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
& 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
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
& 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