-
BIDIRECTIONAL EEG
NEUROFEEDBACK TRAINING OF
THETA COHERENCE IMPROVES
VISUAL ATTENTION
Ksenia Folomeeva &
Ove Mathias Langerud Nesheim
Master of Philosophy in Psychology
Cognitive Neuroscience discipline at the Department of
Psychology
UNIVERSITY OF OSLO
May 2015
-
II
-
III
Bidirectional EEG neurofeedback training of theta coherence
improves
visual attention
By Ksenia Folomeeva & Ove Mathias Langerud Nesheim
Submitted as a master thesis in Cognitive Neuroscience
Department of Psychology
University of Oslo
May 2015
-
IV
Copyright Ksenia Folomeeva and Ove Mathias Langerud Nesheim
2015
Bidirectional EEG neurofeedback training of theta coherence
improves visual attention
Authors: Ksenia Folomeeva and Ove Mathias Langerud Nesheim
Supervisors: Bruno Laeng, Markus Handal Sneve, Svetla
Velikova
http://www.duo.uio.no
-
V
-
VI
Abstract
Authors: Ksenia Folomeeva and Ove Mathias Langerud Nesheim
Title: Bidirectional EEG Neurofeedback Training of Theta
Coherence Improves Visual
Attention
Supervisors: Bruno Laeng, Markus Handal Sneve (co-supervisor)
and Svetla Velikova
(external supervisor)
Neurofeedback (NF) has the potential to enhance cognitive
functioning through
learned regulation of brainwave activity. However, NF for
optimizing performance in healthy
people is still in its infancy and currently not fully explored.
Here, we present an experiment
where 12 subjects undergo 10 sessions of a novel NF protocol
with eyes-closed bidirectional
theta coherence training. This protocol was selected based on
several ideas: contemporary
neuroscience suggests that neural coherence support neuronal
communication, and high task-
related coherence is often observed with higher performance. At
the same time, brain’s theta
waves have been shown to be particularly involved in attentional
processes. In addition, it
can be argued that neural flexibility should encompass the
ability to regulate up and down in
accordance with the cognitive demands of the environment. In
order to evaluate the success of
the NF training in the experimental group, a multiple object
tracking (MOT) task was
administered both pre- and post-training while both
electroencephalogram (EEG) and
pupillometry were recorded simultaneously. A passive control
group performed the test twice
for comparisons, with the same time lag. The results indicate
that NF training was successful
in enhancing attentional processes, since behavioural
improvements were found in both
accuracy and response time (RT) during MOT, and only in the NF
group. In addition, lower
task-related pupil dilations suggested that less mental effort
was deployed during post-training
MOT by the experimental group compared to the control group. The
baselines of resting EEG
recorded before each NF session were compared to the initial
baseline and revealed
widespread increases in coherence in all frequency bands.
Analysis of task-related EEG
indicated higher levels of longitudinal coherence in the
experimental group during the post-
training MOT. However, we cannot exclude that confounding
variables related to changes in
motivational factors could make comparisons between the control
group and experimental
group problematic. We can only tentatively conclude that the
novel NF protocol employed in
the current experiment shows promising support for beneficial
effects of bidirectional theta
NF on cognition. The current experiment should be regarded as an
exploratory study. The NF
protocol was developed in collaboration with Smartbrain AS
(Oslo, Norway) and their
experts. All the collection and analysis of data was done by
Ksenia Folomeeva and Ove
Mathias Langerud Nesheim (authors of the thesis).
-
VII
-
VIII
Acknowledgements
We would like to thank Prof. Bruno Laeng (supervisor) for his
advice, feedback and
guidance on theoretical issues and pupillometry and, most of
all, for helping to organize the
collaboration which made this project possible.
We would like to thank Dr. Markus Handal Sneve (co-supervisor)
for guidance on the
design of the experiment, helping with the generation of MOT
videos as well as for valuable
comments during the writing of the thesis.
We would like to thank Svetla Velikova (MD, PhD) for her advice,
helping to develop
the Neurofeedback (NF) protocol and guiding NF sessions, EEG
analyses and interpretations.
Also, we are grateful for the hospitality at SmartBrain AS and
Haldor Sjåheim’s support
along the way.
A special thank goes to Jonas Meier Strømme for helping us with
writing the python
script during long winter nights. We also thank Fredrik Svartdal
Færevaag, Bendik Holm and
Pelle Bamle for participating in the pilot testing of the MOT
task, EEG and pupillometry
recordings.
-
Contents
Introduction
............................................................................................................................................
1
Attentional systems of the brain and multiple object tracking
(MOT) ............................................ 1
Pupillometry and attention
................................................................................................................
2
EEG and attention
...............................................................................................................................
4
Theta coherence
..............................................................................................................................
5
EEG Neurofeedback and attention
....................................................................................................
6
Hypothesis and predictions
................................................................................................................
8
Methods
................................................................................................................................................
10
Participants
.......................................................................................................................................
10
Procedure and design
.......................................................................................................................
10
Tasks and Equipment
........................................................................................................................
11
MOT task
.......................................................................................................................................
11
Pupillometry
..................................................................................................................................
12
EEG recordings
..............................................................................................................................
12
Neurofeedback protocol\training
.................................................................................................
13
Preprocessing and analysis of data
..................................................................................................
14
Pupillometry
..................................................................................................................................
14
Behavioral
data..............................................................................................................................
15
EEG analysis
...................................................................................................................................
15
Results
...................................................................................................................................................
18
MOT results
.......................................................................................................................................
18
Analysis of
accuracy.......................................................................................................................
18
Analysis of RT
.................................................................................................................................
20
Pupillometry results
.........................................................................................................................
22
EEG results
........................................................................................................................................
23
Regression analysis of resting baseline EEG.
.................................................................................
23
Full-spectrum analysis
...................................................................................................................
25
Coherence during MOT1 and MOT2.
............................................................................................
27
Discussion
.............................................................................................................................................
31
Limitations and future directions
.....................................................................................................
34
Conclusion
.............................................................................................................................................
35
References
........................................................................................................................................
36
Appendix
...............................................................................................................................................
44
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
1
Introduction
The main goal of cognitive neuroscience is to understand how the
mind/brain works
but an important target is also to find practical applications
of this knowledge. Perhaps
reflecting the challenges of the Information Age, there has
recently been an increasing interest
in techniques of cognitive enhancement. Attention, the ability
to focus on some information
while ignoring the rest seems fundamental to cognitive processes
like memory and learning,
as well as our interaction with other human beings. People
practice meditation or use brain-
boosting pills and even play brain-training games as attempts to
train attention. One approach
endorsed by neuroscientists is based on the “operant
conditioning” of brainwaves, generally
known as neurofeedback (NF). Recent advances in technology have
made NF more accessible
to researchers and practitioners seeking to improve attention.
Typically, investigators have
focused on up-regulating EEG power values, like the sensorimotor
rhythm (Egner &
Gruzelier, 2001), beta (Egner & Gruzelier, 2004) and frontal
midline theta (Fm-theta)
(Enriquez-Geppert, Huster, Figge, & Herrmann, 2014).
Alternative NF protocols involve
training of EEG coherence, but these have been less explored.
Coherence can be interpreted
as a measure of functional connectivity of distant brain regions
(Fries, 2005; Mitchell,
McNaughton, Flanagan, & Kirk, 2008), which makes it a
relevant target for NF. In the current
experiment, we set out to test the efficacy of a novel NF
protocol, involving both up- and
down-regulation of theta coherence, in order to enhance
sustained visual attention in healthy
participants. The outcome measure was a behavioral task for
divided visual attention while
simultaneously monitoring usage of cognitive load or mental
effort by recording pupil
dilations (Kahneman, 1973).
Attentional systems of the brain and multiple object tracking
(MOT)
Visual attention selects information relevant to our internal
and external goals, while
ignoring distractions. When playing a game of football, humans
rely on visual attention to
attend to the ball, team mates and opponents while ignoring the
crowd, the referee and other
distracting aspects. This process is likely to be ensured by
top-down control, attributed to a
dorsal frontoparietal attention network (Corbetta, Patel, &
Shulman, 2008). If the football
field suddenly gets invaded by hooligans, bottom-up processes
kicks in to redirect behavior
from the game to the unexpected situation. This process is
supported by a ventral attentional
network, working as an alarm system. The ability to operate
among relevant and irrelevant
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
2
sensory stimuli is therefore achieved through interaction of the
top-down and bottom-up
networks (Corbetta, Kincade, & Shulman, 2002).
The Multiple Object Tracking (MOT) task was developed by
Pylyshyn and Storm
(1988) in order to study early visual processes of spatial
indexing. During MOT, the subject is
required to visually track several moving targets among
distractors while fixating on a central
cross on a computer screen. This task may pose a continual
demand on the visuo-attentional
system. The MOT-paradigm has been used to test different models
of how the attentional
system is capable of tracking several objects at once through
serial and/or parallel processes
(Howe, Cohen, Pinto, & Horowitz, 2010; Pylyshyn & Storm,
1988) . On average, the
participants are able to track 4-5 targets in a single trial
(Pylyshyn & Storm, 1988), however,
the performance depend on the targets\distractors ratio, speed
of the moving objects and
tracking time (Alvarez & Franconeri, 2007). By varying these
parameters, the cognitive load
can be operationalized and studied.
Moreover, as shown by neuroimaging studies, a tracking network
including regions in
the frontal, parietal and occipital cortices is engaged during
MOT (Alnæs et al., 2014; Culham
et al., 1998; Howe, Horowitz, Morocz, Wolfe, & Livingstone,
2009), covering areas of the
dorsal frontoparietal attention network (Corbetta et al., 2008).
Subcortical activations during
tracking (compared to passive viewing) has been found in the
thalamus with the pulvinar
nucleus, the basal ganglia and the locus coeruleus (LC) among
others (Alnæs et al., 2014).
Moreover, activity in the dorsal attention network and the LC
has been found to be closely
linked to task-related pupil dilations during MOT (Alnæs et al.,
2014; Murphy, O'Connell,
O'Sullivan, Robertson, & Balsters, 2014), paralleling the
finding that LC activity correspond
with the demands of attentional tasks (Raizada & Poldrack,
2007). Based on the stability of
the pupil dilation towards the MOT task, which was showed in 9
individuals in a follow-up
study after a few years (Alnæs et al., 2014), pupillometry can
be considered a reliable
estimate of attentional effort .
Pupillometry and attention
The allocation of limited attentional resources relates to the
psychological construct of
‘mental effort’, as a special kind of arousal according to
Kahneman (1973). Through a series
of experiments on different mental tasks, a subject’s pupil
dilation has proved to be a sensitive
measure of mental effort (Laeng, Sirois, & Gredebäck, 2012).
For example, Beatty (1982)
claimed that fluctuations of the mental activity could be
detected through changes in pupil
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
3
size recorded simultaneously with the task performance. A
task-related pupillary response can
be compared to an event-related brain potential recorded by EEG:
task-related pupil-size
changes appear within a short time gap (100 or 200 ms) following
task onset (Beatty, 1982b).
Classical experiments have shown that the dilation of the pupil
follows second by second
alterations in short-term memory load (Kahneman & Beatty,
1966), is sensitive to the level of
abstraction in a language processing task (Wright &
Kahneman, 1971), is sensitive to the
difficulty of mental arithmetic problems (Hess & Polt,
1964), can be used to signal perceptual
thresholds for visual detection (Hakerem & Sutton, 1966;
Kahneman, Beatty, & Pollack,
1967) and indicates the level of performance during tasks
requiring sustained attention
(Beatty, 1982a).
Ahern and Beatty (1979) have investigated the association
between a subject’s
pupillary response to arithmetic problems and his or her
Scholastic Aptitude Test (SAT)
score. The participants who had higher SAT scores showed less
pupil dilation (suggesting use
of less mental effort in order to complete the task) compared to
the participants with lower
scores. More recent studies have confirmed that pupillometry can
be a reliable measurement
of attentional effort during the performance of a task
(Gilzenrat, Nieuwenhuis, Jepma, &
Cohen, 2010; Laeng, Ørbo, Holmlund, & Miozzo, 2011; Wierda,
van Rijn, Taatgen, &
Martens, 2012).
The pupil dilation response from cognitive processing is thought
to stem from the
release of norepinephrine (NE) in the LC through inhibitory
connections to the Edinger-
Westphal nucleus (EWN; (Wilhelm, Ludtke, & Wilhelm, 1999).
The EWN in turn innervates
ciliary ganglion supporting the sphincter pupillae muscle,
controlling the constriction of the
pupil. While the pupil size can vary considerably (2 mm - 8 mm)
with the amount of light that
impinges on the retina, the diameter variations stemming from
mental effort are much
smaller. Cognitively evoked pupil dilations are rarely larger
than 0.5 mm (Beatty & Lucero-
Wagoner, 2000). In a model of LC activity (Aston-Jones &
Cohen, 2005), two modes are
described. The tonic mode is an exploration mode where behavior
is adaptively adjusted to
the environmental changes. In the phasic mode, attention is
filtered to optimize performance
of task-specific behavior. LC phasic activity therefore signals
task-related activity. In order to
capture the cognitively evoked pupil dilations, investigators
usually subtract the tonic pupil
dilation (baseline) from the phasic response (task-related pupil
dilation).
One current model suggests that the LC-NE system is also partly
responsible for the
deactivation of the ventral attention network during focused
attention (Corbetta et al., 2008;
Thatcher, 1992, 1998). In the study described above, the pupil
dilation response was shown to
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
4
predict activity in the dorsal attention network and LC better
than a simple “load” variable,
operationalized as the number of targets to-be-tracked in the
MOT task (Alnæs et al., 2014).
Thus, for the present experiment, pupil dilations were recorded
to give an index of mental
effort during MOT, in turn reflecting subcortical activations
related to mental effort and
sustained attention.
EEG and attention
Electroencephalography (EEG) enables the user to study
electrical activity stemming
from neural circuits in the brain. The application of
quantitative EEG (qEEG) allows
transformation of the EEG signal from the time domain to the
frequency domain by the
application of Fourier analysis (Cooley & Tukey, 1965). The
transformed EEG signal is
characterized by amplitude (measured in μV), power (representing
the squared amplitude) and
frequency (measured in Hz). On the basis of their frequencies,
brain rhythms are subdivided
into the following main bands: delta (1-3Hz), theta (4-7Hz),
alpha (8-12Hz), beta (13-30 Hz)
and gamma (30-50Hz). These bands may be functionally distinct,
and can reveal oscillatory
brain activity related to cognitive processes.
Regarding attention, theta has received particular interest from
investigators (Ishii et
al., 1999; S. Makeig et al., 2004; Missonnier et al., 2006). It
covers the frequencies in the
range 4-7 Hz and has been named after the thalamus to which the
origin of cortical theta has
been attributed (Walter & Dovey, 1944). The thalamus sends
rhythmic activity to the cortex
by means of pacemaker cells, which participate in producing
rhythmic EEG activity (Steriade,
2005). Another source of EEG recorded theta is the anterior
cingulate cortex (ACC) (Asada,
Fukuda, Tsunoda, Yamaguchi, & Tonoike, 1999), producing the
frontal midline theta (Fm-
theta) activity widely implicated in attentional processes and
cognitive demand (Mitchell et
al., 2008). Increased Fm-theta is observed during mental
calculation (Harmony et al., 1999),
visuo-spatial N-back tasks (Smith, McEvoy, & Gevins, 1999),
the Sternberg memory task
(Fernandez et al., 2000) episodic memory tasks (Klimesch,
Schimke, & Schwaiger, 1994) and
video game playing (Pellouchoud, Smith, McEvoy, & Gevins,
1999). Several researchers
have tried to determine whether Fm-theta reflects attentional
processes or working memory
(WM) processes (Gomarus, Althaus, Wijers, & Minderaa, 2006;
Sauseng, Hoppe, Klimesch,
Gerloff, & Hummel, 2007), and suggest that Fm-theta reflect
attention. Furthermore,
frontoparietal theta coherence was found to indicate integration
of sensory information into
executive functions (Sauseng et al., 2007).
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
5
Theta coherence. In addition to the theta power, EEG theta
coherence has been
demonstrated to correlate with tasks of attention (Makeig et
al., 2002) and working memory
(Klimesch, 1999). Coherence reflects the synchronization of
activity between two EEG
electrodes and can have values between 0 (no coherence) and 1
(maximum coherence).
Coherence is calculated by correlating the spectral content of
the two electrodes over a certain
time window within distinct frequency bands, and provides a
measure of the signals’ linear
dependence. If spectral content contained within specific
frequency bands correlates
continuously over the time period, coherence is high
(Saltzbertg, Burton, Burch, Fletcher, &
Michaels, 1986), even in the presence of highly uncorrelated
activity in other frequencies. A
possible confounding variable when analyzing coherence is
increased power of a source
localized between the two synchronized electrodes, whose signal
reaches both electrodes
(Fein, Raz, Brown, & Merrin, 1988).
Thatcher and colleagues (Thatcher, Krause, & Hrybyk, 1986)
have developed a model
of EEG coherence showing that it depends on cortico-cortical
interactions and strength of
synaptic connections between the brain regions (Thatcher, 1992,
1998). Therefore, coherence
can be defined as “Coherence = (Nij*Sij)”, where N stands for
the number of cortico-cortical
connections, and S stands for the strength of those connections.
According to his model,
increased coherence could be attributed either to an increase in
numbers or strength of
synaptic connections between two areas in the cortex. Findings
from studies on patients with
neurogenic pain, however, suggest that EEG coherence might also
reflect an active output
pathway from thalamus to the cortex as the amount of
thalamocortical coherence was
comparable to the amount of cortical coherence in the theta
range (Sarnthein & Jeanmonod,
2008).
Processing of complex information is likely to require
functional integration across a
number of distant brain regions. Coherence analysis between EEG
electrodes during
performance on a specific task could be used to measure this
integration, however, relatively
few studies have pursued this possibility. The highly
influential communication-through-
coherence hypothesis claims that distant brain regions are only
able to communicate
efficiently when they oscillate coherently (Fries, 2005).
Coherent oscillation could allow the
excitability of a region in the network to be predictive,
creating “temporal windows” for
effective communications (Fries, 2005; Pajevic, Basser, &
Fields, 2014). Thalamic nuclei
with wide projections to the cortex have cells with intrinsic
oscillating properties that render
the nuclei ideal “broadcasting centers” of rhythmic activity
(Steriade, 2005). These cells’
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
6
influences on cortical regions could establish functional
selectivity through the specific
distributed rhythms, including theta rhythms.
In support of the above accounts, increased frontoparietal theta
coherence has been
observed during the retention period of a working memory task
(Von Stein & Sarnthein,
2000). Increased theta coherence across the scalp has been
reported during encoding of
correctly recalled nouns (Weiss, Müller, & Rappelsberger,
2000). During the performance of
both verbal and spatial intelligence tests, people with higher
IQ’s have been shown to display
higher long-distance coherence in the theta band (Anokhin,
Lutzenberger, & Birbaumer,
1999), which is thought to reflect their brain’s ability to
establish integration of the involved
cortical regions. Regarding EEG during resting state, coherence
values seem to be less
predictive of task performance than task-related EEG (Anokhin et
al., 1999). Some have
found negative correlation between resting EEG coherence and
intelligence (Thatcher, North,
& Biver, 2005), while others reported a positive
relationship in alpha coherence (Marosi et al.,
1995).
As theta power and coherence have been proved essential for
performance in cognitive
tasks, we should note that EEG neurofeedback training often
focuses on this frequency in
order to improve the attentional abilities.
EEG Neurofeedback and attention
During EEG neurofeedback, the EEG signal is analyzed real-time,
and when the
subjects manage to regulate their brain activity above a certain
threshold for a fixed period of
time, a type of visual or auditory reward is fed back. Over
time, the subject learns to produce
more of the desired brain activity. The exact strategy used by
the trainee in order to learn to
regulate the brain may vary considerably among trainees, ranging
from positive thinking,
relaxing, and visual imagery and so on. In fact, conscious
awareness of how one learns to
regulate the brain activity in accordance with the NF training
may not be a prerequisite for
successful learning (Gruzelier, 2014b).
Today, there is no established standard for how to analyze the
learned control of brain
wave activity caused by NF during training sessions. Gruzelier
(2014b) describes three main
types of analysis present in the literature. Across-session
learning involves analyses of
changes from session to session in the ability to regulate brain
activity during the actual NF
training. Within-session learning involves analysis of NF
training during certain periods
within one NF session. Baseline increments analysis is used to
investigate changes in pre-
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
7
training baseline EEG recordings from session to session.
Looking at baseline increments
might be the most direct way of analyzing lasting changes from
NF training (Gruzelier,
2014b; Ros et al., 2013). Even though changes are hypothesized
to occur in the trained
frequency bands and electrodes, NF training might also induce
changes in electrodes and
frequency bands outside the trained ones. Gruzelier (2014b)
points out that analysis of the full
frequency spectrum is rare, yet should be an essential
requirement for NF studies. This
requirement is followed in the analysis of the present
experiment.
NF in a clinical setting has been applied for years in order to
improve dysfunctions in
brain activity related to different disorders, including ADHD
(Lofthouse, Arnold, & Hurt,
2012), autism spectrum disorders (Coben, Linden, & Myers,
2010), cerebral stroke (Bearden,
Cassisi, & Pineda, 2003), and consequences of brain and
spinal cord damages (Cavinato et al.,
2011) among other disorders. More recently, investigators have
also turned their attention to
cognitive enhancement of healthy subjects and have applied
neurofeedback for improvement
of sustained attention (Egner & Gruzelier, 2004; Egner &
Gruzelier, 2001), musical
performance (Egner & Gruzelier, 2003), working memory
(Vernon et al., 2003) or visuo-
motor skills (Ros et al., 2009) etc. With the possibility to
modulate neural oscillations,
neurofeedback has the potential to inform cognitive neuroscience
of more than just
correlations between cognitive tasks and brain oscillations.
In seeking to enhance a cognitive function, NF-studies aims to
modulate the EEG-
waves activity related to that function, to analyze NF induced
changes in tonic or phasic EEG,
and to investigate cognitive improvement by some cognitive test.
For example, training up
the amplitude of SMR (sensorimotor rhythm) and beta1 (12.5–16
Hz) have shown an effect
on sustained attention in healthy participants (Egner &
Gruzelier, 2004; Egner & Gruzelier,
2001). Enriquez-Geppert and colleagues (Enriquez-Geppert et al.,
2014) investigated nineteen
participants undergoing eight sessions of NF on Fm-theta to
improve executive functioning
(EF). Importantly, outcome measures of NF success and EFs were
compared to twenty-one
participants who had undergone pseudo-neurofeedback. During
pseudo-neurofeedback,
participants typically receive random feedback or feedback from
someone else’s brain. The
experimental group was able to up-regulate Fm-theta amplitude
better and showed improved
EFs in two out of four tests, compared to the
pseudo-neurofeedback group. Wang and Hsieh
(2013) reported similar findings of Fm-theta amplitude training
(including pseudo-
neurofeedback) where the NF-group improved working memory and
attention. A recent
review suggest that NF seems to have great potential as a method
of improving cognitive
functioning (Gruzelier, 2014a).
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
8
While a number of studies have investigated neurofeedback of
Fm-theta amplitude in
relation to cognitive performance, fewer studies have
investigated regulation of theta
coherence. In fact, we are not aware of any study optimizing
performance employing NF
training of coherence. In this article, we present an experiment
where twelve human subjects
undergo ten sessions of theta coherence neurofeedback. The
training protocol tested here
includes increasing of theta power and cyclic increase and
decrease of theta coherence on
interhemispheric electrodes. A test of MOT is given pre- and
post-training with simultaneous
task-related EEG and pupillometry recordings. Pre-training EEG
baselines are also analyzed
to evaluate the effect of NF training on the tonic EEG.
The NF protocol applied here was developed in collaboration with
Smartbrain AS
(Oslo, Norway) in order to explore a novel NF approach, and
therefore there are not previous
published data on it. The training was done with eyes-closed, as
theta rhythm were shown to
be more profound on the EEG recordings with eyes-closed (Barry,
Clarke, Johnstone, Magee,
& Rushby, 2007). Therefore, an auditory reward was used as a
feedback signal. The protocol
starts with an increase of theta power, since the enhancement of
power facilitates the
enhancement of the coherence, which is trained in the next step
of the protocol. The rationale
for training theta coherence is that this band is related to
attentional processing (Makeig et al.,
2002; Mitchell et al., 2008; Sauseng et al., 2007). Also, theta
coherence might reflect the
integration of information in task-relevant regions through a
temporal window ensuring
coherent activity (Fries, 2005), possibly supported by the
recruitment of rhythmic thalamic
activity (Steriade, 2005). Both up- and down-regulation is
trained due to the finding that high
coherence correlates with high cognitive performance (Anokhin et
al., 1999; Weiss et al.,
2000), but is not required during rest. This way, theta
coherence might become more
“adaptive” to task demands by training the ability to turn on
and off coherence. As pointed
out by Gruzelier (2014b), most NF studies have chosen
unidirectional NF training due to
often reported correlations between cognitive performances and
either heightened or lowered
EEG power/coherence. However, it can be argued that learned
control should include the
ability to regulate activity in both up and down directions,
according to task demands.
Hypothesis and predictions
Our main hypothesis is that training of neuronal flexibility, in
the sense of repeated
up- and down-regulation of theta coherence, will facilitate
cognitive performance.
If the theta coherence training is successful, we expect trained
individuals to show
improved accuracy and response time on MOT when comparing
performance before training
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
9
to that after training, while also deploying less mental effort
(as indexed by task-related pupil
dilations) after the training. Furthermore, during task-related
EEG, we expect higher theta
coherence to be associated with higher cognitive performance,
similar to what has been
observed in several studies (Anokhin et al., 1999; Weiss et al.,
2000). For both resting and
task-related EEG, changes in other frequency bands can be
expected due to the frequently
reported non-specific effects of NF (Gruzelier, 2014b), and
these bands will therefore also be
analyzed. However, as we are not aware of any bidirectional NF
protocols reminiscent of the
protocol deployed in the present experiment, we are not
specifically predicting the direction
of the effect of NF on the resting EEG. Since the present NF
training does not specifically
target the “tracking network” (Howe et al., 2009), large effect
sizes should not be expected.
For both the behavioral measures and pupillometry, the
experimental group is predicted to
change significantly more than the control group. In general,
for the control group, a stable
pattern of performance and neurophysiological measures is
expected.
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
10
Methods
Participants
Twenty-nine volunteer participants were recruited by asking
students at campus and
by announcing the project on Facebook. However, given the
demanding schedule to be met
for this experiment, 7 participants terminated the experiment
due to their busy work schedules
before completion. A total of 8 females and 15 males were able
to participate in all training
sessions and tests (mean age: 26.09, range 19-36, SD=4.68).
Every participant read a
document with inclusion criteria for the study, making sure that
none of the participants had a
mental disorder or history of head trauma, or was currently on
medications that could affect
cognition. One participant was shifted from the experimental
group to the control group after
having finished the first day of MOT testing because of a
hearing impairment, making him
unsuitable for the auditory neurofeedback sessions.
Procedure and design
The experiment included an experimental group (N=12) and a
control group (N=11).
The experimental group included 4 females (mean age: 25.25;
range 21-29) and 8 males
(mean age: 25.13; range 20-31), whereas the control group
included 4 females (mean age:
23.5; range 19-31) and 7 males (mean age: 28.5; range 22-36).
Both groups performed pre-
training MOT (MOT1) and post-training MOT (MOT2) tasks during
which EEG and
pupillometry were recorded simultaneously. The experimental
group underwent 10 NF
sessions over 5 weeks, twice per week, in between MOT1 and MOT2.
Before each NF
session, resting baseline EEG was recorded for later analysis. A
mixed repeated measures
design including two within-subject factors (load and session)
and one between-subject factor
(group) was employed and the dependent variables were MOT
accuracy, response time (RT)
and task-related pupil dilations. A pre-test post-test design
was employed to compare task-
related EEG.
The MOT sessions, during which task-related EEG and pupillometry
were recorded,
were done in the Cognitive Laboratory of the University of Oslo
(Oslo, Norway). NF sessions
were conducted in SmartBrain’s clinic (Oslo, Norway). On the
first day of the experiment, all
participants signed a consent form which described the process
of the experiment, main
benefits and risks of the study. Prior to data collection, the
project was consulted with the
local ethical committee.
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
11
Figure 1. The procedure of one trial in the MOT task
Tasks and Equipment
MOT task. Videos for MOT task were generated using MATLAB
(MathWorks,
Natick, MA) and the psychophysics Toolbox extension (Brainard,
1997) prior to
programming the experiment, and saved as video clips in ‘.wmv’
format. The experiment was
programmed and run in E-prime 2.0 (Psychology Software Tools,
Inc) using the MOT videos
to build the MOT trials. Participants were seated approximately
60 cm from a 22 inches Dell
(Dell Inc, TX, USA) monitor with 1600*1024 resolution and asked
to fixate on a central
fixation point during the task.
In the MOT task, participants were requested to track the
several targets among the
distractors (Figure 1). Following the presentation of the
fixation cross, 12 blue objects
appeared on the computer screen. A short target-assignment phase
followed, where 2-5
objects were marked as red. After that, all objects were shown
in blue and started to move for
a total duration of 8 secs. At the end of a trial, after all
objects stopped moving and only one
of them (either a target or a distractor; 50% probability) was
highlighted in red (probe), the
participant’s task was to judge whether the selected object was
among the targets or not.
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
12
They indicated their choice by key presses, which also yielded a
response time (RT) for the
decision. Key ‘n’ was used for a ‘no’ response, and ‘b’ was used
for a ‘yes’ response. Half of
the trials included valid probes and half of them were invalid.
The valid and invalid probes
were presented in a randomized order.
The task had 4 load conditions, presented in random order within
each part of MOT
session (for definition of ‘part’ see below):
load2 (2 targets, 10 distractors);
load3 (3 targets, 9 distractors);
load4 (4 targets, 8 distractors);
load5 (5 targets; 7 distractors).
Thus the amount of objects on the screen was kept the same
during every load
condition, making visual crowding constant. The objects were 0.3
degrees in diameter,
moving with a speed of 6 degrees/per second; the minimum
distance allowed between the
objects was 1.6 degrees from edge to edge. The objects always
moved straight and when they
reached the edge of the display or bumped into another object,
their trajectories changed to a
random angle (full range allowed). All the MATLAB generated
videos for MOT were
visually inspected for “bad videos” including crowding of
objects or others flaws. Those
videos were excluded and replaced.
Each MOT session was divided into 4 parts separated by 5 min
breaks in order to
avoid excessive tiredness of the participants. Each part
consisted of 48 MOT-trials (12 MOT-
movies per load). The first part also included 8 practice trials
so that the participants and the
experimenters could make sure the instructions were understood.
Therefore, one complete
MOT-session included 200 movies (different for each MOT
session), 50 movies per load of
the task (including practice).
Pupillometry. The pupillometry recordings were conducted using
the iView X R.E.D.
eye-tracking system (Sensio-Motoric Instruments, Germany). Data
was recorded with the
iView X 2.7 software at a sampling rate of 60 Hz. Before every
MOT part, a personal 9 point
calibration procedure was performed on a 22 inches Dell (Dell
Inc, TX, USA) monitor with
1600x1024 resolution. The illumination of the room was kept
constant during both MOT
sessions.
EEG recordings. EEG was recorded during the first MOT session
and the last and
also before each session of the neurofeedback training (resting
baseline) for a total of 12
measurements (10 EEG recordings of a baseline and 2 EEG
recordings during MOT1 and
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
13
MOT2). The latter recording allowed us to assess transfer
effects of NF training on successive
pre-training baselines for the experimental group. The baselines
were recorded with eyes-
closed since the NF training was also done with eyes-closed. For
both MOT sessions and NF
sessions, the EEG preparation procedures were the same. The
participants were asked to
minimize body movements, control gaze and tongue movements in
order to avoid artefacts
All EEG recordings for each participant were done approximately
at the same time of the day
in order to avoid differences caused by the normal circadian
changes in EEG activity (Frank
et al., 1966). The distance between the nasion and inion was
measured in order to determine
the suitable size of the cup for each participant and to fit the
cup properly on the head. In
order to clean the ears, the NuPrep, mild abrasive gel was used.
After putting on the cup, the
ECI electrogel was applied to each electrode in order to provide
appropriate signal detection.
Different EEG systems were used during MOT and NF due to the
availability of the
equipment for the current project.
EEG recordings during the MOT task were done with the
Brainmaster Discovery 24E
acquisition system (BrainMaster, OH, USA), using 19-electrodes
caps (FP1, FP2, F3, F4, Fz,
F7, F8, C3, C4, Cz, T3, T4, T5, T6, P3, P4, Pz, O1 and O2) in
accordance with the 10-20
system (Jasper, 1958). BrainAvatar software was used for data
storage. Impedance for each
electrode and for each ear was adjusted to < 10 kΩ, as
measured by a 1089NP Checktrode
EEG Impedance meter.
For neurofeedback sessions Deymed TruScan EEG acquisition system
(32 channels;
Deymed, Czech Republic) was used together with 19-electrodes
caps (FP1, FP2, F3, F4, Fz,
F7, F8, C3, C4, Cz, T3, T4, T5, T6, P3, P4, Pz, O1 and O2).
Impedance for each electrode
and for each ear was adjusted to < 10 kΩ as measured by the
Truscan Acquisition software.
The participants sat in a comfortable armchair with eyes closed,
in a quiet room, with constant
temperature and light conditions. Each session required
approximately 40 minutes to
complete.
Neurofeedback protocol\training. The NF protocol was set up on
the commercially
available software Deymed TruScan (Deymed, Czech Republic). The
training lasted 30 min
and included 10 rounds:
a) 3 min increasing theta power in Cz;
b) 9 min training of theta coherence on F3-F4: 3min up, 3 min
down, 3min up;
c) 9 min training of theta coherence on C3-C4: 3min up, 3 min
down, 3min up;
d) 9 min training of theta coherence on P3-P4: 3min up, 3 min
down, 3min up.
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
14
In the TruScan software, the coherence values are calculated 16
times per second.
These values are averaged by the software because of the highly
erratic nature of EEG.
Participants were rewarded with a short beep sound whenever
their coherence values between
the respective electrodes were maintained above 60% for at least
half a second. This threshold
was kept constant throughout all NF training sessions. The
loudness of the reward signal was
adjusted by asking the subject to select a comfortable
level.
An auditory reward was selected as it allowed participants to
keep their eyes closed as
the eyes-closed condition has been shown to be characterized by
more profound theta rhythm
on EEG recordings (Barry et al., 2007).
Positive relationships were maintained between the experimenters
and the participants,
as this was thought to be important for the success of NF
training (Gruzelier, 2014b). The
experimenters showed interest in the condition of the
participants and their feelings regarding
the experiment and tried to be flexible regarding time-slots for
the training to make the
process of participating more pleasant.
Preprocessing and analysis of data
Pupillometry. The SMI R.E.D. I-View system uses a patented
algorithm to calculate
pupil diameter and adjusting for head movements, while a form of
linear interpolating is used
to replace eye-blinks and other outliers in the raw data stream.
To further preprocess the
pupillometry data, a custom made script was written in Python.
Pupillometry baselines were
collected between 300 ms and 0 ms before tracking start, when
all objects were present in
blue color. This interval was chosen as a baseline as no changes
in color or movement
occurred, and were therefore thought to exclude task-related
cognitive processing.
Furthermore, the baselines were subtracted from the average
pupil size between 2.5 seconds
and 7 seconds after tracking start. This sampling interval was
chosen because cognitively-
evoked pupil dilations arise slowly at the beginning of each
tracking period and reach an
asymptote around 2.5 secs (see Alnæs et al., 2014). In addition,
one can expect pupil dilations
related to preparatory processes for responses towards the end
of the tracking (Richer &
Beatty, 1985). Baselines were subtracted from the average
task-related pupil size in all trials
in order to obtain a measure of average task-related pupil
dilation. The data were further
separated by the amount of targets to-be-tracked in each trial,
and only trials in which a
correct answer was given were included for further analysis (in
total, approximately 7750
correct trials, with around 340 trials per participant, 84 per
load).
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
15
The pupillometry data were also checked for the presence of
outliers using the outlier
detection rule described by Hoaglin and Iglewicz (1987), where
the upper and lower boundary
is defined as:
Upper = Q3 + (2.2*(Q3-Q1))
Lower = Q1 – (2.2*(Q3-Q1)),
where Q3 is the 75th
percentile, Q1 is the 25th
percentile and 2,2 a constant multiplier.
Shapiro-Wilk’s test was applied in order to control data for
normality. A mixed effects
analysis of variances (ANOVA) with MOT session (2 levels: MOT1
and MOT2) and load
(load2; load3; load4; load5) as within-subject factors and Group
(NF and control) as between-
subjects factor was used for the task-related pupil dilation as
the dependent variable for the
experimental and control group. After that, repeated measures
ANOVAs were performed
separately for the control and experimental groups with load and
session as within-subject
factors. A planned comparison with paired t-tests comparing the
MOT1 and MOT2 for each
load was applied in order to compare the task-related pupil
dilation for the experimental group
and control group, separately.
Behavioral data. MOT results were averaged across each load in
MOT1 and MOT2.
Shapiro-Wilk’s test was applied in order to control data for
normality and the outlier detection
rule was applied in order to exclude outliers (Hoaglin &
Iglewicz, 1987).
Two separate mixed effects analysis of variances (ANOVA) with
MOT session (2
levels: MOT1 and MOT2) and load (load2; load3; load4; load5) as
within-subject factors and
Group (NF and control) as between-subjects factor were used for
accuracy and reaction time
as the dependent variables. After that, four repeated measures
ANOVAs with load and session
as within-subject factors were done separately for the
experimental and control groups, for
accuracy and RT. Paired t-tests comparing the MOT1 and MOT2 for
each load were applied
in order to test differences in accuracy and RT for the
experimental and control group,
separately, and according to the predictions made before the
experiment.
EEG analysis. The obtained EEG data were visually inspected and
artifacts were
removed using NeuroGuide Deluxe (Applied Neuroscience Inc.,
Florida, USA) software
version 2.8.3. Both computerized and manual artifact rejection
were applied. In order to
assure the quality of the selected EEG data, test-retest
reliability function of the NeuroGuide
was kept higher than 0.90 for each EEG record, according to the
recommendation of the
NeuroGuide (NeuroGuide Help Manual, 2002-2014). Further
statistical analyses of the data
were performed using Neurostat option of the NeuroGuide
software.
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
16
The statistical analysis in Neuroguide involves Fast Fourier
transform (FFT), a
technique used to identify the frequency components of the EEG
signal. For the analysis EEG
recording is divided into 2 sec. segments or epochs, which are
submitted to frequency analysis
(Kaiser & Sterman, 2000). In NeuroGuide, these epochs are
sampled at a 128 samples\sec.
sampling rate which results in 256 digital time points with a
frequency range from 5 to 40 Hz
with resolution of 0.5 Hz. As segmentation of the epochs in FFT
is likely to produce ‘sharp
edges’ (non-zero values at the beginning and at the end of each
epoch) and differences in
amplitude could result in errors of spectral information, known
as ‘leakage’; mathematical
functions called ‘windows’ are applied for each of the epochs
(Kaiser & Sterman, 2000). In
NeuroGuide, cosine taper windows are used for this purpose. Each
2 second FFT includes 81
rows (0 to 40 Hz frequencies) by 19 columns (electrode
locations), resulting in 1539 elements
cross-spectral matrix for each individual.
Although, the mathematical windows are useful for avoiding the
leakage problem,
they could smooth the frequency peaks at both edges of the
epoch, ending up in analyzing
only the central frequencies of the epoch and reducing the
signal power. However, the
multiple overlapping windows could be a solution (Kaiser &
Sterman, 2000) and the best
quality of data was shown to be achieved by 4 windows per epoch
or 75% overlapping. In
NeuroGuide, an EEG sliding average of 256 FFT cross-spectral
matrixes are computed for
each individual, editing EEG by advancing in 64-point steps.
The FFT is recombined with the 64-point sliding window for 256
FFT cross-spectrum
for EEG record. All the 81 frequencies for each 19 electrode
locations are log10 transformed in
order to correct the data for the normal distribution. The total
amount of 2 second windows is
entered into paired t-tests and is used to compute the degrees
of freedom for the statistical
analysis.
In order to evaluate the effect of NF on successive resting EEG
baselines, the mean
theta coherence for each of the trained electrode pairs was
subjected to linear regression
analysis with number of NF sessions as the predictor. To follow
the recommendation from
Gruzelier (2014b), a full-spectrum analysis followed and
included two comparisons on a
group level. Average EEG baseline recordings from sessions 1-3
were compared to average
baselines from sessions 4-6 and 8-10 using uncorrected paired
t-test analysis in NeuroGuide.
Full p-value tables are attached in the Appendix.
EEG recording during MOT1 and MOT2 were preprocessed in EEGLAB
by
MATLAB (Math-Works, Natick, MA) in order to select the EEG
epochs corresponding with
the specific load condition (2 to 5). Only trials with the
correct response given were included
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
17
into the analysis to maximize the likelihood that the same
cognitive process is involved in
each comparison. Thereafter, EEG epochs of each load were merged
together. Each EEG-
recording was manually cleaned from eye-movements, blinks, jaw
tension, or other body
movement artifacts, using NeuroGuide Deluxe (Applied
Neuroscience Inc., Florida, USA)
version 2.8.3.
The comparison between EEG recordings during MOT1 and MOT2 was
done for the
experimental and control groups separately with paired t-tests.
The EEG coherence was
compared for the recordings, averaged across all loads. By means
of this analysis the overall
trend in coherence changes could be revealed for the
experimental and control groups.
Additionally, paired t-tests were done separately for each load
in order to explore the
differences in coherence levels from the easiest to the most
difficult load of the task.
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
18
Results
MOT results
The outlier detection rule was applied in order to control for
possible outliers (Hoaglin
& Iglewicz, 1987) and none were detected. The Shapiro-Wilk
did not reach significance for
accuracy (experimentalMOT1 p=.186; experimentalMOT2 p=.09;
controlMOT1 p=.415;
controlMOT2=.687) or RT (experimentalMOT1 p=.526;
experimentalMOT2 p=.273;
controlMOT1 p=.921; controlMOT2 p=.359), meaning that the data
did not differ
significantly from the normal distribution.
Analysis of accuracy
An independent t-test was conducted in order to compare the
accuracy during MOT1
in the experimental and control groups to see if the groups
differed before the experimental
group commenced training. The accuracy of each load condition
was averaged for each
participant, giving each participant a total MOT accuracy score,
used for the independent t-
test. The analysis indicated that the groups did not differ by
accuracy level at the initial stage
of the experiment, (t(21)=-.246; p=.808; d=.10; see Figure
2).
Mixed repeated measures ANOVA with two within-subject factors (2
sessions and 4
loads) and one between-subject factor (group) revealed a
significant main effect of load
(F=46.754; p=.000; ŋ2=.69), but no significant effect of session
(F=.031; p=.862; ŋ
2=.001). A
significant interaction effect between Group and MOT session
(F=6.622; p
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
19
However, no interactions between session and load (F=1.976;
p=.127); load and group
(F=.633; p=.597); or session, load and group (F=1.575; p=.204)
were found.
Repeated measures ANOVA for the experimental group revealed a
significant main
effect of the load (F=27.070; p=.000; ŋ2=.711), but no
significant main effect of session
(F=2.791; p=.123; ŋ2=.202). A significant interaction effect
between the load and session was
shown (F=3.053; p=.042; ŋ2=.217). A planned comparison with 4
paired t-tests was
conducted comparing MOT1 and MOT2 for each load condition
(Figure3A) and Bonferroni-
corrected to a significance level of p
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
20
Analysis of RT
Independent t-test was conducted in order to compare the
reaction time during MOT1
in the experimental and control groups to see if the groups
differed before the experimental
group commenced training. The RT of each load condition were
averaged for each
participant, giving each participant a total MOT RT score, used
for the independent t-test. The
analysis revealed that the groups did not differ by RT at the
onset, t(21)=.706; p=.488; d=.29;
see Figure 4.
Mixed effects ANOVA for RT revealed a significant main effect of
the load
(F=42.453, p=.000, ŋ2=.87), but no main effect of session
(F=2.673, p=.117, ŋ
2=.113). A
significant interaction effect of session and group (F=5.665;
p
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
21
Figure 5. The diagram presents the difference in response time
during MOT tasks in the experimental (A) and control (B) group for
each load condition. The bars represent the between-subjects
SEM.
analysis for the experimental group was conducted by means of 4
paired t-tests and was
Bonferroni-corrected to a significance level of p
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
22
Pupillometry results
The data were tested for normality using the Shapiro-Wilk test
and for the outliers
using the outlier detection rule. The Shapiro-Wilk test did not
reach significance for any of
the groups, meaning that the data were normally distributed, and
no outliers were detected.
An independent t-test was conducted in order to compare the
pupil dilation during
MOT1 in the experimental and control groups to establish whether
the groups were
comparable from the start. The analysis indicated that the
groups did not differ by pupil
dilation during MOT1 (t(21)=.706; p=.196; d=.55; Figure 6).
The mixed repeated measures ANOVA indicated a significant main
effect of load
(F=30.714; p=.0001; ŋ2=.594) and a significant main effect of
session (F=6.815; p=.016;
ŋ2=.245). However, no significant interaction between load and
group (F=.256; p=.857;
ŋ2=.012); session and group (F=1.040; p=.320; ŋ
2=.047) and load, session and group (F=.767;
p=.597; ŋ2=.035) were found.
A repeated measures ANOVA for the experimental group revealed a
significant main
effect of load (F=21.860; p=.000; ŋ2=.665) and session
(F=16.474; p=.002; ŋ
2=.600).
However, no significant interaction between load and session was
shown (F=.330; p=.804;
ŋ2=.029). A planned comparison with paired t-tests was conducted
in order to compare the
pupil dilation during MOT1 and MOT2 for the experimental group
for each load and
Bonferroni-corrected to a significance level of p
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
23
significance for load3 in the MOT1 (M=.06; SD=.101) and MOT2
(M=-.021; SD=.107)
conditions; t(11)=2.347; p = .039, d=.768. However, no
significant difference for load4
(p=.076) and load5 (p=.139) was detected.
A repeated measures ANOVA for the control group showed a
significant main effect
of load (F=10.735; p=.000; ŋ2=.518), but no significant main
effect of session (F=.739;
p=.410; ŋ2=.069) or interaction between session and load
(F=.523; p=.670; ŋ
2=.05). Paired t-
tests in the control group (Figure 7B) were Bonferroni-corrected
to a significance level of
p
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
24
Figure 8. The graphs present mean theta coherence across all
resting baselines for the trained electrode pairs: F3-F4 (A), C3-C4
(B), P3-P4 (C). The regression lines are drawn with dotted lines.
Bars represent SEM.
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
25
Full-spectrum analysis
In NeuroGuide the coherence between inter-hemispheric and
intra-hemispheric
electrodes are calculated. The analysis output includes
topographical maps with t-values and
the corresponding significance level for comparison. The
interhemispheric and
intrahemispheric coherence in different frequency bands (delta,
theta, alpha and beta
correspondingly) are presented with figures.
For a type of full spectrum analysis reported here, it is
recommended to use a
Figure 9. Figure 9A. The absolute power comparisons during
resting baseline EEG between sessions 1-3 and 8-10. Figure 9B and
9C. FFT coherence comparisons during resting baseline EEG for
session 1-3 versus session 4-6 (9B) and session and session 1-3
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
26
statistical correction for the multiple comparisons made.
However, there is a trade-off
between increasing the likelihood of finding a significant
effect in topographically and
frequency specific regions and getting the full overview, even
though the chance of a type 1
error increases. To follow Gruzelier’s (2014b) recommendation to
do full spectrum analysis,
uncorrected data is reported in Figure 9, 10 and 11, while full
p-value tables can be found in
the appendix.
Figure 9A displays absolute power comparisons during eyes-closed
baseline
recordings for session 1-3 versus session 8-10. The Figure
presents the likelihood of obtaining
the t-value for each of the comparisons. White color indicates
no significant change. Except
for a change in Fz delta power, the neurofeedback protocol did
not alter participant’s baseline
absolute power values. This is an important finding as it
strengthens the validity of the
analysis of coherence, since power changes could confound the
analysis of coherence (Fein et
al., 1988).
Furthermore, the neurofeedback protocol led to statistically
significant changes in
participant’s baseline coherence values in all frequency bands,
especially in the lower
frequencies including the trained theta band. Looking at the
development from Figure 9B to
Figure 9C, it is evident that more significant changes in
coherence took place with more
sessions of neurofeedback training. With the exception of a
decrease in theta Fp2-F3
coherence between session 1-3 and 4-6, all significant changes
were due to increased
coherence. More longitudinal coherence changes occurred only in
the session 1-3 versus 8-10
comparison, with differences evident in occipital and frontal
areas (O1-F7) theta, occipital
and central areas (O1-C3) beta in the left hemisphere and
between frontal electrodes (F3-F4)
theta as examples. Coherence changes were most notably
intrahemispheric, and most so in the
left hemisphere. Of the trained electrode pairs F3-F4, P3-P4 and
C3-C4, only P3-P4 showed
significant changes in the trained theta band, occurring in the
session 1-3 versus 8-10
comparison.
In summary, the neurofeedback training protocol led to the
increases in coherence and
this could not be explained by absolute power changes. Changes
were observed in the both
intra- and inter-hemispheric coherence. Increased central
inter-hemispheric coherence was
observed in delta and beta frequencies, parietal increased
coherence was found in delta, theta
and beta bands and there was increased frontal theta coherence.
Regarding the intra-
hemispheric coherence an increase of occipito-central,
occipito-parietal and occipito-frontal
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
27
Figure 10. Comparison of the coherence between MOT1 and MOT2 in
MOT-task averaged across all loads for the experimental and control
groups.
coherence in the left hemisphere was observed in all bands.
Clearly, the NF protocol induced
changes in the resting EEG specifically related to coherence and
not power.
Coherence during MOT1 and MOT2. Figure 10 shows the coherence
changes for
MOT2 compared to MOT1, averaged across all loads.
Experimental group. For the experimental group, interhemispheric
changes involved
decreasing coherence between parietal electrodes in delta
frequency; occipital in theta;
temporal for beta and frontal for alpha and beta. The delta band
showed reduction of the
coherence in frontal, central and parietal midline areas of the
left hemisphere and across
frontal, central, parietal and occipital electrode in the right
hemisphere. The intrahemispheric
changes for the theta frequency reveal increased longitudinal
coherence across frontal, central,
parietal and occipital areas of both hemispheres. Similar
patterns were detected for alpha and
beta bands, where increased coherence was shown between
occipital, parieto-temporal,
central and frontal regions of the left hemisphere; and in
frontal, central and parietal areas of
the right. In the beta frequency interhemispheric changes also
involved increased coherence
between parieto-temporal electrodes.
Control group. For the control group, the interhemispheric
results in delta band
revealed decreasing coherence between frontal, parieto-temporal
and occipital regions, in
addition, coherence between frontal, central, parietal, temporal
and occipital electrodes
decreased in both hemispheres. The theta and beta bands revealed
decreased coherence
between temporal and parietal electrodes of the both
hemispheres. Decreased coherence in
theta frequency was detected between the occipital and frontal
areas of the both hemispheres.
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
28
The intrahemispheric coherence decreased between parietal,
temporal and central electrodes
in the right and left hemispheres. At the same time, coherence
in the theta band increased, as
seen in the frontal and central areas of the both hemispheres.
Interhemispheric coherence
increased between the parietal electrodes in theta, alpha and
beta frequency. The alpha band
showed increased coherence in the frontal regions of both
hemispheres; decreased coherence
in the parietal, central and frontal areas of the left
hemisphere and in occipital, parieto-
temporal and central areas in the right. The beta frequency
decreased in occipital, parieto-
temporal, temporal regions in the left hemisphere and
occipito-temporal areas in the right.
The intrahemispheric increased coherence was shown in the beta
band between frontal,
central and parietal areas.
In Figure 11, the changes in task-related EEG coherence are
presented for the
experimental and control group separately for different loads of
the task. Only the EEG
recordings during correct responses were included.
Experimental group. For the load2 in the experimental group
there was increased
coherence in alpha and beta frequency between occipital and
central areas of the left
hemisphere. For load3 changes involved increased coherence
between occipital and frontal
areas (O1 and F7) in theta band, occipital and central areas (O1
and C3) in alpha and
occipital, parietal and frontal regions (O1-P3, O1-C3, O1-F3,
O1-F7) of the left hemisphere in
beta frequency. For load4 coherence increased in frontal and
central regions of the right
hemisphere in theta range, occipital and central regions in the
left hemisphere and frontal
areas of the right in alpha frequency; and through the left
frontal, central and occipital regions
in beta. The highest load involved increased coherence in
occipital-frontal areas of the both
hemispheres and right frontal and central regions of the left
hemisphere in theta frequency;
frontally (Fp2-F4) in alpha. The beta frequency showed a
significant increase between frontal,
central and occipital regions of the left hemisphere and
frontal-parietal in the right
hemispheres, while decreased coherence is disclosed in frontal
areas (Fp1-F7) in theta, alpha
and beta frequency bands. Interestingly, with increasing of the
task difficulty, there was a
parallel increase of the coherence between MOT1 and MOT2. For
the easier load-levels of the
task, the enhancement of the coherence involved the left
parieto-occipital regions and
included changes in alpha and beta bands. For the most difficult
load level, increased
coherence was observed in theta and beta bands and consisted of
increased fronto-occipital
coherence bilaterally (with the left hemisphere prevalence in
beta band).
Control group. For the control group, the intrahemispheric
changes involved decreases
coherence in the parietal areas in delta frequency and in
occipital areas for theta. The delta
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
29
frequency coherence decreased also in temporal and parietal
areas in the left hemisphere and
in the parietal and frontal areas in the right. There was an
increase frontally in theta and in
central and parietal regions in beta. For load3 decreased
coherence in delta frequency was
more expressed than in the lower load: it involved the decrease
in intrahemispheric coherence
in the parietal areas and in interhemispheric coherence through
the temporal, central, parietal
and frontal areas in both hemispheres. The theta, beta and alpha
bands revealed decreased
coherence between frontal, temporal and parietal regions; at the
same time, the alpha
frequency increased frontally and centrally and beta frequency
showed increased coherence
between the parietal electrodes. The intrahemispheric results
showed coherence decreases
between the parieto-temporal and occipital regions in delta,
theta and beta. There was also
decrease in coherence in central and parietal areas in alpha
frequency. Increased coherence
revealed in the frontal areas for theta, alpha and beta
frequencies. Finally, for the load5, delta,
theta and beta bands decreased in coherence between hemispheres
in the posterior regions;
there was a significant decrease in delta coherence between T6
and F4, T5 and C3, and other
electrodes in occipital, parietal and temporal regions of the
left hemisphere. The theta
frequency showed decreases in parieto-temporal and frontal areas
of the right hemisphere, as
well as occipital electrode between hemispheres. At the same
time, significant increases in
theta coherence were detected within frontal regions in both
hemispheres. The alpha
frequency coherence decreased in occipital and parieto-temporal
areas of the right
hemisphere. There was an increase in coherence for the beta
frequency in frontal and central
regions of both hemispheres.
As the sample size in the current experiment was small, it was
not possible to match
the participants by EEG parameters and, therefore, reduce the
impact of individual EEG
differences across groups. The comparison between task-related
EEG recordings in the
experimental and control groups were made for MOT1 and revealed
differences between the
groups. Therefore, similar comparisons for MOT2 were not
conducted as the emphasis was
placed on within-group comparisons.
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
30
Fig
ure
11.
Com
pariso
n o
f th
e c
oh
ere
nce b
etw
een M
OT
1 a
nd
MO
T2 f
or
each
lo
ad
of th
e M
OT
-task for
the e
xperim
enta
l and
co
ntr
ol gro
ups. T
he
red lin
es r
epre
sent
incre
ased c
oh
ere
nce w
hile
blu
e lin
es r
epre
sent
decre
ase
d c
ohere
nce. T
hin
ner
lines indic
ate
s α
-valu
es less than o
r e
qua
l to
0.5
.
The m
ediu
m s
ized lin
e ind
icate
s α
-valu
es less th
an o
r equa
l to
.0
25. T
he t
hic
kest lin
es ind
icate
α-v
alu
es less t
han o
r eq
ual to
.01
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
31
Discussion
The current study sought to investigate a novel neurofeedback
protocol involving
bidirectional, eyes-closed, theta coherence training in order to
improve visual attention. At the
behavioral level, the experimental group showed improved
accuracy scores and decreased RT
during a subset of load conditions of the MOT task (Figure 3A
and Figure 5A). The control
group did not show such improvements in any of the measures
(Figure 3B and Figure 5B).
Pupil diameters were recorded during MOT to provide an estimate
of mental effort, and a
significant main effect of session was found, revealing that
processes related to learning
effects or familiarity with the task may have caused
task-related pupil dilations to decrease
from before training to after the training (i.e., from MOT1 to
MOT2). Planned t-tests showed
that the experimental group’s decrease in pupil dilations
reached significance for a subset of
load conditions (i.e., the easiest, from MOT1 to MOT2; see
Figure 6A), implying that the
experimental group spent less mental effort during MOT after NF.
Furthermore, the theta
coherence training led to changes in the resting baseline EEG
recorded before every training
session (Figure 8, Figure 9B and Figure 9C). Regression analysis
revealed a significant linear
increase in theta coherence with more NF training over the
trained electrode pairs. The full
spectrum analysis further revealed changes in all recorded
frequency bands and outside the
trained electrode pairs. The analysis of task-related EEG in the
experimental group showed
increased fronto-occipital, fronto-parietal, and fronto-central
intrahemispheric coherence
during MOT2 (Figure 10 and Figure 11). The control group showed
widespread decreased
coherence during MOT2 compared to MOT1 (Figure 10 and Figure
11). Despite the
limitations regarding the use of a passive (rather than active)
control group and a small
sample size, which both warrant a cautious interpretation of the
results, we discuss below
some possible mechanisms that could mediate the effect of NF on
MOT performance.
We have trained 12 individuals with EEG based bidirectional
theta coherence NF and
comparisons of successive pre-training baselines indicated a
pattern of increased coherence in
all frequency bands recorded and, in fact, also outside the
trained electrode pairs. To our
knowledge, bidirectional NF protocols have not been investigated
in relation to improving
cognition in healthy participants and, therefore, we did not
predict the direction of effects of
NF on the resting EEG. Although coherence changes outside the
target of training could seem
surprising, they are often observed (Gruzelier, 2014b), and can
be explained by reference to
Thatcher’s two-compartment model of coherence (Thatcher et al.,
1986). That is, Thatcher
collected resting EEG data from 189 individuals and showed that
interhemispheric coherence
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
32
in the frontal areas (F3-F4) was closely linked to the
intrahemispheric coherence in the
posterior areas (for example, between O1 and P3). At the same
time, interhemispheric
interaction in central electrodes (C3-C4) was associated with
the central and frontal
intrahemispheric coherence. Similar pattern have been shown in
the coherence between
parietal electrodes. Thus, the present results might be
interpreted in line with Thatcher’s
model, as training of the interhemispheric electrodes (F3-F4,
C3-C4 and P3-P4) influenced
anterior-posterior coherence within each of the hemispheres.
Furthermore, as discussed in the introduction, some cortical
theta rhythms have been
shown to originate from the thalamus (Steriade, 2005), and this
subcortical structure has been
proposed to play an important role in EEG changes following NF
training. Lubar (1997)
suggested that, in general, NF works through enhancing of the
cortical rhythms generated by
several thalamic pace-makers (Steriade, 2005) which have their
diverse connections with the
cortex. In this way, NF may alter widespread regions of the
cortex by targeting a small
number of electrodes for training. Moreover, there are currently
few NF studies reporting the
full spectrum of EEG changes following training, but the
prevailing outcome is in line with
non-specificity (Gruzelier, 2014b), meaning that changes can
also occur outside the trained
band. This is in line with the present findings, which supports
the view that full spectrum
EEG analysis should be included in NF studies. However, in order
to make conclusions about
the specific mechanism mediating NF training source-localization
more data are required. For
example, an alternative source of Fm-theta may be in the ACC
(Sauseng et al., 2007).
Given the large variety of different NF protocols that have been
reported to have
beneficial effects on cognition (Gruzelier, 2014a), there might
be a common mechanism, at
least across some protocols, by which NF leads to improved
cognitive performance. Ghaziri
and colleagues (2013) investigated structural changes with
diffusion tensor imaging after 3
months of NF training of beta amplitude, leading to improved
performance on a sustained
attention test. Increased fractional anisotropy (FA) values were
found in pathways related to
WM and sustained attention, perhaps indicating microstructural
changes related to
myelination, axon caliber and fiber density. Correspondingly,
myelination remains sensitive
to experience throughout adulthood (Young et al., 2013). In a
recently proposed model
(Pajevic et al., 2014), conduction velocity variation supported
by myelination works as a
major mechanism for neural plasticity. Activity-dependent
modulation of myelin could
functionally influence the oscillatory coupling of distant brain
regions (Pajevic et al., 2014).
Taken together, myelin plasticity might be a strong candidate
common mechanism by which
some NF-induced cognitive improvement could be explained. All
though speculative,
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
33
mechanisms of myelin plasticity might have mediated the effects
of NF on MOT in the
present experiment.
In the present experiment, the experimental group showed higher
levels of longitudinal
coherence during MOT2 compared to MOT1, as well as higher levels
of accuracy and faster
RT. This finding is in line with other studies reporting
correlations between high behavioral
performance and increased task-related coherence (Anokhin et
al., 1999; Von Stein &
Sarnthein, 2000; Weiss et al., 2000). The
communication-through-coherence (CTC)
hypothesis suggests that flexible coherence between oscillating
areas is essential for cognitive
performance (Fries, 2005). It is argued that coherence provides
a temporal window for
effective communication to interacting neuronal assemblies.
Perhaps the increase in task-
related coherence reflects more effective communication between
areas of the tracking
network, where information from occipital, parietal and frontal
areas can be integrated (Howe
et al., 2009). Looking at Figure 11, it was evident that the
experimental group recruited more
coherence during MOT2 between frontal, central, parietal and
occipital cortices. Interestingly,
relatively more significant changes related to longitudinal
coherence were shown comparing
the higher loads against lower loads. During the task-related
EEG, the experimental group
showed a higher number of significant changes in the
fronto-parietal, fronto-central and
fronto-occipital coherence in the beta band, compared to the
theta band (Figure 10). To some
degree, there are overlaps between patterns of changes in the
resting EEG and the task-related
EEG which raises the question of whether the task-related EEG
was really “task-related”.
Task-related pupil dilations decreased from MOT1 to MOT2 as
reflected by a main
effect of session in the mixed ANOVA. Paired t-tests indicated
that only the experimental
group’s decrease in pupil dilations reached significance and for
a subset of load conditions
from MOT1 to MOT2 (Figure 7A). A decrease in mental effort as a
result of learning effects
(Sibley, Coyne, & Baldwin, 2011) could be expected in both
groups, but as the experimental
group was the only group showing significant changes from MOT1
to MOT2, the findings
suggest that NF led task-related pupil dilations in the
experimental group to decrease further.
However, as the interaction between group and session was not
found in the mixed ANOVA
for pupillary responses, we cannot conclude that the
experimental group showed significantly
lower pupillary responses than the control group. If the
decrement observed in the
experimental group is caused by NF, our data suggest that NF
might lead to cortical and
subcortical changes related to higher cognitive performance.
Whether these changes are
actually caused by NF or not, it seems that both decreased
task-related pupil dilations and
increased task-related coherence during MOT is associated with
higher performance on a
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
34
group level. As the NF protocol applied here was not
specifically aimed at MOT related EEG
activity, our results indicate that the NF training might have
led to higher performance
through enhancement of some general attentional/cognitive
mechanism. If so, the medium
effect sizes (d=0.5 or higher; (Cohen, 1977) reported here are
perhaps surprising. However,
the possibly NF induced improvement of the experimental group is
likely mixed with learning
effects and motivational factors.
Alternatively, a slight non-significant decrease in MOT
performance that coincided
with a significant decrease in coherence in all frequency bands
for the control group, could
suggest that the control group had lost some motivation by MOT2.
Typically, lower task-
related coherence is reported during lower levels of performance
(Anokhin et al., 1999; Weiss
et al., 2000). In fact, participants in the control group
returned to the re-test session only in
order to take part in the experiment and were generally
interested in experiencing the EEG
recording and pupillometry. By the second session the novelty of
the approach would be thus
lost on them. If motivation could cause a change in performance
and coherence, the same
argument should be applied to increased performance and
coherence for the experimental
group. In this case, the participants in the experimental group
may have built up more
motivation by participating in the cost-free neurofeedback
training sessions, which may have
given a particular meaning to the final MOT2 session that
concluded the training.
Limitations and future directions
The present experiment has several limitations that should be
addressed in further
research. First of all, there was only a passive control group
and the inclusion of an active
control group would further strengthen the current results. It
is plausible that the mere
participation in a brain training study significantly alters a
person’s motivation and
performance level. For example, a control group could have
received pseudo-neurofeedback,
as was done in Enriquez-Geppert and colleagues (2014) study.
Typically, pseudo-
neurofeedback involves random feedback that should not alter the
functioning of the
receiver’s brain. Including such an active control group, one
could more confidently conclude
that changes in the experimental group were due to the specific
NF training protocol. Given
the time limit and demand on data sampling, it was not feasible
to include such an additional
group in the Master’s project. A possible follow up-study should
address this problem.
However, based on findings from other NF-studies that have
included pseudo-neurofeedback
-
Bidirectional EEG Neurofeedback Training of Theta Coherence
Improves Visual Attention
35
groups (