suggested below (Figure 4).

Figure 4. A Multi-stage Model of Success in Neurofeedback

In this model, neurofeedback training is conceived of being dependent on multiple

consecutive stages in the process. It begins with participants engaging in the process and

directing their attention appropriately towards the task (1-3). Without this stage there cannot

be further progression and learning will be unsuccessful (5). If this stage is satisfactory,

participants enter the performance stage whereby they actively experiment with efferent

processes similar to how it is conceived in dual-process theory (3, 3a, 3ab; Lacroix, 1981).

Those who can direct their attention appropriately and quickly find good strategies will

perform well. The final learning stage occurs when participants have found effective

strategies and received sufficient neurofeedback so that they can begin to shortlist these

techniques and refine them further (3a, 4). The accrued techniques will then be available in

later training and be further refined as necessary. However, this effect can potentially be

overridden at any stage by the loss of attention through distraction, fatigue or reduced

motivation. Note that the current study examines practice related change which is likely to be

somewhere in between performance and learning. As learning and performance are separate

processes, good performance would not necessarily relate to good learning or we would

expect good learners to also be good performers. Rather, some participants may initially

perform well but lose interest or fatigue and so fail to build on their success, whereas others

may struggle at first but respond quickly when they find success. By this theory we would

only expect a threshold level of “performance” required for learning as participants must at

some point find a functional strategy on which to build.

Within this model participant factors are conceived of as influencing learning and

performance in two ways. Firstly, in their ability to influence attentional resources. For

example, state anxiety may pull attention towards unrelated thoughts or physical symptoms

(e.g. increased heart rate) rather than the task at hand. Secondly, participant factors may

facilitate the achievement threshold level of performance though their link with effective

strategies, thus increasing the chance of learning. Examples of this second type of influence

may be task specific; such as in the findings of Emmert and colleagues (2017) that self-

reported pain management strategies predict the ability to modulate pain related areas in a

course of neurofeedback. Some variables may tap into both of these influences. For example,

verbal comprehension is, as with many IQ related tests, related to attention but may also help

people to verbally process and label emotions in aid of adequate emotional regulation in the

current study.

Limitations

The results of this secondary data analysis should be interpreted with caution. This was a

small sample which can only detect larger effect sizes. Missing data made the sample smaller

still for some analyses and may have led to bias if there were similarities between individuals

who missed out certain questionnaires. Performance session to session for individuals was

often highly variable. This may be a feature of this type of training and has been found in

other studies (e.g. Chiew et al., 2012). The variability may in part be due to the difficulty of

the task in that attention must be sustained for extended periods to keep track of delayed

feedback in spite of an array of internal and external distractors and fatigue. The above

factors led to large confidence intervals in some cases and as such, larger scale research

would be required to confirm these findings. This study did not have a direct measure of two

key variables discussed in the literature; that of motivation and attention and so cannot draw

conclusions about psychological predictors generally. Additionally, this study is limited in its

ability to equate practice-related change over the four sessions to learning. Multiple groups of

blocks and a transfer run would be required to be more confident in this aspect. Moreover, the

findings are limited in how much they may be generalised to other populations and modalities

of feedback. The sample itself was affluent and had high IQ relative to the rest of the

population. Future research into this field should consider in its design that psychological

factors are often collinear and may have a complex relationship with performance or learning,

thus leading to subtler effects.

Future Research

The effect of individual differences in neurofeedback success are often considered as a

secondary priority if at all. However, neurofeedback must be justified in its efficiency

especially in fMRI research which is relatively resource heavy. There is a need for research

which either manipulates individual differences or can group participants according to such

variables to clarify their predictive ability. There needs to be a clearer description of learners

and non-learners and their performance over the course of different sessions to better

understand these groups as they have been labelled. There is also a great need for follow up

studies with those that do not benefit from neurofeedback to offer intervention; such as with

the provision of coaching for neurofeedback strategies or anxiety management techniques.

All future research into individual differences should at a minimum include measures of

attention, IQ, anxiety (or relevant self-regulation control beliefs) and motivation. This

research will allow tailored interventions to be developed for non-responders in future to

improve the efficacy of neurofeedback.

References

Aldao, A., Nolen-Hoeksema, S., & Schweizer, S. (2010). Emotion-regulation strategies

across psychopathology: A meta-analytic review. Clinical Psychology Review, 30,

217–237.

Alkoby, O., Abu-Rmileh, A., Shriki, O., & Todder, D. (2017). Can we predict who will

respond to neurofeedback? A review of the inefficacy problem and existing predictors

for successful EEG neurofeedback learning. Neuroscience,12.050

Angold, A., Costello, E, Messer, S., Pickles, A, Winder, F, & Silver, D. (1995). The

development of a short questionnaire for use in epidemiological studies of depression

in children and adolescents. International Journal of Methods in Psychiatric

Research, 5, 237-249

Attwell, D., & Iadecola, C. (2002). The neural basis of functional brain imaging signals.

Trends Neurosci, 25(12), 621-625.

Bambauer, K. Z., Soumerai, S. B., Adams, A. S., Zhang, F., & Ross-Degnan, D. (2007).

Provider and patient characteristics associated with antidepressant nonadherence: the

impact of provider specialty. The Journal of clinical psychiatry, 68(6), 867-873.

Barnett, J.H. & Fletcher, P.C. (2008) in: Cappa, S. F. (2008). Cognitive neurology : A

clinical textbook. Oxford: Oxford University Press.

Barsaglini, Sartori, Benetti, Pettersson-Yeo, & Mechelli. (2014). The effects of

psychotherapy on brain function: A systematic and critical review. Progress in

Neurobiology, 114, 1-14.

Beck, A.T., Steer, R.A., Ball, R., and Ranieri, W.F. (1996). Comparison of beck depression

inventories -IA and-II in psychiatric outpatients. J. Pers. Assess. 67, 588–597.

Blankertz B, Sannelli C, Halder S, Hammer EM, Ku¨ bler A, Mu¨ ller K-R, et al. (2010)

Neurophysiological predictor of SMR-based BCI performance. Neuroimage 51:1303–

1309.

Borkenau, P., & Ostendorf, F. (1993). NEO-Fünf-Faktoren Inventar:(NEO-FFI); nach Costa

und McCrae. Hogrefe.

Brocki, K., & Bohlin, G. (2004). Executive Functions in Children Aged 6 to 13: A

Dimensional and Developmental Study. Developmental Neuropsychology, 26(2), 571-

593.

Burde, W., & Blankertz, B. (2006). Is the locus of control of reinforcement a predictor of

brain-computer interface performance?. In Proceedings of the 3rd International

Brain-Computer Interface Workshop and Training Course (Vol. 2006, pp. 108-109).

Caria, A., Sitaram, R., & Birbaumer, N. (2012). Real-time fMRI: a tool for local brain

regulation. The Neuroscientist: a review journal bringing neurobiology, neurology

and psychiatry, 18(5), 487.

Carskadon, M. A., & Acebo, C. (1993). A self-administered rating scale for pubertal

development. Journal of Adolescent Health, 14(3), 190-195.

Chervonsky, E., & Hunt, C. (2018). Emotion Regulation, Mental Health, and Social

Wellbeing in a Young Adolescent Sample: A Concurrent and Longitudinal

Investigation. Emotion, Emotion, 2018.

Chiew, M., Laconte, S. M., & Graham, S. J. (2012). Investigation of fMRI neurofeedback of

differential primary motor cortex activity using kinesthetic motor imagery.

NeuroImage, 61(1), 21-31.

Cohen Kadosh, K., Luo, Q., de Burca, C., Sokunbi, M. O., Feng, J., Linden, D. E., & Lau, J.

Y. (2016). Using real-time fMRI to influence effective connectivity in the developing

emotion regulation network. Neuroimage, 125, 616-626.

Daum, I., Rockstroh, B., Birbaumer, N., Elbert, T., Canavan, A., & Lutzenberger, W. (1993).

Behavioural treatment of slow cortical potentials in intractable epilepsy:

neuropsychological predictors of outcome. Journal of Neurology, Neurosurgery

&amp; Psychiatry, 56(1), 94.

deCharms, R. C., (2008). Applications of real-time fMRI. Nature Reviews Neuroscience,

9(9), 720-9.

deCharms, R. C., Maeda, F., Glover, G. H., Ludlow, D., Pauly, J. M., Soneji, D., . . . Mackey,

S. C. (2005). Control over brain activation and pain learned by using real-time

functional MRI. Proceedings of the National Academy of Sciences, 102(51), 18626-

18631.

Diaz Hernandez, L., Rieger, K., & Koenig, T. (2016). Low motivational incongruence

predicts successful EEG resting-state neurofeedback performance in healthy adults.

Neuroscience.12.005

Di Bona, Laura, Saxon, David, Barkham, Michael, Dent-Brown, Kim, & Parry, Glenys.

(2014). Predictors of patient non-attendance at Improving Access to Psychological

Therapy services demonstration sites. Journal of Affective Disorders, 169, 157.

Enriquez-Geppert S, Huster RJ, Scharfenort R, Mokom ZN, Vosskuhl J, Figge C, Herrmann

CS (2013) The morphology of midcingulate cortex predicts frontal-midline theta

neurofeedback success. Front Hum Neurosci 7:453

Emmert, K., Breimhorst, M., Bauermann, T., Birklein, F., Rebhorn, C., Van De Ville, D., &

Haller, S. (2017). Active pain coping is associated with the response in real-time

fMRI neurofeedback during pain. Brain Imaging and Behavior, 11(3), 712-721.

Emmert, Kopel, Sulzer, Brühl, Berman, Linden, . . . Haller. (2016). Meta-analysis of real-

time fMRI neurofeedback studies using individual participant data: How is brain

regulation mediated? NeuroImage, 124(Pt A), 806-812.

Garnefski, N., Kraaij, V., & Spinhoven, P. (2001). Negative life events, cognitive emotion

regulation and emotional problems. Personality and Individual Differences, 30(8),

1311-1327

Gee, D.G., Humphreys, K.L., Flannery, J., Goff, B., Telzer, E.H., Shapiro, M., ... Tottenham,

N., (2013). A developmental shift from positive to negative connectivity in human

amygdala–prefrontal circuitry. J. Neurosci. 33 (10), 4584–4593.

Goethe, J. W., Woolley, S. B., Cardoni, A. A., Woznicki, B. A., & Piez, D. A. (2007).

Selective serotonin reuptake inhibitor discontinuation: side effects and other factors

that influence medication adherence. Journal of Clinical Psychopharmacology, 27(5),

451-458.

Gratz, K., & Roemer, L. (2004). Multidimensional assessment of emotion regulation and

dysregulation: Development, factor structure, and initial validation of the Difficulties

in Emotion Regulation Scale. Journal of Psychopathology and Behavioural

Assessment, 26, 41–54.

Grosse-Wentrup, M., & Schölkopf, B. (2013). A review of performance variations in SMR-

based Brain− Computer interfaces (BCIs). In Brain-Computer Interface

Research (pp. 39-51). Springer, Berlin, Heidelberg.

Gruzelier, J., Hardman, E., Wild, J., & Zaman, R. (1999). Learned control of slow potential

interhemispheric asymmetry in schizophrenia. International Journal of

Psychophysiology, 34(3), 341-348.

Gruzelier, J. H., & Kaiser, J. (1996). Syndromes of schizotypy and timing of

puberty. Schizophrenia Research, 21(3), 183-194.

Halder S, Varkuti B, Bogdan M, Ku¨ bler A, Rosenstiel W, Sitaram R, et al. (2013)

Prediction of brain-computer interface aptitude from individual brain structure. Front

Hum Neurosci 7:1–9.

Haller, S.P.W., Raeder, S.M., Scerif, G., Cohen Kadosh, K., & Lau J.Y.F. (2016). Measuring

online interpretations and attributions of social situations: Links with adolescent

social anxiety. J Behav Ther Exp Psychiatry. 50:250–256.

Hamedi, M., Salleh, S. H., & Noor, A. M. (2016). Electroencephalographic motor imagery

brain connectivity analysis for BCI: a review. Neural computation, 28(6), 999-1041.

Hammer, E. M., Halder, S., Blankertz, B., Sannelli, C., Dickhaus, T., Kleih, S., . . . Kübler,

A. (2012). Psychological predictors of SMR-BCI performance. Biological

Psychology, 89(1), 80-86.

Hardman, E., Gruzelier, J., Cheesman, K., Jones, C., David Liddiard, H., Schleichert, N., &

Birbaumer, N. (1997). Frontal interhemispheric asymmetry: self-regulation and

individual differences in humans. Neuroscience Letters, 221(2), 117-120.

Hebb, D.O. (1949). The Organization of Behavior. New York: Wiley & Sons.

Herting, Johnson, Mills, Vijayakumar, Dennison, Liu, . . . Tamnes. (2018). Development of

subcortical volumes across adolescence in males and females: A multisample study of

longitudinal changes. NeuroImage, 172, 194-205.

Janssen, T., Bink, M., Geladé, K., Mourik, R., Maras, A., & Oosterlaan, J. (2016). A

randomized controlled trial into the effects of neurofeedback, methylphenidate, and

physical activity on EEG power spectra in children with ADHD. Journal of Child

Psychology and Psychiatry, 57(5), 633-644.

Jeunet, C., N’Kaoua, B., Subramanian, S., Hachet, M., & Lotte, F. (2015). Predicting mental

imagery-based BCI performance from personality, cognitive profile and

neurophysiological patterns. PloS one, 10(12), e0143962.

Jeunet, C., N’Kaoua, B., & Lotte, F. (2016). Advances in user-training for mental-imagery-

based BCI control: Psychological and cognitive factors and their neural correlates.

In Progress in brain research (Vol. 228, pp. 3-35). Elsevier.

Johnston, S., Boehm, S., Healy, D., Goebel, R., & Linden, D. (2010). Neurofeedback: A

promising tool for the self-regulation of emotion networks. Neuroimage, 49, 1066-

1072.

Kessler, R. C., Berglund, P., Demler, O., Jin, R., Merikangas, K. R., & Walters, E. E. (2005).

Lifetime prevalence and age-of-onset distributions of DSM-IV disorders in the

National Comorbidity Survey Replication. Archives of general psychiatry, 62(6), 593-

602.

Kleih, S. C., & Kübler, A. (2013). Empathy, motivation, and P300 BCI

performance. Frontiers in human neuroscience, 7, 642.

Kleih, S. C., Nijboer, F., Halder, S., & Kübler, A. (2010). Motivation modulates the P300

amplitude during brain–computer interface use. Clinical Neurophysiology, 121(7),

1023-1031.

Kober, S. E. et al. (2013) Learning to modulate one’s own brain activity: the effect of

spontaneous mental strategies. Front. Hum. Neurosci. 7, 695.

Kotchoubey B, Strehl U, Holzapfel S, Blankenhorn V, Fro¨ scher W, Birbaumer N (1999)

Negative potential shifts and the prediction of the outcome of neurofeedback therapy

in epilepsy. Clin Neurophysiol 110(4):683–686.

Lacroix, J. (1981). The Acquisition of Autonomic Control Through Biofeedback: The Case

Against an Afferent Process and a Two‐Process Alternative. Psychophysiology, 18(5),

573-587.

La Greca, A. M., & Lopez, N. (1998). Social anxiety among adolescents: linkages with peer

relations and friendships. Journal of Abnormal Child Psychology, 26, 83–94.

Lang, P. J. (1975). Acquisition of heart rate control: Method, Theory and clinical

implications. In D. C. Fowles (Ed.), Clinical applications of psychophysiology. New

York: Columbia University Press, 167-191.

Laynard, R., Clark, D., Knapp, M., & Mayraz, G. (2007). Cost-benefit analysis of

psychological therapy. National Institute Economic Review, 202(1), 90-98.

Leeb, R., Lee, F., Keinrath, C., Scherer, R., Bischof, H., & Pfurtscheller, G. (2007). Brain–

computer communication: motivation, aim, and impact of exploring a virtual

apartment. IEEE Transactions on Neural Systems and Rehabilitation

Engineering, 15(4), 473-482.

Lenox, B., (2008) in: Cappa, S. F. (2008). Cognitive neurology: A clinical textbook. Oxford:

Oxford University Press.

Linden, D. E. J. (2006). How psychotherapy changes the brain–the contribution of functional

neuroimaging. Molecular psychiatry, 11(6), 528.

Linden, D. E., Habes, I., Johnston, S. J., Linden, S., Tatineni, R., Subramanian, L., . . .

Goebel, R. (2012). Real-time self-regulation of emotion networks in patients with

depression. PLoS One, 7(6), e38115.

Luciano, J. V., Algarabel, S., Tomás, J. M., & Martínez, J. L. (2005). Development and

validation of the thought control ability questionnaire. Personality and Individual

Differences, 38(5), 997-1008.

Marxen, M., Jacob, M. J., Muller, D. K., Posse, S., Ackley, E., Hellrung, L., . . . Smolka, M.

N. (2016). Amygdala Regulation Following fMRI-Neurofeedback without Instructed

Strategies. Frontiers in Human Neuroscience, 10(2016).

Marzbani, H., Marateb, H., & Mansourian, M. (2016). Neurofeedback: A Comprehensive

Review on System Design, Methodology and Clinical Applications. Basic and

Clinical Neuroscience, 7(2), 143-58.

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.

Moher, D., Liberati, A., Tetzlaff, J., Altman, D. G., & Prisma Group. (2009). Preferred

reporting items for systematic reviews and meta-analyses: the PRISMA

statement. PLoS medicine, 6(7), e1000097.

Nan W, Rodrigues JP, Ma J, Qu X, Wan F, Mak PI, Rosa A (2012). Individual alpha

neurofeedback training effect on short term memory. Int J Psychophysiol 86(1):83–

87.

Neuper, C., Scherer, R., Reiner, M., Pfurtscheller, G., 2005. Imagery of motor actions:

differential effects of kinesthetic and visual-motor mode of imagery in single trial

EEG. Brain Research. Cognitive Brain Research 25, 668–677.

Neumann, N., & Birbaumer, N. (2003). Predictors of successful self-control during brain-

computer communication. Journal of Neurology, Neurosurgery & Psychiatry, 74(8),

1117-1121.

Nicolas-Alonso, L., & Gomez-Gil, J. (2012). Brain Computer Interfaces, a Review. Sensors

(Basel, Switzerland), 12(2), 1211-1279

Nicholson, A. A., Rabellino, D., Densmore, M., Frewen, P. A., Paret, C., Kluetsch, R., . . .

Lanius, R. A. (2017). The neurobiology of emotion regulation in posttraumatic stress

disorder: Amygdala downregulation via real‐time fMRI neurofeedback. Human Brain

Mapping, 38(1), 541-560. doi:10.1002/hbm.23402

Nijboer, F., Birbaumer, N., & Kübler, A. (2010). The Influence of Psychological State and

Motivation on Brain–Computer Interface Performance in Patients with Amyotrophic

Lateral Sclerosis – a Longitudinal Study. Frontiers in Neuroscience, 4.

doi:10.3389/fnins.2010.00055

Nijboer, F., Furdea, A., Gunst, I., Mellinger, J., McFarland, D. J., Birbaumer, N., & Kübler,

A. (2008). An auditory brain–computer interface (BCI). Journal of Neuroscience

Methods, 167(1), 43-50.

Niv, S. (2013). Clinical efficacy and potential mechanisms of neurofeedback. Personality

and Individual Differences, 54(6), 676-686.

Paret, C., Kluetsch, R., Zaehringer, J., Ruf, M. Demirakca, T., Bohus, M. . . . Schmahl. C.,

(2016). Alterations of amygdala-prefrontal connectivity with real-time fMRI

neurofeedback in BPD patients. Social Cognitive and Affective Neuroscience, 11(6),

952-960.

Paret, C., Kluetsch, R., Ruf, M., Demirakca, T., Hoesterey, S., Ende, G., & Schmahl, C.

(2014). Down-regulation of amygdala activation with real-time fMRI neurofeedback

in a healthy female sample. Frontiers in behavioral neuroscience, 8, 299.

Rheinberg, F., Vollmeyer, R., & Burns, B. D. (2001). FAM: Ein Fragebogen zur Erfassung

aktueller Motivation in Lern-und Leistungssituationen (Langversion,

2001). Diagnostica, 2, 57-66.

Robineau, F., Meskaldji, D. E., Koush, Y., Rieger, S. W., Mermoud, C., Morgenthaler, S., . . .

Scharnowski, F. (2017). Maintenance of Voluntary Self-regulation Learned through

Real-Time fMRI Neurofeedback. Frontiers in Human Neuroscience, 11.

Royal College of Psychiatrists (2004) Psychotherapy and Learning Disability. Council

Report CR116. London: Royal College of Psychiatrists

Ruiz, S., Lee, S., Soekadar, S. R., Caria, A., Veit, R., Kircher, T., . . . Sitaram, R. (2013).

Acquired self-control of insula cortex modulates emotion recognition and brain

network connectivity in schizophrenia. Hum Brain Mapp, 34(1), 200-212.

Sarkheil, P., Zilverstand, A., Kilian-Hütten, N., Schneider, F., Goebel, R., & Mathiak, K.

(2015). fMRI feedback enhances emotion regulation as evidenced by a reduced

amygdala response. Behavioural Brain Research, 281, 326-332.

Scharnowski, F. et al. (2015) Manipulating motor performance and memory through real-

time fMRI neurofeedback. Biol. Psychol. 108, 85–97.

Scharnowski, F., & Weiskopf, N. (2015). Cognitive enhancement through real-time fMRI

neurofeedback. Current Opinion in Behavioral Science.

Scheinost D, Stoica T, Wasylink S, Gruner P, Saksa J, Pittenger C, Hampson M (2014)

Resting state functional connectivity predicts neurofeedback response. Front Behav

Neurosci 8:338.

Schuhfried G (1989) Wiener Testsystem PC/S. Cognitrone, Mödling, Austria

Sellers, E. W., Arbel, Y., and Donchin, E. (2012). “BCIs that use p300 event-related

potentials,” in Brain-Computer Interfaces: Principles and Practice, eds J. Wolpaw and

E. W. Wolpaw (New York, NY: Oxford University Press), 215–226.

Sepúlveda, P., Sitaram, R., Rana, M., Montalba, C., Tejos, C., & Ruiz, S. (2016). How

feedback, motor imagery, and reward influence brain self-regulation using real-time

fMRI. Hum. Brain Mapp. 37, 3153–3171

Sitaram, R., Ros, T., Stoeckel, L., Haller, S., Scharnowski, F., Lewis-Peacock, J., . . . Sulzer,

J. (2016). Closed-loop brain training: the science of neurofeedback. Nat Rev Neurosci,

18(2), 86-100.

Skinner, B. F. (1945). The operational analysis of psychological terms. Psychol. Rev. 52,

270–277

Sorger, B., Reithler, J., Dahmen, B., & Goebel, R. (2012). A Real-Time fMRI-Based Spelling

Device Immediately Enabling Robust Motor-Independent Communication. Current

Biology, 22(14), 1333-1338.

Sorger, Scharnowski, Linden, Hampson, & Young. (2019). Control freaks: Towards optimal

selection of control conditions for fMRI neurofeedback studies. NeuroImage, 186,

256-265.

Spetter, M. S., Malekshahi, R., Birbaumer, N., Luhrs, M., van der Veer, A. H., Scheffler, K., .

. . Hallschmid, M. (2017). Volitional regulation of brain responses to food stimuli in

overweight and obese subjects: A real-time fMRI feedback study. Appetite, 112, 188-

195.

Spielberger, C., Gorsuch, R., & Luchene, R. (1976). Manual for the State-Trait Anxiety

Inventory. Palo Alto, CA: Consulting Psychologists Press

Spielberger, C.D., (2010). State-Trait Anxiety Inventory. In: the Corsini Encyclopedia of

Psychology. John Wiley & Sons, Inc.

Staunton, G. & Cohen Kadosh, (2019). A systematic review of the psychological factors that

influence neurofeedback learning outcomes. NeuroImage, 185, 545-555.

Subramanian, L., Hindle, J., Johnston, S., Roberts, M., Husain, M., Goebel, R., & Linden, D.

E. J. (2011). Real-time fMRI Neurofeedback for Treatment of Parkinson's Disease.

Journal of Neuroscience, 31(45), 16309-16317.

Thibault, R. T., MacPherson, A., Lifshitz, M., Roth, R. R., & Raz, A. (2018). Neurofeedback

with fMRI: A critical systematic review. Neuroimage, 172, 786-807.

Vidaurre, C., & Blankertz, B. (2010). Towards a Cure for BCI Illiteracy. Brain Topography,

23(2), 194-198.

Wan, F., Nan, W., Vai, M.I. & Rosa, A., (2014) Resting alpha activity predicts learning

ability in alpha neurofeedback. Front Human Neurosci 8:500.

Weber E, Koberl A, Frank S, Doppelmayr M (2010) Predicting Successful learning of SMR

neurofeedback in healthy participants: Methodological considerations. Appl

Psychophysiol Biofeedback 36(1):37–45.

Wechsler, D. (1981). The psychometric tradition: Developing the Wechsler Adult

Intelligence Scale. Contemporary Educational Psychology.

Wechsler, D. (1999). Wechsler Abbreviated Scale of Intelligence. San Antonio, TX: The

Psychological Corporation.

Wells, A., & Davies, M. I. (1994). The Thought Control Questionnaire: A measure of

individual differences in the control of unwanted thoughts. Behaviour research and

therapy, 32(8), 871-878.

Wierzgaaa, P., Zapaaa, D., Wojcik, G., & Masiak, J. (2018). Most Popular Signal Processing

Methods in Motor-Imagery BCI: A Review and Meta-Analysis. Frontiers in

Neuroinformatics, 12, 78.

Witte, M., Kober, S. E., Ninaus, M., Neuper, C., & Wood, G. (2013). Control beliefs can

predict the ability to up-regulate sensorimotor rhythm during neurofeedback

training. Frontiers in human neuroscience, 7, 478.

Wood, G., & Kober, S. (2018). EEG Neurofeedback Is Under Strong Control of Psychosocial

Factors. Applied Psychophysiology and Biofeedback, 43(4), 293-300.

Zich, C., Haller, S. P., Luehrs, M., Lisk, S., Lau, J. Y., & Kadosh, K. C. (2018). Modulatory

effects of dynamic fMRI-based neurofeedback on emotion regulation networks during

adolescence. bioRxiv, 347971.

Zotev, V., Krueger, F., Phillips, R., Alvarez, R. P., Simmons, W. K., Bellgowan, P., ... &

Bodurka, J. (2011). Self-regulation of amygdala activation using real-time fMRI

neurofeedback. PloS one, 6(9), e24522.

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(11), e79184.

Zotev, V., Phillips, R., Yuan, H., Misaki, M., & Bodurka, J. (2014). Self-regulation of human

brain activity using simultaneous real-time fMRI and EEG neurofeedback.

NeuroImage, 85, 985-995.

Zotev, V., Phillips, R., Misaki, M., Wong, C., Wurfel, B., Krueger, F., . . . Bodurka, J.

(2018). Real-time fMRI neurofeedback training of the amygdala activity with

simultaneous EEG in veterans with combat-related PTSD. NeuroImage: Clinical,

19(C), 106-121.

Zuberer, A., Minder, F., Brandeis, D., & Drechsler, R. (2018). Mixed-Effects Modeling of

Neurofeedback Self-Regulation Performance: Moderators for Learning in Children

with ADHD. Neural Plasticity, 2018, 2464310.

List of Appendices to Empirical Paper

Page

Appendix A – Ethical Approval 116

Appendix B – Supplementary Materials to Method 119

Appendix C - Questionnaires 121

Appendix D – Tests for Normality and Outliers 149

Appendix E – Assumptions Testing 153

Appendix F – Significant Correlations between Predictor Variables 156

Appendix G – Significant Correlations between Predictor Variables 158

Appendix AEthical Approval

Study Title: Emotion regulation abilities in adolescence

Internal Reference Number / Short title: ERA study

Med IDREC Ref: MS-IDREC-C2-2015-023Date and Version No: 14.08.2018_v7

Chief Investigator:

Title and name Dr Kathrin Cohen Kadosh

Position Research Fellow

Add1 University of Oxford

Add2 Oxford

Postcode

Tel :

E-mail :

OX2 6AE

01865 271 349

kathrin.cohenkadosh@psy.ox.ac.uk

Investigators: Dr Jennifer Lau, King’s College London

Catharina Zich, University of Oxford

Rhiannon Thompson, University of Oxford

Simona Haller, University of Oxford

Stephen Lisk, King’s College London

Samantha Krausse

Graham Staunton (Trainee Clinical Psychologist)

Joakim Wiencken

Nicola Johnstone

Sponsor: University of Oxford

Funder: European Commission 7th Framework Programme for Research,

Technological Development and Demonstration

Chief Investigator

Signature:

Dr Kathrin Cohen Kadosh

Dr Cohen Kadosh declares that there is no conflict of interest

Confidentiality Statement

This document contains confidential information that must not be disclosed to anyone other

than the authorised individuals from the University of Oxford, the Investigator Team and

members of the Medical Sciences Interdisciplinary Research Ethics Committee (Medical

Sciences IDREC), unless authorised to do so.

Appendix BSupplementary Materials to

Method

61

Localizer task

The localizer task lasted 9 min. and comprised thirty trials. Each trial started with a social

scene presented for four volumes (3.73 s). Scenes depicted negative, rejecting social

situations viewed from the perspective of a (female) protagonist depicted from the back. The

participants were instructed to interpret the scene freely (appraisal). This appraisal event was

followed by a positively valenced interpretative statement (4 volumes, 3.73 s), after which the

same scene was shown again for four volumes (3.73 s). For the duration of the second scene

presentation, participants were instructed to reappraise the social scene in accordance with

the interpretative statement (reappraisal). Finally, participants were asked to rate how much

they were able to change (reappraise) their thoughts and feelings from the first to the second

presentation of the scene. Participants indicated their perceived success on a Likert scale

ranging from no change (1) to much change (4) via button press on each trial. A fixation

cross was presented for one volume between two trials. Stimulus presentation was controlled

via BrainStim 1.1.0.1 (open source stimulation software, Maastricht University,

http://svengijsen.github.io/BrainStim/).

Neurofeedback task – Instructions

All participants were blind to their NF display parameters and were instructed as follows:

“You will see a thermometer with a green rim on the screen. The red bars show how much

the regions that are important for emotions are active. Your job is to get these regions as

active as possible! So try to get this thermometer up as much as possible. Similar to the task

before, try to control your thoughts towards a positive feeling. When the thermometer does

not have a green rim, the thermometer is not working. However, even if the thermometer is

not working, your task will be the same and we are still measuring how much your brain is

active. The two different thermometers will alternate.”

62

Appendix CQuestionnaires

63

Note for Questionnaires

Due to reasons involving copywrite, copies of the questionnaires sued in this study are not included for publishing as an e-thesis. For further information

please contact the researchers.

64

Appendix DTests for Normality and

Outliers

Table 4: Shapiro Wilks Statistics for Normal DataShapiro-Wilk

Statistic df Sig.

65

STAI (trait) .971 21 .756STAI (state) .963 21 .588

SAS .934 21 .164CERQ .970 21 .723TCQ .932 21 .152

TCA-Q .963 21 .578MFQ .892 21 .024PDS .764 21 .000IQ .969 21 .706

Age .847 21 .004Initial FC .953 21 .392

Total Correlations .960 21 .524Learning Slope .970 21 .732

Control .971 21 .747Maturity .947 21 .302

Table 5: Skewness and Kurtosis for non-Normal DataSkewness (SE) Kurtosis (SE)

MFQ .959 (.501) 1.050 (.972)Age -.251 (.501) -1.309 (.972)PDS 1.240 (.464) 1.60 (.902)

Figure 5: Histogram to show distribution of scores on the Mood and Feelings Questionnaire

66

Figure 6: Histogram to show distribution of Age in Years

Figure 7: Histogram to show distribution of scores on the Puberty Development Scale

67

Figure 8: Stem and Leaf Diagram to show Outliers for the Puberty Development Scale

68

Appendix EAssumptions Testing

69

Table 6: KMO and Bartlett’s Test of Sphericity for Principal Components AnalysisKMO and Bartlett's Test

Kaiser-Meyer-Olkin Measure of Sampling Adequacy. .696

Bartlett's Test of Sphericity Approx. Chi-Square 107.255

df 45

Sig. .000

Table 7: Table to show Cook’s Distance for Multiple RegressionsCook’s Distance

Minimum Maximum Mean SD3 Factors with Learning Slope

.000 .481 .066 .119

3 Factors with Performance

.000 .401 .058 .092

Figure 9: Scatterplot for Multiple Regression with Learning Slope

70

Figure 10: Scatterplot for Multiple Regression with Performance

Table 8: Tests for Sphericity with ANOVAs of Leaners and Non-learnersMauchly’s Test of Sphericity

Mauchly’s W

App. Chi Squared

df Sig. Epsilon Greenhouse-Geisser

Huynh- Feldt

Lower Bound

Learners over time

.383 10.287 5 .069 .627 .739 .333

Non-Learners over time

.780 .3161 5 .676 .888 1.000 .333

71

Appendix FSignificant Correlations

between Predictor Variables

Table 9: Significant Correlations between Other Predictor VariablesPearson Correlation

r Sig. (2-tailed) NSTAI (trait) with STAI (state) .765 .000 25

72

STAI (trait) with SAS .633 .000 28STAI (trait) with CERQ -.809 .000 28STAI (trait) with TCQ -.489 .008 28

STAI (trait) with TCA-Q -.843 .000 24STAI (trait) with MFQ .732 .000 26

STAI (state) with CERQ -.625 .001 25STAI (state) with TCA-Q -.748 .000 21STAI (state) with MFQ .510 .009 25

SAS with CERQ -.445 .015 28SAS with TCA-Q -.587 .003 24SAS with MFQ .591 .001 26MFQ with TCQ -.485 .012 26

MFQ with TCA-Q -.662 .001 22MFQ with CERQ -.662 .000 26CERQ with TCQ .569 .002 28

CERQ with TCA-Q .656 .001 24Age with Puberty Development Scale .617 .001 25

Age with IQ-Matrix Reasoning .526 .004 28Puberty Development Scale with IQ-Matrix

Reasoning.499 .011 25

Table 10: Significant Correlations between Predictor Variables and Dependent Variables

Pearson Correlationr Sig. (2-tailed) N

IQ-Verbal Reasoning with Learning Slope .373 .050 28Initial FC with Performance -.560 .002 28

73

Appendix GPrincipal Components

Analysis

Table 11: SPSS output for Eigenvalues of PCA

Total Variance Explained

Component

Initial Eigenvalues

Extraction Sums of Squared

Loadings

Total % of Variance Cumulative % Total % of Variance

74

1 4.491 44.911 44.911 4.491 44.911

2 1.832 18.321 63.232 1.832 18.321

3 1.378 13.777 77.009

4 .638 6.380 83.389

5 .489 4.893 88.281

6 .404 4.041 92.323

7 .323 3.231 95.553

8 .230 2.305 97.858

9 .134 1.342 99.200

10 .080 .800 100.000

Figure 11. Scree Plot of Predictor Variables

Table 12: Forced Two-Component Principal Components Analysis

Component Matrixa

Component

1 2

STAI (trait) .937

75

STAI (state) pre .828

SAS-A .771

CERQ pre -.803

TCA-Q pre -.874

MFQ pre .745

TC-Q pre -.480

PDS .837

Age (years) Testing .641

IQ total .689

Extraction Method: Principal Component

Analysis.a

a. 2 components extracted.

Part Two: Research

A SYSTEMATIC REVIEW OF THE PSYCHOLOGICAL

FACTORS THAT INFLUENCE NEUROFEEDBACK LEARNING

76

OUTCOMES

Word count: 7658

Statement of Journal Choice: This work has been submitted and accepted for publishing by

Neuroimage. Neuroimage had an impact factor within the top 6.5% of journals in 2017. A

substantial amount of fMRI neurofeedback studies are published in this journal. Additionally,

at the time of submission there was a special edition on neurofeedback that was highly

relevant to this research.

School of PsychologyFaculty of Health and Medical Sciences

University of SurreyApril 201

Abstract

Real-time functional magnetic resonance imaging (fMRI)-based neurofeedback represents the

latest applied behavioural neuroscience methodology developed to train participants in the

self-regulation of brain regions or networks. However, as with previous biofeedback

approaches which rely on electroencephalography (EEG) or related approaches such as brain-

77

machine interface technology (BCI), individual success rates vary significantly, and some

participants never learn to control their brain responses at all. Given that these approaches are

often being developed for eventual use in a clinical setting, this represents a significant hurdle

which requires more research. Here we present the findings of a systematic review which

focused on how psychological variables contribute to learning outcomes in fMRI-based

neurofeedback. However, as this is a relatively new methodology, we also considered

findings from EEG-based neurofeedback and BCI. 271 papers were found and screened

through PsycINFO, psycARTICLES, Psychological and Behavioural Sciences Collection, ISI

Web of Science and Medline and 20 were found to contribute towards the aim of this survey.

Several main categories emerged: Attentional variables appear to be of importance to both

performance and learning, motivational factors and mood have been implicated as moderate

predictors of success, while personality factors have mixed findings. We conclude that these

factors should be measured routinely in neurofeedback research. There is a need for better

controlled studies that, for example, manipulate psychological variables and define clear

thresholds for a successful neurofeedback effect. This research will allow interventions to be

developed for non-responders and better selection procedures in future to improve the

efficacy of neurofeedback.

Introduction

Functional magnetic resonance imaging (fMRI)-based neurofeedback is a technique in the

strong tradition of biofeedback approaches that utilises the latest developments of real-time

data processing and pattern analysis to train participants in the self-modulation of neural

networks (e.g. Sitaram et al., 2017). The training is based on the procedure whereby

participants learn to control brain activity in specific brain regions or networks using real-

time signals from their own brain (Johnston, Boehm, Healy, Goebel, & Linden, 2010;

78

Scharnowski & Weiskopf, 2015; Sitaram et al., 2017).

We currently have a number of neurofeedback techniques at our disposal which utilise brain

activity in different ways, three of which are considered in this review. The first involves

fMRI, which works by detecting the concentration of oxygenated and deoxygenated

haemoglobin in the neural vasculature. This is reflective of the metabolic demands of the

underlying neural activation (Attwell & Iadecola, 2002). Real-time fMRI-based

neurofeedback represents a particularly promising approach, due to its high spatial resolution

and whole-brain coverages. It is also suitable for probing deep subcortical structures, which

allows for the extraction of information from distributed activation patterns, and the mapping

of functionally connected networks, important in clinically relevant processes such as

emotional regulation (Sorger, Reithler, Dahmen, & Goebel, 2012).

The second technique uses Electroencephalography (EEG), which measures the signals from

extracellular field potentials in the brain using electrodes over the surface of the scalp. In

real-time EEG, the signals are transformed into frequencies (e.g., delta (0–4 Hz), theta (4–7

Hz), alpha (8–12 Hz), beta (12–30 Hz) and gamma (>30 Hz) bands) before key features of

the signals are extracted (Nicolas-Alonso & Gomez-Gil, 2012). These features can be used to

interpret multivariate patterns in the signal called event-related potentials (ERP) and slow

cortical potentials (SCP). Each approach has unique advantages and disadvantages. For

example, fMRI can access deeper, subcortical areas due to the superior spatial resolution,

which is in millimetres (e.g. Glover, 2011) as opposed to centimetres for EEG (e.g. Babiloni

et al., 2001). EEG however has superior temporal resolution, accurate to milliseconds as

opposed to seconds for fMRI (Weiskopf, 2012). Both EEG and fMRI-based neurofeedback

79

training have both been found to be successful and are sometimes used together (Keynan &

Hendler, 2017; Zotev, Phillips, Yuan, Misaki, & Bodurka, 2014).

The final technique is brain-computer interfacing (BCI), where brain activity is linked up in

direct communication with a device, such as a robotic arm. Typically, BCI utilises EEG

(Jeunet, N’Kaoua, & Lotte, 2016) but theoretically could involve any imaging technique.

EEG setups will often target P300 or event related potentials and sensorimotor rhythms

(SMR), which are oscillations in the 8 to 30 Hz frequency range stemming from the

sensorimotor cortex (Papanicolaou & Cheyne, 2017). Actual movement, as well as movement

observation and imagination, induce a reduction in the amplitude of SMR, which is known as

event-related desynchronization. SMR are often used because they are relatively easy to

influence in an experimental setup (Nicolas-Alonso & Gomez-Gil, 2012).

One of the reasons why neurofeedback has been taken up with enthusiasm by the research

community is that it responds to a huge clinical need for mechanism-driven therapies in

neurology and psychiatry. Advances in neuroimaging and other neuroscience techniques have

produced a wealth of information about the neural networks that can contribute to clinical

disorders (Linden et al 2012) and thus inform treatment (Linden, 2006). Neurofeedback now

offers the opportunity to harness this information to pinpoint and modify dysfunction and

potential compensatory mechanisms in individual participants.

Despite being a new and emerging field, fMRI-based neurofeedback has already been used

successfully to change brain responses in clinical populations, such as participants with

schizophrenia (Ruiz et al., 2013) depression (Linden et al., 2012), obesity (Spetter et al.,

2017), chronic pain (deCharms et al., 2005) and Parkinson’s (Subramanian et al., 2011). In

the emotion processing domain, fMRI-based neurofeedback has proven particularly useful for

up-regulating or down-regulating the brain regions involved in healthy adults’ emotional

80

responses (Johnston et al., 2010; Paret et al., 2014, Zotev et al., 2011, Zotev, Phillips, Young,

Drevets & Bodurka, 2013) and also in children and adolescents (Cohen Kadosh et al., 2016).

The neural changes from fMRI-based neurofeedback are instantaneous but are also

documented to last for up to 14 months in some cases (Megumi et al., 2015, deCharms et

al.2005; Robineau et al., 2017). Critically, it has been shown that lasting effects of training

not only result in the ability to influence one’s own brain activity, but are also related

behavioural changes (Caria, Sitaram, & Birbaumer, 2012).

Despite these overall successes, it has been repeated found that not everyone who receives

neurofeedback is able to influence their brain activity. These participants are often referred to

in the neurofeedback literature as non-responders, non-performers or non-regulators and

represent 30-50% of the population (Alkoby, Abu-Rmileh, Shriki, & Todder, 2017). Note

that there is a similar issue in the BCI literature known as ‘BCI Illiteracy’ in which BCI is

unsuccessful for 10-30% of participants (e.g. Vidaurre & Blankertz, 2010). Given the not

negligible proportion of non-responders, there is a strong need to understand the inter-

individual differences in neurofeedback performance.

It is the aim of this systematic review thus to explore this problem. Specifically, here we will

assess how different psychological and cognitive factors relate to learning outcome in

neurofeedback, and more importantly whether these could help to account for the differences

between responders and non-responders. Factors implicated so far can be divided into

psychological, learning dynamics or neurophysiological factors (Diaz Hernandez, Rieger, &

Koenig, 2016). Variables can be further classified as being inter or intrapersonal (Grosse-

Wentrup & Schölkopf, 2013), fluctuating states or stable traits (Jeunet, N’Kaoua, & Lotte,

2016) and explanatory or correlational variables (Grosse-Wentrup & Schölkopf, 2013). This

review is focused on the psychological factors which relate to the learning outcome. This area

81

is a priority as psychological variables are more likely to be open to intervention. Variables

such as age offer insight, but it is difficult to improve the efficiency of neurofeedback based

on this knowledge other than through selection processes (Grosse-Wentrup & Schölkopf,

2013). It is beyond the scope of this review to explore generic factors manipulated by the

experimenter or other categories of predictor variables mentioned above. For the purposes of

the present study, relevant literature includes any study, experiment or review exploring

psychological variables in relation to neurofeedback outcomes. The psychological variables

considered here will be all those implicated by the literature so far and include attentional,

motivational, mood and personality factors (Alkoby et al., 2017). As the research body is

only just starting to receive due attention and particularly for fMRI-based neurofeedback

(Enriquez-Geppert et al., 2014), a wide range of populations and methodologies will be

considered. That is, we will include both quantitative and qualitative research, using both

EEG- and fMRI-based neurofeedback and also the BCI literature.

With the aim of this study satisfied, it is expected that there will be a clearer picture of where

future research is required to understand the psychological differences between regulators

and non-regulators. This can also begin to inform how psychological variables can potentially

be addressed with tailored neurofeedback interventions to improve outcomes.

Method

A review of all published research up and until 2018 was completed using the online

databases PsycINFO, psycARTICLES, Psychological and Behavioural Sciences Collection,

ISI Web of Science and Medline. We note a descriptive systematic review approach was

adopted here due to the significant variety of the methodologies and approaches used in the

studies reviewed, which makes combining quantitative results problematic (Thibault et al.,

2018). Eight additional papers were identified through researchers in the field. Search terms

82

were chosen to focus the literature search on real-time neurofeedback and BCI. The terms

included were “real-time fMRI”, “real-time EEG”, “BCI”, “neurofeedback”, biofeedback”,

“neurotherapy”, “bioneurofeedback” with both abbreviated and non-abbreviated terms

applied. In addition, search terms were chosen to narrow the focus to psychological

performance and learning variables, which were “factors”, “variable”, “predict”, “learning”,

“response”, “mood”, “depression, “anxiety”, “attention”, “motivation”, “psychological”,

“cognitive”, “personality” and “outcome” which were broadened with the asterisk function in

search engines (e.g. “respon*”). These two sets of search terms were combined with the “or”

function on search engines.

Papers were screened in two phases; initially based on the relevance of their titles and

abstracts allowing for some level of flexibility and then secondly the full body was assessed

(See Figure 1. for a flow chart). However, we noted that often the studies did not include key

findings about psychological variables in abstracts (as they were secondary findings) and so

to counteract this, the method sections of all recent fMRI studies were examined by default.

From this further examination, 33 articles were found that did not contribute to the current

review due to a lack of consideration of psychological predictors. Abstracts were always

examined as papers did not always included secondary findings in the title. Literature reviews

and meta-analyses were always explored in full due to their utility in synthesising evidence.

This approach is susceptible to bias in that the researcher may have naturally varied in their

flexibility of exploring studies further based on titles and abstracts. Papers which cited

neurofeedback or BCI (brain-computer interface) along with variables, predictors, factors,

response or learning outcome were included. Papers which did not mention neurofeedback

were excluded; for example, studies referring to static neuroimaging. Given the relative

novelty of fMRI-based neurofeedback, all methodologies were considered.

83

To review the studies critically, two quality appraisal checklists were used from appendices F

and G of guidance from the National Institute for Health and Care Excellence (NICE, 2012).

These checklists were designed to be applied to research for interventions that will eventually

be used in a clinical setting. This is relevant to neurofeedback as many studies aim to

investigate its therapeutic potential. The checklists assist in the appraisal of group based

quantitative intervention studies (27 item scale) and correlational research (19 item scale)

84

20 full text articles included in qualitative synthesis

33 fMRI full text articles excluded as no

inclusion/mention of psychological predictors

54 full text articles assessed for eligibility

217 records excluded270 abstracts screened

270 records after duplicates removed

10 records identified through other sources

271 records identified through database searching

Figure 1: Flowchart of the systematic review following the procedure set-out in (Moher, Liberati, Tetzlaff,

Altman, & PRISMA group, 2009)

respectively. Each covers five sections: population, method of selection, outcomes, analyses

and summary. Both checklists are useful for neurofeedback research due to many studies

involving a single cohort correlational methodology as well as control group studies. Each

item is rated from being unlikely to be a source of bias (++), a possible source of bias (+) and

a significant source of potential bias (-). When the item cannot be answered from the report,

there are additional responses of not reported (NR) and not applicable (NA).

Results

20 articles were found linking psychological factors to neurofeedback performance. EEG and

BCI literature is reviewed below separately from fMRI based neurofeedback. Implicated

factors are grouped into the categories of motivation, mood, attention, personality and more

isolated research was grouped into the “other factors” category. The results of the analysis of

these studies using critical appraisal tools are summarised in Table 1. The studies are also

indexed for clarity and organised alphabetically, as some studies explore multiple

psychological variables, with EEG and BCI studies represented in Table 2 and fMRI studies

in Table 3.

85

Table 1. Table to Show Ratings for all Studies based on each Item of the NICE (2012) Quality Appraisal Checklists

Table Index, Researchers and

Design

Section One: Population

Section Two: Method of Allocation to Intervention

Section Three: Outcomes Section Four: Analyses Section Five:

Summary

Limitations

2: 1 Burde & Blankertz, (2006): Non-comparative

exploratory research. Correlational

1.1: NR1.2: NR1.3: NR

2.1: +2.2: NR2.3: NA

2.4: -2.5: NA

3.1: +3.2: +3.3: -

3.4: NA3.5: NA

4.1: NR4.2: -4.3: -4.4: +

5.1: +5.2: NR

Few details about participants and selection

bias. Univariate. No confidence intervals

2: 2 Daum et al. (1993). Observational

before and after Study

1.1: +1.2: NR1.3: +

2.1: NR2.2: ++2.3: NA2.4: +

2.5: NA

3.1: ++3.2: ++3.3: +

3.4: NA3.5: NA

4.1: NR4.2: ++4.3: -

4.4: NR

5.1: +5.2: +

No details of selection of participants. Assigned participants by median

split. No confidence intervals or power

discussion, small sample2: 3 Diaz Hernandez, Rieger and Koenig

(2016). Correlational

1.1: ++1.2: NR1.3: +

2.1: -2.2: +

2.3: NA2.4: ++2.5: NA

3.1: ++3.2: +

3.3: ++3.4: NA3.5: NA

4.1: NR4.2: ++4.3: ++4.4: ++

5.1: +5.2: NR

No confidence intervals. Collinear variables

missing, not reported whether representative of

potential treatment populations

2: 4 Enriquez-Geppert et al. (2014).

RCT with pseudo feedback control

group

1.1: +1.2: NR1.3: NR

2.1: ++2.2: ++2.3: +2.4: +

2.5: NR

2.6: ++2.7: ++2.8: ++2.9: NA2.10:NA

3.1: ++3.2: ++3.3: ++

3.4: ++3.5: NA3.6: NA

4.1: ++4.2: NA4.3: NR

4.4: ++4.5: ++4.6: NR

5.1: ++5.2: NR

Missing recruitment information and links to treatment populations and power analysis.

Measures are not robust.

2: 5 Gruzelier et al. (1999). Within Subjects, non-comparative. Correlational

1.1: ++1.2: +1.3: +

2.1: NR2.2: ++2.3: NA2.4: +

2.5: NA

3.1: ++3.2: +3.3: +

3.4: NA3.5: NA

4.1: NR4.2: +4.3: +

4.4: NR

5.1: +5.2: ++

No confidence intervals. No controls for

participants who dropped out. No direct measure of attention for conclusion

drawn

86

2: 6 Hammer, et al. (2012). Correlational,

non-comparative

1.1: +1.2: NR1.3: NR

2.1: NR2.2: +

2.3: NA2.4: +

2.5: NA

3.1: ++3.2: +

3.3: ++3.4: NA3.5: NA

4.1: +4.2: ++4.3: +

4.4: NR

5.1: +5.2: NR

Selection procedure unstated. Lacking details

of participant means.

2: 7 Hardman et al., (1997). Non- randomised

Controlled Trial

1.1: +1.2: NR

1.3: -

2.1: NR2.2: ++2.3: NR2.4: NR2.5: +

2.6: -2.7: -

2.8: ++2.9: NA2.10:NA

3.1: ++3.2: ++3.3: +

3.4: ++3.5: NA3.6: NA

4.1: NR4.2: NR4.3: NR

4.4: NR4.5: ++4.6: +

5.1: -5.2: -

Healthy participants unrepresentative of clinical populations.

Selection and randomisation

unreported2: 8 Jeunet et al.

(2015). Non-comparative, correlational

1.1: ++1.2: NR1.3: NR

2.1: NR2.2: ++2.3: NA2.4: ++2.5: NA

3.1: ++3.2: ++3.3: ++3.4: NA3.5: NA

4.1: NR4.2: ++4.3: ++4.4: ++

5.1: +5.2: NR

No recruitment information or

generalisation to clinical populations possible

2: 9 Kikkert (2015). Non-Comparative

Correlational

1.1: ++1.2: +

1.3: NA

2.1: +2.2: +

2.3: NA2.4: NA2.5: NA

3.1: +3.2: +

3.3: ++3.4: NA3.5: NA

4.1: -4.2: ++4.3: -4.4: -

5.1: -5.2: NA

Exploratory study with small sample and only

visual/descriptive statistic comparisons

made2: 10 Kleih and Kubler (2013).

Comparative Before and After study

1.1: ++1.2: +1.3: -

2.1: NA2.2: ++2.3: ++2.4: -

2.5: ++

2.6: +2.7: ++2.8: ++2.9: +

2.10: NA

3.1: ++3.2: ++3.3: ++

3.4: ++3.5: NA3.6: NA

4.1: ++4.2: NA4.3: NR

4.4: NR4.5: -

4.6: NR

5.1: -5.2: NA

Median split allocation. Compared means rather

than using regression analysis.

2: 11 Kleih et al. 2010. Randomised

Control Trial

1.1: ++1.2: NR

1.3: -

2.1: ++2.2: ++2.3: NR2.4: NR2.5: ++

2.6: ++2.7: ++2.8: ++2.9: NA2.10:NA

3.1: ++3.2: ++3.3: ++

3.4: ++3.5: NA3.6:NA

4.1: ++4.2: NA4.3: NR

4.4: NR4.5: ++4.6: +

5.1: ++5.2: -

Not generalisable to clinical populations.

Motivation at ceiling in all groups. Did not

manipulate motivation.2: 12 Leeb et al. (2007). Within

subjects.

1.1: +1.2: NR1.3: NA

2.1: NR2.2: ++2.3: ++2.4: ++

3.1: -3.2: +3.3: -

3.4: NA

4.1: -4.2: -4.3: -

4.4: NR

5.1: -5.2: NA

Few participants and sourcing undescribed. No

direct quantitative measure of motivation.

87

2.5: NA 3.5: NA Contamination of fatigue2: 13 Nijboer et al.

(2008). Non-Comparative, Correlational

1.1: ++1.2: NR1.3: NR

2.1: NR2.2: ++2.3: NA2.4: ++2.5: NA

3.1: +3.2: ++3.3: ++3.4: NA3.5: NA

4.1: -4.2: ++4.3: ++4.4: +

5.1: +5.2: NA

Quality of life scale lacks reliability and validity details. Small sample with no discussion of

power/confidence2: 14 Nijboer, Birbaumer and

Kubler (2010) Non-Comparative, Correlational

1.1: ++1.2: +

1.3: ++

2.1: NR2.2: ++2.3: NA2.4: ++2.5: NA

3.1: +3.2: ++3.3: ++3.4: NA3.5: NA

4.1: -4.2: ++4.3: ++4.4: -

5.1: +5.2: NA

Quality of life scale not a reliable measure of

anxiety. Small sample. Relies on qualitative

summary2: 15 Witte et al.

(2013). Double Blind RCT

1.1: +1.2: NR1.3: NR

2.1: ++2.2: ++2.3: ++2.4: ++2.5: ++

2.6: ++2.7: ++2.8: ++2.9: NA2.10: NA

3.1: ++3.2: ++3.3: ++

3.4: ++3.5: NA3.6:NA

4.1: ++4.2: NA4.3: +

4.4: +4.5: ++4.6: NR

5.1: +5.2: NR

Not generalisable to clinical populations. No

detials of confidence/power

3: 1 Chiew, Laconte & Graham (2012). Non-randomised Controlled trial

1.1: ++1.2: NR1.3: NR

2.1: NR2.2: ++2.3: NR2.4: NR2.5: ++

2.6: ++2.7: ++2.8: ++2.9: NA2.10: NA

3.1: ++3.2: ++3.3: ++

3.4: ++3.5: NA3.6: NA

4.1: ++4.2: NA

4.3: -

4.4: NR 4.5: ++4.6: +

5.1: +5.2: NR

No details of randomisation or power and confidence for small

sample.

3: 2 Emmert et al., (2017) Non-

comparative, within subjects.

Correlational

1.1: ++1.2: NR1.3: NR

2.1: ++2.2: ++2.3: ++2.4: -

2.5: NA

3.1: ++3.2: ++3.3: +

3.4: NA3.5: NA

4.1: NR4.2: -

4.3: ++4.4: +

5.1: +5.2: NR

No other predictor variables considered.

Correlation statistics not reported

3: 3 Marxen et al. (2016) Non-

comparative, within subjects.

Correlational

1.1: +1.2: NR1.3: NR

2.1: NR2.2: ++2.3: NA2.4: +

2.5: NA

3.1: ++3.2: ++3.3: ++3.4: NA3.5: NA

4.1: NR4.2: ++4.3: ++4.4: +

5.1: +5.2: NR

Four participants with missing data removed. Some contamination of effects of respiration in

BOLD signal

88

3: 4 Nicholson et al. (2017). Non-

comparative, within subjects.

Correlational

1.1: ++1.2: ++1.3: ++

2.1: NR2.2: ++2.3: NA2.4: +

2.5: NA

3.1: ++3.2: ++3.3: ++3.4: NA3.5: NA

4.1: -4.2: ++4.3: ++4.4: +

5.1: +5.2: ++

No inclusion of respiration control. No

power analysis or confidence intervals;

small participant sample3: 5 Zotev et al.

(2011). RCT1.1: ++1.2: NR1.3: NR

2.1: ++2.2: ++2.3: NR2.4: NR2.5: ++

2.6: ++2.7: ++2.8: ++2.9: NA2.10: NA

3.1: ++3.2: ++3.3: +

3.4: ++3.5: NA3.6: NA

4.1: NR4.2: NA4.3: NR

4.4: ++4.5: ++4.6: +

5.1: +5.2: NR

Other confounding variables not controlled for. No reports of power or confidence intervals.

89

Overview of Real-Time EEG Neurofeedback and BCI literature

Attention: Many studies consider attention to be important for neurofeedback. Even when

investigating other variables, such as mood or motivation, studies often attribute the variation

in performance to the effect that theses variables have on the attentional resources of

participants. However, only a handful of studies were found which directly explore this

variable in their analyses. Both attention span and general ability to concentrate are

implicated. For example, Daum and colleagues (1993, Table 2: 2) found participants with

greater control over SCPs had greater scores of attention span (block tapping), digit span and

verbal learning tests from the Wechsler Scale of Intelligence (Wechsler, 1981). Success was

based on the significance of SCP differentiation between baseline and active feedback trials

in terms of performance in each session and learning over the trials and transfer run. This was

a small study which recruited 14 drug-refractory participants with epilepsy and so the power

of the results is questionable. However, in line with these findings, a larger study by Hammer

and colleagues (Hammer et al., 2012, Table 2: 6) measured attention using the Cognitrone

(Schuhfried, 2007); a performance measure of attention and ability to concentrate. This study

focused on SMR to assess the influence of motivation on neurofeedback success. BCI

performance for SMR control was measured through the percentage of correct responses of

moving a cursor using imagined movement of hands and feet. The authors found that the

ability to concentrate on the task accounted for 19% of the variance in performance. Note

though that the runs took a considerable amount of time (up to 5 hours or more) and one

might expect concentration to be of greater importance under such conditions. Both of these

studies lack some details about their participants which leave possible biases unconsidered;

namely how participants were recruited and means obtained for the different questionnaires

used.

90

Rather than measure attention specifically, other studies infer concentration from

performance over neurofeedback trials. These results highlight that attention and fatigue may

be important, but they lack validity without an appropriate measure to back up their

conclusions. Gruzelier and colleagues (Gruzelier, Hardman, Wild & Zaman, 1999, Table 2:

5) designed an EEG-based study to teach the reversal of neurological asymmetry in 25

participants with schizophrenia. They displayed the SCP negativity between hemispheres on

the sensory-motor areas of the brain using an on-screen rocket. Participants were divided into

good and average performers based on their neurofeedback ability. Average performers were

not in fact slower learners, instead they dropped off in performance over time compared to

good performers. This difference was most pronounced in the final block of three (each with

20 trials), leading the researchers to believe that the effects of fatigue on concentration made

the difference between good and average performers, rather than learning ability. Note that

this study had some drop-outs for which controls were not discussed. This may have

introduced some bias to the final sample.

Motivation and Mood: Motivational factors appear to be among the most well-researched

psychological variables in neurofeedback learning outcomes. Motivation can be defined in

various ways, but many BCI studies utilise the QCMBCI, adapted from the Questionnaire for

Current Motivation (Rheinberg, Vollmeyer, Burns, 2001), which has four subscales including

“mastery confidence”, “interest”, “fear of incompetence” and “challenge”. For example,

Nijboer and colleagues (Nijboer et al., 2008, Table 2: 14) used an SMR-BCI setup. Sixteen

University students were instructed to visualise a movement that created their largest event-

related desynchronization. The feedback given was either auditory by generating the correct

sound (harps or bongos) or visual and involved moving a cursor across a screen to hit a target

within a time frame. The authors were interested in how the accuracy of SMR manipulation

91

was influenced by self-reported motivation. Nijboer and colleagues used the QCMBCI before

each of the three training sessions. In visual feedback, they found mastery confidence

correlated with accuracy (target hits), while fear of incompetence was associated with

decreased accuracy. Additionally, they tracked those participants who scored worse over time

and found that their mastery confidence was decreasing, and their fear of incompetence

increasing, along with the deteriorating performance. However, this was not found for the

auditory feedback condition. In fact, a relationship between improved accuracy and fear of

incompetence was found in the auditory condition, suggesting that their anxiety lead to

improved performance. The authors also found that improved mood prior to each training

session correlated positively with improved SMR BCI performance (accuracy of cursor

movements). They used a quality of life measure (Skalen zur Erfassung der Lebensqualität,

SEL,; Averbeck et al., 1997) with ten Likert-scale questions relating to mood in the current

situation. The authors noted that they could not derive a causal relationship of these factors

due to the design of the study being correlational.

The authors followed up this study with similar study, which also included a condition

looking at P300 control (Nijboer, Birbaumer & Kubler, 2010, Table 2: 14). The P300 is an

ERP component that reflects attentional and working memory capacity (Kok, 2001). In this

study, participants had to influence P300 to focus on characters in a letter matrix to copy a

word, while the setup of the SMR BCI matched that of Nijboer et al. (2008). Seven

participants with amyotrophic lateral sclerosis were organised into BCI training protocols for

SMR BCI (1 participant), P300 BCI (3 participants) or both (3 participants). Within the P300

condition, a positive relationship between mastery confidence and letter selection accuracy

was found in one participant whereas a negative relationship between fear of incompetence

and accuracy was found in another. In the SMR group, challenge was positively correlated to

92

accuracy in one participant. No other significant correlations were found including with

mood; contradicting Nijboer et al., (2008). It is interesting that fear of incompetence related

to outcome and anxiety on the measure of mood did not. Both of the above studies had a low

number of participants and lack statistical power. Additionally, the quality of life measure

used to assess mood may not be robust. The latter study actually found significant

correlations between the number of sessions and the psychological variables (e.g. many of the

participants increased in motivational scores over time) and performance also increased in all

SMR participants according to number of sessions. The correlations of psychological

variables and performance may therefore be masked by the greater influence of the number of

sessions.

These smaller studies depict how individual and variable the effects of motivation may be.

An example of a larger study is that of Kleih and colleagues (Kleih, Nijboer, Halder &

Kübler, 2010, Table 2: 11) who explored how self-reported motivation affected the ability of

33 psychology students to influence the P300 wave to spell words with a letter matrix. The

authors found that self-reported motivation on a visual analogue scale (1-10) could predict

21.4% of the variance in P300 amplitude in sessions. They also found that a subscale of the

QCMBCI called “challenge” significantly predicted spelling accuracy in single trials. This

would suggest that participants who perceived the training as more of a challenge performed

better. No other motivational scales predicted accuracy but note that accuracy in this study

averaged at 99% and motivation in general was found to be very high. Therefore, for both

accuracy and motivation (particularly on the “interest” and “mastery confidence” subscales of

the QCMBCI) there were significant ceiling effects. Note that the primary aim of the study

was to manipulate motivation with financial rewards (25-50 cents) for some groups of

participants. However, their rewards had no effect on self-reported motivation. According to

93

Self-Determination Theory (Deci & Ryan, 1985), introducing extrinsic motivation from

financial rewards would only serve to interfere with the high intrinsic motivation reported by

participants in the study (Sarkheil et al., 2015), which could explain the problem.

Perhaps, implicating the "interest” qualities of motivation, Leeb and colleagues (Leeb et al.

2007 Table 2: 12) used a virtual reality BCI task where participants had to navigate through a

virtual apartment by increasing activity in motor regions through motor imagery (measured

through EEG). Participants were previously trained with a cue-based feedback task where

they would be presented with a smiley face on-screen if they responded correctly. All

participants returned to this task following the virtual apartment task. Leeb et al., found that

when participants started the virtual apartment task for the first time their performance (the

number of errors they made) changed. That is, following the change to the virtual apartment,

participants who were initially performing poorly on the cue-based task would perform

better, while participants who were already performing well made more errors at first.

Moreover, when participants returned to the cue-based task in the final session, all

participants performed significantly worse. Participants reported that when they started the

virtual apartment task they found it new and exciting and so Leeb and colleagues attributed

the findings to motivation. Those that performed well were distracted by the new interesting

tasks where those that performed poorly had increased motivation and engagement from it.

Based on learning, one would expect the final session to elicit the least errors as participants

were the most experienced, but this was not the case and performance was worse even than

their first training session. The authors therefore considered that a drop in motivation due to

boredom may be potent enough to override the learning effect. However, the results should

be taken with caution as motivation was never measured, and fatigue may also have had an

impact on performance as participants were tested for several hours.

94

While the QCMBCI appears to be a thorough assessment of motivation, another way of

interpreting the concept of motivation comes from the idea of motivational incongruence

(MI). MI is defined as the extent of mismatch between a person’s goals and motivation and

what one truly perceives that they have achieved in their life (Diaz Hernandez et al., 2016).

Grawe (2004) suggests that those with high MI have internally conflicted processes in their

goal directed behaviour, which may interfere with learning tasks such as neurofeedback. Diaz

Hernandez and colleagues (Diaz Hernandez et al., 2016, Table 2: 3) conducted a study that

involved learning to influence class D microstates (EEG-based potentials, linked to positive

symptoms of psychosis). They used the motivational incongruence scale (Holtforth & Grawe,

2003) as a predictive measure. Participants learned through auditory cue-based feedback (i.e.

participants were instructed to increase the volume of a sound) and learning was measured

both across different sessions and within individual sessions. They found that MI accounted

for 36% of the variance of learning (i.e. increase of microstate D contribution compared to

baseline) between training sessions and 42% of the variance within a session. The authors

concluded that MI is linked to learning in neurofeedback. However, other key variables were

not considered. For example, those whom are less satisfied with their goals may score higher

on depression, which might mean that depression is a better predictor of success. The study

only considered trait anxiety in this respect, which they measured with the state version of the

State-Trait Anxiety Inventory (STAI; Spielberger, 2010) and had no significant findings

(though levels were very low in the group they tested).

Other studies have found no link with motivation or even a negative relationship. For

example, the earlier mentioned study by Hammer and colleagues (Hammer et al., 2012,

Table 2: 6) found no significant link between motivation measured by the QCMBCI and

SMR influence. They suggested that the lack of findings may relate to the fact that their

95

participants (a sample of 83 people mostly in higher education) were more homogenous than

the general population in terms of motivation. Another study that found no motivational

effect was by Kleih and colleagues (Kleih and Kübler, 2013, Table 2:10). They attempted to

manipulate motivation with presentations and used customised motivational questionnaire to

assess intrinsic versus extrinsic motivation for P300-BCI. Participants were grouped into

motivated and unmotivated groups via a median split. Motivated participants viewed an

interesting presentation about BCI, whereas the unmotivated group experienced a boring

presentation. However, the presentation did not influence the motivation of groups according

to scores on a visual analogue scale of motivation (measured before and after the

presentations). Regardless, the intrinsically motivated group remained at a greater mean score

of motivation but did not perform better in P300 motivation. They also performed worse in a

d2 test (Brickenkamp, 2002) after the presentations, which measures both speed and accuracy

of visual scanning. The authors theorised that the motivated group may have consumed

attentional resources during the presentations. Note that the use of a median split where a

range of motivational scores for participants in each group are treated as one value, may have

had less statistical power than a regression analysis.

Another negative result for motivation comes from Enriquez-Geppert and colleagues

(Enriquez-Geppert et al., 2014, Table 2: 4) whom explored the neurofeedback of frontal-

midline theta (fm-theta) waves with EEG through an on-screen coloured square (with red

suggesting an increase in fm-theta as opposed to blue). Two independent raters looked at the

change in fm-theta frequency and rated participants as 1 to 5 on a Likert scale (1 was no

response where 5 was strong response) and if scores of 1 and 2 were agreed upon, the

participant was considered a non-responder. A quarter of the 31 participants did not respond

to training according to this rating but this was not found to be linked to motivation,

96

commitment or task difficulty; with these factors being self-reported according to a 7-point

Likert scale. Instead, the authors suggested it was due to the strategy taken and mid-cingulate

fissurisation interfering with fm-theta. It is also noteworthy that this study found that real or

sham feedback did not affect the motivation of participants. These findings are interesting but

more robust measures and detailed measures (e.g. QCMBCI) are required to draw more valid

conclusions.

In terms of mood, the earlier mentioned study into asymmetry in schizophrenia (Gruzelier,

Hardman, Wild & Zaman, 1999, Table 2: 5) found that depression and anxiety assessed

using the Positive and Negative Symptom Scale (PANSS; Kay, Fiszbein & Opfer, 1987),

correlated negatively with performance. Another study explored mood in relation to

hemispheric functional connectivity (Hardman et al., 1997, Table 2: 7) involving

neurofeedback through an on-screen rocket ship. This time a subset of participants was

instructed to imagine positive emotions to activate the left hemisphere and negative emotions

to activate the right, while the other participants were given no strategy. There were no

differences between groups and both were successful by the third trial in influencing

connectivity. Participants completed the Personality Syndrome Questionnaire (PSQ;

Gruzelier & Kaiser, 1996), which has four subscales including “active”, “withdrawn”,

“unreality” and “differential attentional processes inventory”. Through their observation of

individual data, they found “withdrawal” to link with greater rightward shifts in connectivity

(Hardman et al., 1997, Table 2: 7). The withdrawal dimension includes elements such as

social anxiety or being emotionally withdrawn. This suggests that pre-training emotional

traits can predict the success of certain tasks within neurofeedback, though the findings

within a small sample should be interpreted with caution.

Personality and other factors: Overall there is inconsistent support for personality as an

97

influential variable. For example, in a study reviewed above, Diaz Hernandez et al. (2016,

Table 2: 3) found no relationship with personality (along with “body awareness”) with the

ability to influence microstates. These variables were assessed through the Five-Factor

Personality Inventory (Borkenau and Ostendorf, 1993) and the Body Awareness

Questionnaire (Shields et al., 1989) respectively. As reviewed above, Hammer and

colleagues (Hammer et al., 2012, Table 2: 6) found no personality factors from a measure of

the “Big Five - Plus One” (Holocher-Ertl, Kubinger, & Menghin, 2003) that significantly

predicted performance. The study by Kleih and Kübler (Kleih & Kubler, 2013, Table 2: 10)

is the only study to date which explored the influence of empathy and found P300 amplitudes

were greater in more empathic participants. The authors wondered if empathy for end users

of BCI measured with the Saarbrücker Personality Questionnaire (SPF; Paulus, 2009) may

increase emotional involvement with the study which interfered with the allocation of

attentional resources.

In terms of other variables along with personality, BCI literature has investigated many.

Reviews such as Jeunet, N’Kaoua and Lotte (2017) show the diversity of potentially relevant

variables although many of the studies are correlational and have small samples. It is also

possible that some of the variables implicated are specific to the requirements of SMR BCI.

For example, in an SMR BCI setup with 18 participants, Jeunet and colleagues (Jeunet et al.,

2015, Table 2: 8) found that abstractness (relating to creativity) and mental rotation were

related to SMR BCI success involving imagined movements. The imagination of movement

may specifically demand spatial abilities and creative thought. However, tension, learning

style and the trait of self-reliance also related to BCI success. Similarly, an exploratory study

by Kikkert (Kikkert, 2015, Table 2:9) into neurofeedback for the influence of EEG

frequency bands found the potential influence of a wide array of variables. This included

98

cognitive style, learning style, motivation, perceived session difficulty, the big five

personality traits and mindfulness. The influence of these variables depended on whether the

changes were phasic or tonic, or whether beta enhancement or theta inhibition was the goal.

Locus of control is the final recurring concept in the EEG and BCI literature. Burde and

Blankertz (Burde & Blankertz, 2006, Table 2: 1) used EEG to measure activity in motor

areas and asked 12 participants to imagine movements in the hands and feet within a single

session to move a cursor to the edge of a screen. The number of successful screen hits a

participant had correlated with their scores on a scale of loci of control over technology

(KUT; Beier, 2004). This study was followed by a double blinded randomised controlled trial

(RCT) study by Witte and colleagues (Witte et al., 2013, Table 2: 15) who also used scores

from the KUT as an independent variable. However, they tapped directly into SMR power

and fed this back with the relative size of on-screen bars. They found that participants who

held stronger beliefs that they could influence technology were actually less able to up-

regulate SMR. The researchers suggested that in this particular study, relaxation is an

important pre-condition for learning to synchronise SMR and control confidence may recruit

non-SMR related neural resources. Based on the findings of other studies (e.g. Nijboer et al.

2008, Table 2: 13) it is possible that fear of incompetence may play role as well if poor

performance may lead to greater negative emotions in those with a greater locus of control

and their resulting higher expectations.

99

Table 2. Table to Show EEG/BCI studies Implicating Psychological Variables in Neurofeedback performance

Results Section Index

Researchers N Relevant Independent Variable(s) Dependent Variable(s) Findings

1. Burde & Blankertz, (2006) 12 Participants Locus of Control for technology Motor region activity Locus of control led togreater accuracy

2. Daum et al. (1993) 14 epilepsy patients

Attention span SCP potentials learning Performance improved in those with higher attention span

3. Diaz Hernandez, Rieger and Koenig (2016)

20 healthy participants

Motivational incongruence, Life satisfaction, Body awareness

Personality, trait anxiety

Self-regulation of EEG microstates

Negative correlation between Motivational incongruence and EEG microstate increase

4. Enriquez-Geppert et al. (2014)

31 healthy participants – money reward

Self-reported motivation, commitment and difficulty

SMR Influence No link - greater perceived difficulty in genuine group

5. Gruzelier et al. (1999) 25 Schizophrenic

patients

Anxiety, depression, psychosis Asymmetry reversal Anxiety and depression linked, positive symptoms of psychosis not.

6. Hammer, et al. (2012) 80 Participants Motivation, Degree of concentration SMR Influence Degree of concentration was a significant predictor

7. Hardman et al. (1997) 16 healthy participants

Withdrawal (loneliness, lack of friends, social anxiety, emotional withdrawal)

Cortical asymmetry More asymmetry in right hemisphere – less successful

8. Jeunet et al. (2015) 18 Participants Mental rotation, tension, self-reliance, learning style and abstractness

SMR BCI success Mental rotation, self-reliance, abstractness positively correlate with BCI performance.

Tension negatively related.

9. Kikkert (2015) 34 students Cognitive style, learning style, motivation, perceived session

difficulty, the big five personality traits and mindfulness

Beta enhancement and Theta inhibition

Beta enhancement correlated with learning style, cognitive style, and locus of control.

Theta inhibition correlated with factors such as mindfulness and reward sensitivity.

100

10. Kleih and Kubler (2013) 21 Healthy Students

Motivation and Empathy BCI performance and P300 No link between motivated and unmotivated group and performance. Empathy did have a

moderate effect11. Kleih et al. 2010 33 psychology

studentsIncentive financial Mood, motivation and P300

amplitudeMotivation predicted P300 amplitude. Motivation unrelated to money reward.

12. Leeb et al. (2007) 10 paid volunteers

Motivation BCI performance Positive correlation between variables

13. Nijboer et al. (2008) 16 healthy volunteers

Mood and Motivation BCI performance Mood and motivation predicted performance

14. Nijboer, Birbaumer and Kubler (2010)

6 adults with ALS

Psychological Well-being (Quality of life, depression, mood, motivation)

BCI performance Challenge and mastery confidence related to BCI performance. Incompetence fear

negatively related. Some participants did not relate. Mood unrelated

15. Witte et al. (2013) 20 healthy participants

Locus of control Sensori-motor rhythm up-regulation

Locus of control negatively correlated with performance

101

Overview of Real-Time fMRI Neurofeedback Literature

In fMRI-based neurofeedback research, only a few studies were found which considered

predictor variables. One small study used the advantages of fMRI to explore the influence of

attentional networks and give further understanding of the importance of attention in

neurofeedback. Chiew and colleagues trained participants to influence BOLD activity in

motor areas of the brain during imagined motor movements. They found that activity in task-

related networks could account for 65.6% of variance in right handed people and 17.6% in

left handed fMRI performance (Chiew, Laconte, & Graham, 2012, Table 3: 1). This task-

positive network has been attributed to attention and engagement in the neurofeedback task

and includes the frontal eye fields (premotor), insula and dorsolateral prefrontal areas.

Activity in this network was associated with performance even in the first training session,

which would indicate that initially the neurofeedback success is not a learning effect but an

attentional one. The authors did not use an external measure of attention and though a control

group with sham feedback was used, it is unclear how participants were recruited and

whether there was randomisation.

In the fMRI research into emotional regulation, there has been an interest in the trait of

susceptibility to emotion and coping strategies as well as more common measures of mood.

Marxen and colleagues (Marxen et al., 2016, Table 3: 3) utilised fMRI-based neurofeedback

to teach 35 healthy participants to influence BOLD signal from the bilateral amygdala with

the aim of eventually removing feedback for more long-term learning. Feedback was visual

and involved moving dots on a screen towards a target. Successful learning was defined as a

demonstration of significantly different BOLD signals in the amygdala between up-regulation

attempts and down-regulation. They found that self-rated susceptibility to anger (from the

Emotional Contagion Scale; Doherty, 1997) was negatively correlated with emotional

102

regulation capacity following fMRI-based neurofeedback. State anxiety and depression

measured through the STAI (Kendall et al., 1976) and Beck Depression Inventory (Beck et

al., 1996), were not found to have a statistically significant influence. However, in terms of

personality, they found that agreeableness as measured through the NEO-Five Factor

Inventory (Costa and Mccrae, 1992) was negatively correlated with regulation capacity,

though only when feedback was not present. The authors did not suggest an explanation, but

it is possible that similar to empathy, those who are agreeable had a desire to please which

was consuming attentional resources. Finally, the authors found that participants’ ratings of

how well they manage emotion using the Emotional Regulation Scale (Gross & John, 2003)

correlated with regulation capacity post fMRI-based neurofeedback training; though only

when feedback was present. There was a potential contaminating variable in this study in that

there was a significant increase in rates of respiration from non-feedback to feedback runs,

which is known to link to an increase of the BOLD signal and so the results require

confirmation.

Another similar fMRI study by Zotev and colleagues (Zotev et al., 2011, Table 3: 5)

investigated amygdala regulation through the use of positive autobiographical memories with

28 healthy volunteers. This study included both an experimental and sham feedback control

condition and participants were matched on variables such as education. They found that

participants who rated themselves as more susceptible to anger exhibited less BOLD

activation in the left amygdala during training. Susceptibility may have a downside in that

some people with high susceptibility or empathy may experience vicarious emotion that is

difficult for them to regulate. It is also noteworthy that both of the above studies contradicted

each other in their findings about the ability to label emotions. That is, Marxen and

colleagues (Marxen et al., 2016, Table 3: 3) found no correlation of amygdala regulation

103

with scores on the “difficulty identifying feelings” scale (Toronto Alexithymia Scale; Bagby

et al., 1986), whereas Zotev and colleagues found with the same scale that those with higher

difficulty identifying feelings, achieved less amygdala regulation from training (Zotev et al.

2011 Table 3: 5). It is possible that the strategy provided (to recall positive memories)

requires better labelling of emotions and ting memories to emotions. Note that Zotev and

colleagues did not measure other predictor variables in their study and so other factors apart

from susceptibility may be involved.

Interestingly, emotional difficulties do not appear to be a hindrance for neurofeedback in

every case. For example, Nicholson and colleagues used fMRI-based neurofeedback to

explore down-regulation of amygdala regulation in 10 participants with post-traumatic stress

disorder (PTSD) following exposure to trauma trigger words. Participants completed three

training sessions and a transfer run without feedback given. Symptom severity was measured

with the Clinically Administered PTSD Scale (Blake et al., 1995). In sessions one and three,

they found that participants with more severe post-traumatic stress disorder symptoms could

achieve greater down-regulation of amygdala activity (Nicholson et al., 2017, Table 3: 4). It

is possible then, that there is more to gain when there are already neurological emotional

regulation deficits. However, it should be noted that this study did not report control for

respiration or movement, which are critical contaminating variables in fMRI research

(Thibault et al., 2018). It is not farfetched to imaging for example, that people of higher

anxiety might move or breathe more in a novel situation and thus appear to increase BOLD

activity more than other participants.

Finally, there is also some evidence that specific pre-existing coping strategies that people

have can predict their success in neurofeedback training. For example, in an fMRI-based

neurofeedback study on coping with pain through heat stimulation, Emmert and colleagues

104

(Emmert et al., 2017, Table 3: 2) assessed self-reported scores of the frequency of active pain

coping strategies measured with the Coping Strategies Questionnaire (Rosenstiel & Keefe,

1983). They found that the use of active pain coping strategies predicted the success of

participants at regulating pain-sensitive regions of the brain, though the statistics for this

association are not reported.

Summary of Findings across the Literature

To conclude, the ability to concentrate and sustain attention is worthy of further research as a

predictor of success in neurofeedback. Attentional abilities should be established prior to

training. It would be helpful to research adapted programmes for people with attentional

difficulties; for example, with shorter sessions or more simple training setups and tasks with

minimal distractors. However, other psychological variables such as motivation are likely to

also influence attention. As such a profile of each participant should be established before

considering how to maximise the efficiency of their training. To monitor motivation,

comprehensive scales such as the QCMBCI should be utilised rather than generic Likert

scales as motivation consists of various factors which may differ in the individual. For

example, participants may be motivated in terms of interest but have little mastery

confidence. Generally, participant motivation in this field of research is high which means

that manipulation of this variable is challenging. There have been successful attempts in the

past to manipulate motivation with monetary reward (e.g. Shibata et al., 2011; Cortese et al.,

2016) and manipulating how interesting the neurofeedback task is in itself may be useful as

in Leeb et al. (2007, Table 1: 12).

With the inconsistency of the results of mood research, it is possible that mood affects

performance in a task-specific way. This area requires further investigation and it is

recommended that researchers consider using robust measures of current mood and anxiety

105

rather than Likert scales (unless used in combination) to ensure adequate reliability and

validity. It seems likely that anxiety can impair training success across a range of

neurofeedback set-ups and it should therefore be monitored regularly through training.

Interventions for anxious non-responders could include offering reassurance and anxiety

management techniques such as relaxed breathing or progressive muscle relaxation (Roth &

Pilling, 2007). Susceptibility to emotion also appears significant to emotional regulation

neurofeedback studies and so should not be overlooked in this branch of neurofeedback

research.

Personality factors such as the “big five” have yet to show consistent evidence of their effect

on neurofeedback. Cases where the factors have been found to be associated with

performance are also possibly explained in terms of anxiety and motivational factors and the

consumption of attentional resources and so multiple variables should be examined. It is also

worth considering other factors which may be specific to the research being carried out. This

is exemplified by the significance of individual pain management strategies in a

neurofeedback study to do with coping with pain (Emmert et al., 2017, Table 2: 2). Finally, it

is worth noting that all of the variables considered above may relate differently the

neurofeedback success, depending on strategies given to participants or the length of the

intervention amongst other methodological variations.

106

107

Table 3. Table to Show fMRI studies Implicating Psychological Variables in NeurofeedbackResults Section Index

Researchers N Relevant Independent Variable(s)

Dependent Variable(s) Findings

1. Chiew, Laconte & Graham (2012)

13 Participants BOLD signal for task related networks (attention)

BOLD signal in motor areas

Neurofeedback success is attentional rather than a learning effect

2. Emmert et al., (2017) 28 Healthy Participants

Self-reported Pain coping ability Pain sensitive region regulation

Active coping strategies predicted neurofeedback success

3. Marxen et al. (2016) 35 Healthy participants

Susceptibility to anger Amygdala regulation Positive correlation between variables

4. Nicholson et al. (2017) 10 PTSD patients Symptom Severity Amygdala regulation Greater symptom severity lead to greater down-regulation

5. Zotev et al. (2011) 28 Volunteers Identifying feelings, Susceptibility to anger

Amygdala regulation Difficulty identifying emotions and susceptibility to anger both

negatively correlated to amygdala regulation

108

Discussion

As a result of significant advances in fMRI-based neurofeedback technologies, the field of

neurofeedback has experienced a surge in research activity and publications (Sulzer et al.,

2013). However, it has also been shown that not all participants benefit from the training and

it appears that comparative to BCI and EEG literature, there are fewer studies which explore

the predictors of success in fMRI-based neurofeedback. Here we conducted a literature

review to explore the psychological variables that are implicated as being influential in

neurofeedback outcomes. Specifically, we looked at attention, motivation, mood and

personality factors along with any other evidence for unique factors such as “body

awareness”. Overall, the research is in a preliminary state, but findings so far suggest that

there are moderate psychological influences. The influence of these variables is complex. For

example, it is not simply a case that better mood relates to better learning; fear of

incompetence can lead to improved performance in some cases while it is detrimental in

others (Nijboer et al., 2008). Attention was found to have the most consistent positive

influence on learning and performance in neurofeedback tasks across fMRI and EEG. All

four studies concerning this variable found such a result.

Out of the eight studies that looked at motivational factors, five found at least some influence

of motivational factors on performance or learning. For the most part, it seems motivational

factors can increase the focus on the task. In this way motivation probably relates at least in

part to the distribution of attentional resources. Other than general Likert ratings of

motivation, more specific variables implicated to improve performance include interest in the

task, confidence in mastering the task and the challenge perceived in the task (Nijboer et al.,

2008; Nijboer, Birbaumer & Kubler, 2010; Kleih et al., 2010). On the other hand, fear of

109

being incompetent in a study and incongruence in goal-seeking behaviour are predictive of

poorer performance and learning (Nijboer, Birbaumer & Kubler, 2010; Diaz Hernandez,

Rieger & Koenig, 2016). Motivation generally is reported to be high by people taking part in

neurofeedback studies, possibly due to how participants are recruited (often through self-

selection) and the because of the treatment potential for participants from clinical

populations.

Regarding mood and emotional factors, anxiety and depression are implicated to have a

negative influence on performance if at all (Gruzelier, Hardman, Wild & Zaman, 1999;

Hardman et al. 1997; Nijboer et al., 2010). The susceptibility people have to the emotion of

others also appears to play a part in emotional regulation learning. For example, susceptibility

to anger has been found to link to impaired emotional regulation in two fMRI-based

neurofeedback studies (Marxen et al., 2016; Zotev et al., 2007). Moreover, under specific

circumstances, the ability to label emotions may be important in emotional regulation

learning (Marxen et al., 2016).

For personality and other variables there were several other possible influential variables

implicated. Personality is mostly found to have inconsistent influence, with few significant

findings found. However, agreeableness has been found to negatively relate to emotional

regulation learning with fMRI-based neurofeedback (Marxen et al. 2016). Two studies found

locus of control to be significantly influential but in contradictory ways; one finding locus of

control over technology to improve performance (Burde &Blankertz, 2006) and the other

found it made it made performance worse (Witte et al., 2013). Empathy was also found in

one study to negatively influence performance (Kleih & Kubler, 2013). Also coping

110

strategies for pain predicted success in reducing activity in pain sensitive regions (Emmert et

al., 2017).

The variety in variables targeted for learning or performance in neurofeedback and BCI

research is extensive. Even within the current survey are studies looking at hemispheric

asymmetry, microstates, SMR, BOLD levels or P300 amplitude. In addition, different areas

or circuits of the brain are often utilised as targets for certain activity to be modified and

methodologies and training protocols vary significantly in length, style of feedback and

whether there are control groups. The variety of research carried out does offer a degree of

cross-methodological validity. It is certainly ever more convincing that people are able to

modify a remarkable variety of neurological activity through training.

Overall, the research into psychological variables is currently limited. Studies often have

small samples, which may reflect a wider issue with the funding of this type of research.

However additionally, many studies do not comment on statistical power which may be very

low, and do not report effect sizes, confidence intervals or details of the selection, allocation

and recruitment of participants. Larger studies may not be required to gain useful insights

into this topic. Qualitative evidence from interviews in one of the smaller studies has

captured interesting nuances in training performance specifically in relation to the

psychology of individuals along their course of training (Nijboer et al., 2010). Such work

highlights that research into motivation should consider the potentially multi-collinear

relationship between psychological variables and how they change through a course of

neurofeedback. In light of this, a variety of measures should be employed habitually by

researchers. At minimum it is recommended that measures of attention, motivation and mood

are administered along with any task-relevant variables.

111

The literature is generally lacking detail in its exploration of non-responders as a group and

would be helpful to understand this population better. It is unclear what difficulties non-

responders face and what their experience of the training is like in comparison to those who

are successful (e.g. what strategies are they using). There is also a question as to whether

barriers to responding can be overcome or whether better selection of participants is required.

It has not yet been tested whether non-responding participants in one study would also not

respond to training in another under different methodologies. It also seems to be unclear as to

what the consensus is as to which participants are considered to be “non-responders” and for

example, what the threshold level of successful response is (Emmert et al., 2016) and whether

this concerns performance in a single session compared to learning from neurofeedback over

several sessions. Some BCI studies set a 70% accuracy threshold to differentiate successful

and non-successful BCI (Alkoby et al., 2017). Alternatively, studies have defined success as

the achievement of significant differences in responses between two conditions, such as

differences in BOLD signals between up and down regulation conditions (e.g. Marxen et al.,

2016) or differences in P300 amplitude from a control group (Kleih et al., 2010). Also,

Gruzelier and colleagues (1999) used two independent raters to judge average from good

performers of hemispheric shifts in negativity, with the aid of graphs. There are therefore a

variety of methods for defining success which will also vary based on the methodology of the

study and grey areas in between these thresholds. The consideration of non-responders as a

single homogenous group at this stage is likely to be overly simplistic.

To conclude, in the future, high quality research is necessary to allow conclusions to be

drawn confidently regarding psychological factors that predict response to neurofeedback.

This means that measures used to assess attention, motivation, mood, and so on should have

112

good reliability and validity and be sensitive to pick up the nuances of such variables. There

needs to be recognition that psychological variables will vary over time and so need to be

monitored throughout training, for example with the use of tracking measures. It would also

be beneficial to consider ways to manipulate these variables in addition to correlational

research. Also, researchers should consider and distinguish between performance and

learning during neurofeedback training and how psychological variables relate to both.

Finally, researchers should consider trialling interventions to reduce the effect of inhibitory

psychological variables on neurofeedback success to increase the efficiency of this promising

technique.

113

References

Alkoby, O., Abu-Rmileh, A., Shriki, O., & Todder, D. (2017). Can we predict who will

respond to neurofeedback? A review of the inefficacy problem and existing predictors

for successful EEG neurofeedback learning. Neuroscience.12.050

Attwell, D., & Iadecola, C. (2002). The neural basis of functional brain imaging signals.

Trends Neurosci, 25(12), 621-625.

Averbeck, M., Brieger, M., Grote-Kusch, M., Leiberich, P., Olbrich, E., Schröder, A.

&Schumacher, K. (1997). Skalen zur Erfassung der Lebensqualität (SEL).. Swets &

Zeitlinger B.V., Frankfurt: Swets Test Services

Babiloni F., Cincotti F., Carducci F., Rossini P.M., Babiloni C. (2001). Spatial enhancement

of EEG data by surface Laplacian estimation: the use of magnetic resonance imaging-

based head models. Clin. Neurophysiol.112(5):724–727.

Bagby, M., Taylor, G. J., & Ryan, D. (1986). Toronto Alexithymia Scale: Relationship with

personality and psychopathology measures. Psychotherapy and

Psychosomatics, 45(4), 207-215.

Beck, A.T., Steer, R.A., Ball, R., and Ranieri, W.F. (1996). Comparison of beck depression

inventories -IA and-II in psychiatric outpatients. J. Pers. Assess. 67, 588–597.

Blake, D.D., Weathers, F.W., Nagy, L.M., Kaloupek, D.G., Gusman, F.D., Charney, D.S.,

Keane, T.M. (1995): The development of a Clinician-Administered PTSD Scale. J

Trauma Stress 8:75–90.

Borkenau P, Ostendorf F (1993) NEO-Fu¨ nf-Faktoren Inventar (NEOFFI) nach Costa und

McCrae [NEO Five-Factor Personality Inventory (NEO-FFI) according Costa and

McCrae]. Go¨ ttingen: Hogrefe.

114

Beier, G. (2004). Kontrollüberzeugungen im Umgang mit Technik: ein

Persönlichkeitsmerkmal mit Relevanz für die Gestaltung technischer Systeme

(Doctoral thesis). Available from GESIS database, Record No. 20040112708.

Brickenkamp, R. (2002). Test d2: Aufmerksamkeits-Belastungs-Test (9., überarbeitete und

neu normierte Auflage). Göttingen: Hogrefe

Burde, W., & Blankertz, B. (2006). Is the locus of control of reinforcement a predictor of

brain-computer interface performance?. In Proceedings of the 3rd International

Brain-Computer Interface Workshop and Training Course (Vol. 2006, pp. 108-109).

Caria, A., Sitaram, R., & Birbaumer, N. (2012). Real-time fMRI: a tool for local brain

regulation. The Neuroscientist: a review journal bringing neurobiology, neurology

and psychiatry, 18(5), 487.

Chiew, M., Laconte, S. M., & Graham, S. J. (2012). Investigation of fMRI neurofeedback of

differential primary motor cortex activity using kinesthetic motor imagery.

NeuroImage, 61(1), 21-31.

Cohen Kadosh, K., Luo, Q., de Burca, C., Sokunbi, M. O., Feng, J., Linden, D. E., & Lau, J.

Y. (2016). Using real-time fMRI to influence effective connectivity in the developing

emotion regulation network. Neuroimage, 125, 616-626.

Cortese, A., Amano, K., Koizumi, A., Kawato, M., & Lau, H. (2016). Multivoxel

neurofeedback selectively modulates confidence without changing perceptual

performance. Nature communications, 7, 13669.

Costa, P.T. Jr.,and Mccrae, R.R.(1992).The 5-factor model of personality and its relevance to

personality-disorders. J. Pers. Disord. 6, 343–359.

115

Daum, I., Rockstroh, B., Birbaumer, N., Elbert, T., Canavan, A., & Lutzenberger, W. (1993).

Behavioural treatment of slow cortical potentials in intractable epilepsy:

neuropsychological predictors of outcome. Journal of Neurology, Neurosurgery

&amp; Psychiatry, 56(1), 94.

deCharms, R. C., Maeda, F., Glover, G. H., Ludlow, D., Pauly, J. M., Soneji, D., . . . Mackey,

S. C. (2005). Control over brain activation and pain learned by using real-time

functional MRI. Proceedings of the National Academy of Sciences, 102(51), 18626-

18631.

Deci, E. L., & Ryan, R. M. (1985). The general causality orientations scale: Self-

determination in personality. Journal of research in personality, 19(2), 109-134.

Diaz Hernandez, L., Rieger, K., & Koenig, T. (2016). Low motivational incongruence

predicts successful EEG resting-state neurofeedback performance in healthy adults.

Neuroscience. doi:10.1016/j.neuroscience.2016.12.005

Doherty, R. W. (1997). The emotional contagion scale: A measure of individual

differences. Journal of nonverbal Behavior, 21(2), 131-154.

Enriquez-Geppert, S., Huster, R. J., Scharfenort, R., Mokom, Z. N., Zimmermann, J., &

Herrmann, C. S. (2014). Modulation of frontal-midline theta by

neurofeedback. Biological psychology, 95, 59-69.

Emmert, K., Breimhorst, M., Bauermann, T., Birklein, F., Rebhorn, C., Van De Ville, D., &

Haller, S. (2017). Active pain coping is associated with the response in real-time

fMRI neurofeedback during pain. Brain Imaging and Behavior, 11(3), 712-721.

116

Emmert, K., Kopel, R., Sulzer, J., Brühl, A. B., Berman, B. D., Linden, D. E. J., . . . Haller, S.

(2016). Meta-analysis of real-time fMRI neurofeedback studies using individual

participant data: How is brain regulation mediated? NeuroImage, 124, 806-812.

Glover, G. H. (2011). Overview of functional magnetic resonance imaging. Neurosurgery

Clinics, 22(2), 133-139.

Grawe K. 2004. Psychological therapy. Seattle.

Gross, J. J., & John, O. P. (2003). Individual differences in two emotion regulation processes:

implications for affect, relationships, and well-being. Journal of personality and

social psychology, 85(2), 348.

Grosse-Wentrup, M., & Schölkopf, B. (2013). A review of performance variations in SMR-

based Brain− Computer interfaces (BCIs). In Brain-Computer Interface

Research (pp. 39-51). Springer, Berlin, Heidelberg.

Gruzelier, J., Hardman, E., Wild, J., & Zaman, R. (1999). Learned control of slow potential

interhemispheric asymmetry in schizophrenia. International Journal of

Psychophysiology, 34(3), 341-348.

Gruzelier, J. H., & Kaiser, J. (1996). Syndromes of schizotypy and timing of

puberty. Schizophrenia Research, 21(3), 183-194.

Hammer, E. M., Halder, S., Blankertz, B., Sannelli, C., Dickhaus, T., Kleih, S., . . . Kübler,

A. (2012). Psychological predictors of SMR-BCI performance. Biological

Psychology, 89(1), 80-86.

Hardman, E., Gruzelier, J., Cheesman, K., Jones, C., David Liddiard, H., Schleichert, N., &

Birbaumer, N. (1997). Frontal interhemispheric asymmetry: self-regulation and

individual differences in humans. Neuroscience Letters, 221(2), 117-120.

117

Holocher-Ertl, S., Kubinger, K.D., Menghin, S., 2003. Big Five Plus One

Persönlichkeitsinventar (B5PO). Test: Software und Manual. Mödling: Wiener

Testsystem/Schuhfried.

Holtforth MG, Grawe K (2003) Der Inkongruenzfragebogen (INK). Zeitschrift für Klinische

Psychologie und Psychotherapie 32:315–323.

Jeunet, C., N’Kaoua, B., Subramanian, S., Hachet, M., & Lotte, F. (2015). Predicting mental

imagery-based BCI performance from personality, cognitive profile and

neurophysiological patterns. PloS one, 10(12), e0143962.

Jeunet, C., N’Kaoua, B., & Lotte, F. (2016). Advances in user-training for mental-imagery-

based BCI control: Psychological and cognitive factors and their neural correlates.

In Progress in brain research (Vol. 228, pp. 3-35). Elsevier.

Jeunet, C., N’Kaoua, B., & Lotte, F. (2017). “Towards a cognitive model of MI-BCI user

training”, 7th international BCI conference. Retrieved from

https://camillejeunet.wordpress.com/publications/

Johnston, S., Boehm, S., Healy, D., Goebel, R., & Linden, D. (2010). Neurofeedback: A

promising tool for the self-regulation of emotion networks. Neuroimage, 49, 1066-

1072.

Kay, S. R., Fiszbein, A., & Opfer, L. A. (1987). The positive and negative syndrome scale

(PANSS) for schizophrenia. Schizophrenia bulletin, 13(2), 261.

Kendall, P.C., Finch, A.J. Jr., Auerbach, S.M., Hooke, J.F., and Mikulka, P.J. (1976). State-

trait anxiety inventory: a systematic evaluation. J. Consult. Clin. Psychol. 44, 406–

412.

118

Keynan, J., & Hendler, T. (2017). Amygdala-Neurofeedback Reduces Traumatic Stress

Vulnerability, Biological Psychiatry, Vol. 81, pp. S157-S158.

Kikkert, A. (2015). Predictors of neurofeedback efficacy: An exploratory study to the

influence of personality and cognitive characteristics on the efficacy of theta and beta

neurofeedback training. Retrieved from

https://openaccess.leidenuniv.nl/bitstream/handle/1887/32590/Kikkert-Anneke-

s1183613-ReMa%20thesis%20CNP-2015.pdf?sequence=1

Kleih, S. C., & Kübler, A. (2013). Empathy, motivation, and P300 BCI

performance. Frontiers in human neuroscience, 7, 642.

Kleih, S. C., Nijboer, F., Halder, S., & Kübler, A. (2010). Motivation modulates the P300

amplitude during brain–computer interface use. Clinical Neurophysiology, 121(7),

1023-1031.

Kok, A. (2001). On the utility of P3 amplitude as a measure of processing capacity.

Psychophysiology, 38, 557-577.

Leeb, R., Lee, F., Keinrath, C., Scherer, R., Bischof, H., & Pfurtscheller, G. (2007). Brain–

computer communication: motivation, aim, and impact of exploring a virtual

apartment. IEEE Transactions on Neural Systems and Rehabilitation

Engineering, 15(4), 473-482.

Linden, D. E. J. (2006). How psychotherapy changes the brain–the contribution of functional

neuroimaging. Molecular psychiatry, 11(6), 528.

Linden, D. E., Habes, I., Johnston, S. J., Linden, S., Tatineni, R., Subramanian, L., . . .

Goebel, R. (2012). Real-time self-regulation of emotion networks in patients with

depression. PLoS One, 7(6), e38115.

119

Marxen, M., Jacob, M. J., Muller, D. K., Posse, S., Ackley, E., Hellrung, L., . . . Smolka, M.

N. (2016). Amygdala Regulation following fMRI-Neurofeedback without Instructed

Strategies. Frontiers in Human Neuroscience, 10(2016).

doi:10.3389/fnhum.2016.00183.

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.

Moher, D., Liberati, A., Tetzlaff, J., Altman, D. G., & Prisma Group. (2009). Preferred

reporting items for systematic reviews and meta-analyses: the PRISMA

statement. PLoS medicine, 6(7), e1000097.

National Institute for Health and Care Excellence (2012). Methods for the development of

NICE public health guidance (third edition). Retrieved

from https://www.nice.org.uk/process/pmg4/chapter/appendix-f-quality-appraisal-

checklist-quantitative-intervention-studies

Nicolas-Alonso, L., & Gomez-Gil, J. (2012). Brain Computer Interfaces, a Review. Sensors

(Basel, Switzerland), 12(2), 1211-1279.

Nicholson, A. A., Rabellino, D., Densmore, M., Frewen, P. A., Paret, C., Kluetsch, R., . . .

Lanius, R. A. (2017). The neurobiology of emotion regulation in posttraumatic stress

disorder: Amygdala downregulation via real‐time fMRI neurofeedback. Human Brain

Mapping, 38(1), 541-560. doi:10.1002/hbm.23402

Nijboer, F., Birbaumer, N., & Kübler, A. (2010). The Influence of Psychological State and

Motivation on Brain–Computer Interface Performance in Patients with Amyotrophic

120

Lateral Sclerosis – a Longitudinal Study. Frontiers in Neuroscience, 4.

doi:10.3389/fnins.2010.00055

Nijboer, F., Furdea, A., Gunst, I., Mellinger, J., McFarland, D. J., Birbaumer, N., & Kübler,

A. (2008). An auditory brain–computer interface (BCI). Journal of Neuroscience

Methods, 167(1), 43-50. doi:10.1016/j.jneumeth.2007.02.009

Papanicolaou, A., & Cheyne, D. (2017). Magnetoencephalography and Magnetic Source

Imaging. In The Oxford Handbook of Functional Brain Imaging in Neuropsychology

and Cognitive Neurosciences (Chapter 6). Oxford University Press.

Paret, C., Kluetsch, R., Ruf, M., Demirakca, T., Hoesterey, S., Ende, G., & Schmahl, C.

(2014). Down-regulation of amygdala activation with real-time fMRI neurofeedback

in a healthy female sample. Frontiers in behavioral neuroscience, 8, 299.

Paulus,C.(2009). Der Saarbrücker Per- sönlichkeitsfragebogen SPF(IRI) zur Messungvon

Empathie: Psychome- trische. Evaluationder deutschen Version des Interpersonal

Reactivity Index. Saarbrücken:University of Saarland.

Rheinberg, F., Vollmeyer, R., & Burns, B. D. (2001). FAM: Ein Fragebogen zur Erfassung

aktueller Motivation in Lern-und Leistungssituationen (Langversion,

2001). Diagnostica, 2, 57-66.

Robineau, F., Meskaldji, D. E., Koush, Y., Rieger, S. W., Mermoud, C., Morgenthaler, S., . . .

Scharnowski, F. (2017). Maintenance of Voluntary Self-regulation Learned through

Real-Time fMRI Neurofeedback.(functional magnetic resonance imaging)(Report)

(Author abstract). Frontiers in Human Neuroscience, 11.

doi:10.3389/fnhum.2017.00131

121

Rosenstiel, A. K., & Keefe, F. J. (1983). The use of coping strategies in chronic low back

pain patients: relationship to patient characteristics and current

adjustment. Pain, 17(1), 33-44.

Roth, A.D., & Pilling, S. (2007). The Competences Required to Deliver Effective Cognitive

and Behavioural Therapy for People with Depression and with Anxiety Disorders.

London, UK: Department of Health. Retrieved

from: https://www.ucl.ac.uk/.../cognitive-and-behavioural-therapy.

Ruiz, S., Lee, S., Soekadar, S. R., Caria, A., Veit, R., Kircher, T., . . . Sitaram, R. (2013).

Acquired self-control of insula cortex modulates emotion recognition and brain

network connectivity in schizophrenia. Hum Brain Mapp, 34(1), 200-212. doi:Doi

10.1002/Hbm.21427

Sarkheil, P., Zilverstand, A., Kilian-Hütten, N., Schneider, F., Goebel, R., & Mathiak, K.

(2015). fMRI feedback enhances emotion regulation as evidenced by a reduced

amygdala response. Behavioural Brain Research, 281, 326-332.

doi:10.1016/j.bbr.2014.11.027

Scharnowski, F., & Weiskopf, N. (2015). Cognitive enhancement through real-time fMRI

neurofeedback. Current Opinion in Behavioral Science.

Schuhfried G (1989) Wiener Testsystem PC/S. Cognitrone, Mödling, Austria

Shibata, Kazuhisa, Watanabe, Takeo, Sasaki, Yuka, & Kawato, Mitsuo. (2011). Perceptual

learning incepted by decoded fMRI neurofeedback without stimulus presentation.

Science, 334(6061), 1413-1415.

Shields SA, Mallory ME, Simon A (1989) The body awareness questionnaire - reliability and

validity. J Pers Assess 53:802–815.

122

Sitaram, R., Ros, T., Stoeckel, L., Haller, S., Scharnowski, F., Lewis-Peacock, J., . . . Sulzer,

J. (2017). Closed-loop brain training: the science of neurofeedback. Nat Rev Neurosci,

18(2), 86-100. doi:10.1038/nrn.2016.164

Sorger, B., Reithler, J., Dahmen, B., & Goebel, R. (2012). A Real-Time fMRI-Based Spelling

Device Immediately Enabling Robust Motor-Independent Communication. Current

Biology, 22(14), 1333-1338. doi:10.1016/j.cub.2012.05.022

Spetter, M. S., Malekshahi, R., Birbaumer, N., Luhrs, M., van der Veer, A. H., Scheffler, K., .

. . Hallschmid, M. (2017). Volitional regulation of brain responses to food stimuli in

overweight and obese subjects: A real-time fMRI feedback study. Appetite, 112, 188-

195.

Spielberger, C.D., (2010). State-Trait Anxiety Inventory. In: The Corsini Encyclopedia of

Psychology. John Wiley & Sons, Inc..

Subramanian, L., Hindle, J., Johnston, S., Roberts, M., Husain, M., Goebel, R., & Linden, D.

E. J. (2011). Real-time fMRI Neurofeedback for Treatment of Parkinson's Disease.

Journal of Neuroscience, 31(45), 16309-16317.

Sulzer, J., Haller, S., Scharnowski, F., Weiskopf, N., Birbaumer, N., Blefari, M. L., . . .

Sitaram, R. (2013). Real-time fMRI neurofeedback: progress and challenges.

Neuroimage, 76, 386-399.

Thibault, R. T., MacPherson, A., Lifshitz, M., Roth, R. R., & Raz, A. (2018). Neurofeedback

with fMRI: A critical systematic review. Neuroimage, 172, 786-807.

Vidaurre, C., & Blankertz, B. (2010). Towards a Cure for BCI Illiteracy. Brain Topography,

23(2), 194-198.

123

Wechsler, D. (1981). The psychometric tradition: Developing the Wechsler Adult

Intelligence Scale. Contemporary Educational Psychology.

Weiskopf, N. (2012). Real-time fMRI and its application to neurofeedback. NeuroImage,

62(2), 682-692.

Witte, M., Kober, S. E., Ninaus, M., Neuper, C., & Wood, G. (2013). Control beliefs can

predict the ability to up-regulate sensorimotor rhythm during neurofeedback

training. Frontiers in human neuroscience, 7, 478.

Zotev, V., Krueger, F., Phillips, R., Alvarez, R. P., Simmons, W. K., Bellgowan, P., ... &

Bodurka, J. (2011). Self-regulation of amygdala activation using real-time fMRI

neurofeedback. PloS one, 6(9), e24522.

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(11), e79184.

Zotev, V., Phillips, R., Yuan, H., Misaki, M., & Bodurka, J. (2014). Self-regulation of human

brain activity using simultaneous real-time fMRI and EEG neurofeedback.

NeuroImage, 85, 985-995.

124

Neuroimage: Guidelines for Authors

125

126

127

Part Three: Clinical Experience

Adult Placement: Community Mental Health Recovery Service:

In this placement, I gained experience in carrying out interventions with clients

of a wide range of ages, backgrounds and with varying intellectual abilities. I

learned to apply core therapeutic models. Predominantly this involved CBT and

I was also supervised on a case applying Psychodynamic theory. I also worked

as a reinforcer for a STEPPS group. I gave a talk on trauma and dissociation at

an Assistant Psychologist meeting. I completed my service-related project on

the topic of managing rates of non-attendance at the team base.

Older Adult Placement: In-patient Wards:

Here I worked on both organic and functional wards with a greater focus

towards Systemic and consultancy-based work. I applied models such as

Positive Behaviour Support and Bronfenbrenner Ecological Model. Focusing on

particular complex cases, I learned skills in influencing staff teams to improve

care and carry out interventions. I also completed some short-term one-to-one

interventions involving coping skills and community-based assessments related

to memory loss.

128

Learning Disability Placement: Community Team for People with Learning

Disabilities:

Supervision at this placement had a strong focus on Systemic thinking. I

therefore applied a range of Systemic models to interventions and in particular,

a Narrative approach. I worked both directly and indirectly with clients

themselves, with families and with care home staff. I learned to adapt

interventions to suit people of varying ability levels. Through carrying out a tree

of life group intervention, I learned the impact of community work to create a

sense of belonging in those with mental health difficulties. I gave teaching to

the community team on the theory of attachment.

Specialist Placement: Neuropsychology Outpatients Clinic:

This placement was particularly specialised in carrying out assessments related

to cognitive difficulties in late adulthood. I became very proficient in carrying

out psychometric assessments. I learned to adapt assessments to people who had

various difficulties with engaging; for example due to difficulties with

impulsivity. I completed two pieces of rehabilitative work with clients

struggling with adjustment to cognitive difficulties. I administered a group

intervention for the family of carers of those with dementia.129

Child and Adolescent Placement: Child and Adolescent Mental Health Service:

I was able to greatly refine my skills in one-to-one interventions on this

placement, paying close attention to the details of high-quality CBT. I also

learned to apply different models such as Systemic and Psychodynamic thinking

in an integrative way at multiple levels; such as one-to-one, with schools and

with families. I learned to use new psychometric assessments suited for

establishing educational needs and was involved in the diagnosis of ASD and

ADHD in the neurodevelopmental pathway at the placement. I supervised

assistant psychologists for cases involving psychometric assessment.

130

Part Four: Assessments

PSYCHD CLINICAL PROGAMMETABLE OF ASSESSMENTS COMPLETED DURING TRAINING

Year I AssessmentsASSESSMENT TITLE

WAIS WAIS Interpretation (online assessment)Practice Report of Clinical Activity

“Jane” – Staged Trauma work with a Female in her mid-30s presenting with difficulties in Managing Emotions

Audio Recording of Clinical Activity with Critical Appraisal

CBT-based Work with a Woman in her mid-30s Presenting with Difficulties in Managing Emotions

Report of Clinical Activity N=1

“Jenny” – CBT based Intervention with a Woman in her Late Teens presenting with Distressing thoughts about her Body Image

Major Research Project Literature Survey

Which Psychological Factors Influence Learning Outcomes From Neurofeedback?

Major Research Project Proposal

Psychological Factors predicting Learning Outcomes from Neurofeedback using Real-time Functional Magnetic Resonance Imaging

Service-Related Project Audit of Non-Attendance and Related Risk Factors in an Adult Community Mental Health Team (2016-2017)

Year II AssessmentsASSESSMENT TITLE

Report of Clinical Activity/Report of Clinical Activity – Formal Assessment

Neuropsychological Assessment with a Women in her early 70s presenting with Language and Memory Difficulties

PPLD Process Account A Case of Improved Cohesion within a Personal and Professional Development Group and its Relevance to Working in the National Health Service

Year III Assessments ASSESSMENT TITLE

Presentation of Clinical Activity

“Matt” – A Male With A Learning Disability In His Late Teens Presenting With Difficulties In Managing Anger And Anxiety

Major Research Project Literature Review

A Systematic Review Of The Psychological Factors That Influence Neurofeedback Learning Outcomes

Major Research Project Predictors Of Successful Modulation Of Emotion-Related 131

Empirical Paper Networks In A Study Of Real-Time F-MRI Neurofeedback With Adolescent Females

Report of Clinical Activity/Report of Clinical Activity – Formal Assessment

“Mr Ayad” Neurorehabilitation With A Man In His Mid-50’s Presenting With Difficulties In Attention, Memory And Processing Speed

132