STATISTICAL ANALYSIS OF FUNCTIONAL MRI DATA USING
INDEPENDENT COMPONENT ANALYSIS
M. Bartés-Serrallonga
1
, J. Sole´-Casals
1
, A. Adan
2,3
, C. Falcón
4,5
, N. Bargalló
6
and J. M. Serra-Grabulosa
2,4
1
Digital Technologies Group, University of Vic, Vic, Spain
2
Departament de Psiquiatria i Psicobiologia Clínica, Universitat de Barcelona, Barcelona, Spain
3
Institute for Brain, Cognition and Behaviour (IR3C), Barcelona, Spain
4
Institut d’Investigacions Biomèdiques August Pi i Sunyer (IDIBAPS), Barcelona, Spain
5
CIBER-BBN, Barcelona, Spain
6
Secció de Neuroradiologia, Servei de Radiologia, Centre de Diagnòstic per la Imatge (CDI)
Hospital Clínic de Barcelona, Barcelona, Spain
Keywords: Functional magnetic resonance imaging, Independent component analysis, BOLD.
Abstract: Functional magnetic resonance imaging (fMRI) is a technique to map the brain, anatomically as well as
physiologically, which does not require any invasive analysis. In order to obtain brain activation maps, the
subject under study must perform a task or be exposed to an external stimulus. At the same time a large
amount of images are acquired using ultra-fast sequences through magnetic resonance. Afterwards, these
images are processed and analyzed with statistical algorithms. This study was made in collaboration with
the consolidated Neuropsychology Research Group of the University of Barcelona, focusing on applications
of fMRI for the study of brain function in images obtained with various subjects. This group performed a
study which analyzed fMRI data, acquired with various subjects, using the General Linear Model (GLM).
The aim of our work was to analyze the same fMRI data using Independent Component Analysis (ICA) and
compare the results with those obtained through GLM. Results showed that ICA was able to find more
active networks than GLM. The activations were found in frontal, parietal, occipital and temporal areas.
1 INTRODUCTION
Functional Magnetic Resonance Imaging (fMRI) is a
technique that provides the opportunity to study
noninvasively which parts of the brain are activated
by different types of stimulation or activity, such as
sight, sound or movement. This technique measures
the Blood Oxygenation Level Dependent (BOLD)
contrast, which is based on the differing magnetic
properties of oxygenated (diamagnetic) and
deoxygenated (paramagnetic) blood. When brain
neurons are activated, there is a change in blood
flow and oxygenation that causes a change in the
Magnetic Resonance (MR) signal which is received
by the receiver coils. A higher level of oxygenated
blood in a located area means that there is an
increase in neural activity in this area. On the other
hand, a lower level means the opposite (D’Esposito
et al., 1999).
In order to capture the effect of BOLD contrast,
the subject lies in the magnet under the influence of
a powerful magnetic field and a particular form of
stimulation is conducted (such as showing images
with a projector). Then, a series of low resolution
brain scans are taken over time. For some of these
scans the stimulus is present and for some others the
stimulus is absent. The low resolution brain images
of the two cases can then be compared in order to
see which parts of the brain were activated by the
stimulus. After the experiment has finished, the set
of images is pre-processed and analyzed.
One problem for fMRI data is that data includes
contributions from many other sources including the
heart beat, breathing and head motion artifacts,
which can cause wrong results (S.A Huettel. et al.,
2004). ICA-based methods have shown to be useful
for analyze data when this is noisy and when regions
involved in a particular task are unknown.
430
Bartés-Serrallonga M., Solé-Casals J., Adan A., Falcón C., Bargalló N. and Serra-Grabulosa J..
STATISTICAL ANALYSIS OF FUNCTIONAL MRI DATA USING INDEPENDENT COMPONENT ANALYSIS.
DOI: 10.5220/0003723504300436
In Proceedings of the International Conference on Neural Computation Theory and Applications (Special Session on Challenges in Neuroengineering-
2011), pages 430-436
ISBN: 978-989-8425-84-3
Copyright
c
2011 SCITEPRESS (Science and Technology Publications, Lda.)
In an attempt to find the components extracted
from data reporting on different subjects or
paradigms and discover which were task-related and
which were noise, we applied a method based on
ICA. In this paper, we present all the steps we did
for this work and we show results obtained from real
activation fMRI experiments conducted on a group
of forty subjects.
2 MATERIALS AND METHODS
The study was performed in a 3 T MRI scanner
(Magnetom Trio Tim, Siemens Medical Systems,
Germany) at the Diagnostic Imaging Centre at
Hospital Clínic (CDIC) using the blood-oxygen
level-dependent (BOLD) fMRI signal. Whereas the
pre-processing of MR images and the regression
model were performed using SPM8 software
(SPM8, Wellcome Department of Cognitive
Neurology, London), the data analysis was carried
out using Group ICA of fMRI Toolbox (Calhoun et
al., 2001). Both pre-processing and analysis software
were run on a Matlab platform (R2009b version).
2.1 Participants
Forty right-handed healthy undergraduate students
[50% women; age range 18–25, mean (+
S.D.) 19.6
(+1.7)] were recruited from the University of
Barcelona. Subjects with chronic disorders, nervous
system disorders or history of mental illness were
excluded, as well as regular drinkers and those on
medication. All participants were non smokers and
low caffeine consumers (< 100mg/day), had
intermediate circadian typology and reported an
undisturbed sleep period of at least 6 h during the
night prior to the fMRI scan sessions. Caffeine may
affect the performance of the task (Serra-Grabulosa
et al., 2010a); Adan and Serra-Grabulosa, 2010). For
this reason the participants abstained from caffeine
intake for a minimum of 12 h and fasted for at least
8 h prior to the first fMRI session. The study was
approved by the ethics committee of Hospital Clínic
de Barcelona. Written consent was obtained from all
participants, who were financially rewarded for
taking part.
2.2 Experimental Design
The functional magnetic resonance imaging was
obtained using gradient echo sequence single-shot
echo-planar imaging, with the following parameters:
TR (repetition time): 2000 ms, TE (echo time): 40
ms, FOV (field of view): 24 x 24 cm, matrix 128 x
128 pixels, flip angle 90, slice thickness: 2 mm, gap
between sections: 0.6 mm, 36 axial slices per scan.
A total of 243 volumes were purchased, with 46
slices each.
During the acquisition of fMRI, in order to
obtain the BOLD contrast, the subjects performed a
sustained attention task (CPT-IP, Continuous
Performance Test-Identical Pairs), which is a
modification of the Cornblatt task (Cornblatt et al.,
1989) and a control task. CPT-IP task was created
with the software Presentation (Neurobehavioral
System, USA). All stimuli were presented to the
subjects through glasses specially designed for use
in the scanner.
The CPT-IP task was performed using a block
design. It started with a block of 35 seconds of
accommodation to the scanner, which had a blank
screen that the subject had to stare at. After this first
block, 9 blocks of CPT were alternated with 9
blocks of control (Figure 1). Preceding each block,
subjects received instructions for what to do in the
next block for a duration time of 5 seconds.
Figure 1: Design of the sustained attention task with
alternation between blocks.
Each of the CPT blocks had a total of 27
numbers formed by 4 digits (1 to 9, without
repeating the same figure), so that 23 of the figures
were different and 4 were repeated. The presentation
time of each number was 450 ms and the interval
between the onsets of each of the 27 consecutive
digits was 750 ms. Subjects’ task was to detect the
repeated figures and respond by pressing a button as
quickly as possible (Figure 2A). The position of the
repeated figures was randomized over the blocks
CPT. Concerning the control block, it always had
the same 4 digits (1 2 3 4) and the task of the
subjects was only to stare at it throughout the
presentation (Figure 2B).
STATISTICAL ANALYSIS OF FUNCTIONAL MRI DATA USING INDEPENDENT COMPONENT ANALYSIS
431
Figure 2: The following figure illustrates the design of the
task blocks. The top (A) exemplifies the figures presented
in the CPT blocks. In this example, you should respond
to the stimulus e3. The bottom (B) exemplifies the figures
presented in the control blocks.
2.3 Data Pre-processing
The data that comes directly out of the scanner is
very noisy. The noise is defined as any variability in
the data that is not explained by our statistical model
(Ashby, 2011), for example when a subject moves
his or her head. The magnitude of this variance is
important because it can cause some errors in the
results of the statistical analysis. If the noise is low,
it will increase the probability of discover true brain
activations related with the task.
To reduce the error variance as much as possible,
functional and structural MRI data were pre-
processed using SPM8 software (http://www.fil.ion.
ucl.ac.uk/spm/software/spm8/) as described in
(http://www.fil.ion.ucl.ac.uk /spm/doc/spm8_manual
.pdf), which aims to improve the signal noise ratio.
This includes the following steps:
1. Converting all the images from DICOM (Digital
Imaging and Communication in Medicine) format
to NIfTI (Neuroimaging Informatics Technology
Initiative) format in order to treat them with SPM8
and Group ICA of fMRI Toolbox.
2. Realigning the images to the same position
according to the coordinates of the anterior and
posterior commissure.
3. Correcting the head movements which may have
occurred in the scanner. In this way, the head
movements can cause artefacts or abrupt changes
in the intensity of the signal which can badly
corrupt fMRI data and in consequence affect the
results of the statistical analysis. The calculations
for the correction are made through
interpolations, performing 3 corrections of
rotation and 3 corrections of translation.
4. Coregistering the functional and structural
images. In this way a correspondence is achieved
point to point between the structural and the
functional images and the activations can be
interpreted.
5. Normalizing the images to minimize the huge
individual differences in the sizes and shapes of
individual brains. All brains need to be of the
same size and orientation in order to be
compared. The aim is to normalize the data into
the standard Montreal Neurological Institute
(MNI) space. This space is used worldwide, so
results are comparable with those from all other
institutes.
6. Finally, apply Gaussian transformations in order
to minimize false positives.
2.4 Implementation of the Regression
Model
After pre-processing step, we proceeded to perform
the regression model to explain brain activations. To
do this, we created a regression line where signal
changes observed in each voxel could be explained
by changes in the proposed task minimizing the
residual error (Figure 3).
2.5 Independent Component Analysis
After pre-processing and regression model creation
steps, we applied ICA analysis to the images. In the
following lines, we will explain the principles of
ICA. Independent components analysis is a
multivariate technique which is very popular and
common in the analysis of fMRI data. A good way
to understanding the basic principles of ICA is
through the typical ICA problem namely cocktail
party (Hyvärinen et al., 2000).
In this situation, some people are attending a
cocktail party speaking all at once. Assume that their
voices are recorded from different microphones
placed around the room. The resultant recordings
will be unintelligible because each microphone will
pick up some mixture of two or more people
speaking simultaneously and some background
noise. As a result, it will be very difficult to
understand even a single speaker. ICA provides an
effective method which can usually solve the
cocktail party problem separating the conversation
of every speaker.
An equivalent to this problem in fMRI is to
assume that instead of speakers, there are functional
independent neural networks that are simultaneously
active during some fMRI experiment. The aim of
ICA is to separate these simultaneous neural
networks from the global mixture as independent
components. The problem that ICA tries to solve
NCTA 2011 - International Conference on Neural Computation Theory and Applications
432
Figure 3: Regression model proposed to explain, for each
voxel of the functional MRI images, the variability in the
signal along the recorded 243 volumes. Each one of the 10
columns corresponds to one of the input variables in the
regression. The first one corresponds to the attention task
in which the subject has to respond to repeated stimuli.
The second one corresponds to the task of looking at
numbers and the third one to the task of initial rest. The
next 6 columns are the values applied to correct the head
movements in the pre-processing step. The last one
represents the error. On the right side of the table the
registered volumes are listed from 1 to 243. For each
variable, white colour indicates that this helps to explain
the variability while black colour indicates the opposite.
can be expressed in matrix notation by the following
equation:
X
=
AS (1)
where
A
is the (unknown) mixing matrix and S is
the (unknown) source matrix. The procedure
consists on recovering S, using only the vector X
with N observations. For that, the aim is to estimate
a weight matrix W, which should be the inverse of
A, up to scale and permutation effects, so that the
original independent signals can be recovered as:
U
= W
X = WAS S
(2)
To estimate the ICA model it’s necessary to
make certain assumptions and restrictions
(Hyvärinen et al., 2001):
1. The components are assumed to be statistically
independent.
2. The components must have non-gaussian
distributions.
3. For sake of simplicity, we assume that the
unknown mixing matrix is square.
4. We cannot determine the variances (energies) of
the recovered independent components.
5. We cannot determine the order of the recovered
independent components.
2.6 ICA Algorithm used
To perform the ICA analysis, as we have mentioned
before, we used the Group ICA of fMRI Toolbox.
This program has the option to make the analysis
using different algorithms, as Jade, Erica, Infomax,
Simbec, Amuse and others.
The chosen algorithm to analyze fMRI data was
Infomax because has been one of the most
commonly used algorithms for fMRI data analysis
and has proven to be quite reliable (Calhoun et al.,
2004).
3 RESULTS
3.1 Selection of the Independent
Components
After ICA analysis we selected some of the
components in order to evaluate results. For that, we
did a multiple regression and a statistic correlation
with every paradigm. We excluded the components
that had a p-value greater than 0.01, and the ones
which were associated to noise. Therefore we
selected 3 components for the CPT task and 3
components for the control task.
3.2 Obtention of the Areas of Interest
After the selection of the independent components,
we performed a T – test with all the subjects and all
the components. We also performed a ‘multiple
regression’ SPM8 analysis to establish the
relationship between CPT-IP-related activations.
The fMRI results were interpreted only if they
attained both a voxelwise threshold p<0.05
(corrected) (cluster extent (k) = 10voxels). The
anatomical location of the activated brain areas was
determined by the Montreal Neurological Institute
(MNI) coordinates. Anatomical labels were given on
the basis of anatomical parcellation developed by
(Tzourio-Mazoyer et al., 2002).
STATISTICAL ANALYSIS OF FUNCTIONAL MRI DATA USING INDEPENDENT COMPONENT ANALYSIS
433
3.3 fMRI Results
Activations found in the CPT task were located (see
Table 1 and Figure 4) bilaterally in frontal lobe (BAs
6, 8, right 9, 10, 11, 24, 32, 44, 45, 46, 47), parietal
(BAs 7, 23, 31, 40), temporal (BAs 21, 22, 34, right
37) and occipital (BAs 18, 19).
The control task showed a pattern of bilateral
activation (see Table 2 and Figure 5) in frontal lobe
(BAs 4, left 6, 8, 9, 10, 11, 24, 32), parietal (BAs
right 2, left 3, right 5, 7, 23, 31, 39, 40), temporal
(BAs 20, 21, 28, 34, 35, 37, 38) and occipital (BAs
17, 18).
4 DISCUSSION
The aim of our study was to analyze fMRI data from
a stimulation paradigm using ICA, and compare the
obtained results with previous ones done in other
study (Serra-Grabulosa et al., 2010b) which
analyzed the same data using general linear
modelling (GLM).
In general terms, obtained results follow a
similar pattern as previous analysis reported in
(Serra-Grabulosa et al., 2010) but with more active
regions. In the following paragraph we will
comment these new activations.
As in the GLM case, ICA analysis of the CPT
task indicated that the used paradigm activates a
network in frontal, parietal and occipital areas. In
addition, the new results showed activations in the
temporal area. The frontal activation obtained was
bilateral and the new included areas were frontal eye
fields (BA 8), dorsolateral prefrontal cortex (right
BA 9), ventral anterior cingulate cortex (BA 24) and
inferior prefrontal gyrus (BA 47). Frontal eye fields
are believed to play an important role in the control
of eye movements and in the management of
uncertainty (Volz et al., 2005) which could be
present during the CPT task. BA 9 is part of
dorsolateral prefrontal cortex and it’s involved in
functions such as working memory, integration of
sensory mnemonic information and the regulation of
intellectual function and action. These functions
were necessary in the CPT task in order to remember
the numbers, to compare them and to decide the
correct answer. BA 24 is part of the anterior
cingulate cortex and many studies attribute functions
such as error detection, anticipation of tasks,
attention (Weissman et al., 2005), motivation, and
modulation of emotional responses to the ACC
(Bush et al., 2000; Posner et al., 1998; Nieuwenhuis
et al., 2001). Thus this area could contribute to
maintain the attention during the task and detecting
the equal numbers. BA 47 has been implicated in the
processing of syntax in spoken and signed
languages. Therefore, this zone could be related to
the processing of the numbers during the task.
Bilateral parietal activations were also found in
the CPT task. These are in the posterior cingulate
cortex, which is associated with Brodmann areas 23
and 31. Imaging studies indicate a prominent role for
the posterior cingulate cortex in pain and episodic
memory retrieval (Nielsen et al., 2005). Thus, this
part of the cortex could contribute to recover the
digits from memory during the task. BA 40 and
more exactly its supramarginal gyrus part, is
involved in reading, both regarding meaning and
phonology (Stoeckel C. et al., 2009). In our case it
may be related with the number recognition.
Another cluster of activation related to the CPT
task, and not found in the previous study, was found
in temporal areas. BA 21 has been connected with
processes as different as observation of motion,
recognition of known faces and accessing word
meaning while reading. BA 22 is an important
region for the processing of speech so that it can be
understood as language. BA 37 includes functions as
face and body recognition, number recognition and
processing of colour information. These regions
could be related to the recognition and the numbers
meaning when were shown. BA 34 is a part of the
entorhinal area which is the main interface between
the hippocampus and neocortex. The entorhinal
cortex (EC)-hippocampus system plays an important
role in autobiographical / declarative / episodic
memories and in particular in spatial memories
including memory formation, memory consolidation
and memory optimization in sleep. Therefore this
area could contribute to processing the numbers
during verbal working memory.
Comparing with GLM, ICA analysis of the control
task also indicated activity in angular gyrus,
posterior cingulate gyrus, frontal gyrus and inferior
and medial temporal gyrus. In addition, ICA results
showed activations in primary motor cortex,
premotor cortex, primary somatosensory cortex,
somatosensory association cortex, perirhinal cortex
and temporopolar area. As in the previous analysis,
the control task showed activations in different brain
areas which were not activated in the CPT task and
probably could reflect an inhibition of processes that
could interfere with the correct execution of the task,
as external and internal monitoring (Gusnard and
Raichle, 2001). This deactivation could optimize
performance in high attentional demanding tasks
(McKiernan et al., 2003).
NCTA 2011 - International Conference on Neural Computation Theory and Applications
434
Figure 4: This figure shows the activations found on
the CPT task. Each colour represents the active region of a
different component.
Table 1: Coordinates (x, y, and z) of the areas of
significance, level of significance (T-Score) and
localization of the voxel (BA) for CPT task.
Coordinates of voxels
T-Score BA
x y z
3 11 49 32.64 6
-51 20 40 11.10 8
42 53 4 22.94 10
48 14 28 23.09 9
0 29 22 29.08 24
6 23 31 31.43 32
-45 12 20 6.18 44
60 14 19 11.70 45
48 32 25 23.71 46
36 23 -5 10.88 47
24 -67 49 25.91 7
0 -25 31 8.24 23
-24 -76 28 20.94 31
-48 -61 43 14.00 40
63 -31 -5 16.72 21
-48 8 -2 16.68 22
9 5 -11 12.15 34
57 -43 -5 10.18 37
-30 -88 4 25.17 18
39 -82 -5 22.88 19
5 CONCLUSIONS
After the analysis ICA has demonstrated to be a
technique with a great potential. Comparing with
GLM-based approaches ICA is able to separate
statistical independent components and identify
Figure 5: This figure shows the activations found on the
control task. Each colour represents the active region of a
different component.
Table 2: Coordinates (x, y, and z) of the areas of
significance, level of significance (T-Score) and
localization of the voxel (BA) for control task.
Coordinates of voxels
T-Score BA
x y z
-33 -22 58 7.98 4
-24 -19 64 7.70 6
-21 38 46 10.07 8
9 53 37 8.64 9
3 53 -5 42.53 10
0 41 -14 9.83 11
0 26 19 10.92 24
-3 44 -2 43.77 32
33 -37 61 6.60 2
-12 -37 67 6.45 3
3 -40 64 7.50 5
15 -52 49 7.44 7
3 -58 16 28.66 23
-6 -64 22 37.85 31
48 -64 28 10.84 39
60 -25 31 10.74 40
-42 -25 13 6.67 41
-57 -7 -20 8.68 20
57 -13 -17 8.94 21
-21 -16 -17 7.81 28
more networks than GLM. The main inconvenience
we observe with ICA is that in some cases it might
identify a large number of components, while only a
few are related with the task. To find those related
components can be a challenge. Therefore it’s
important to estimate an appropriate number of
components in order to better separate the real
STATISTICAL ANALYSIS OF FUNCTIONAL MRI DATA USING INDEPENDENT COMPONENT ANALYSIS
435
activations from noise. Despite these difficulties,
ICA works well and separates noise from real
activations allowing extracting the desired signals.
ACKNOWLEDGEMENTS
This work has been partially supported by the
Secretaria d’Universitats i Recerca of the
Departament d’Economia i Coneixement of the
Generalitat de Catalunya under the grant 2010BE1-
00772 to Dr. Jordi Solé-Casals; by the University of
Vic under de grant R0904; and by grants of the
Ministerio de Educación y Ciencia of the Spanish
Government (SEJ2005-08704) and the Departament
d’Innovació, Universitats i Empresa of the
Generalitat de Catalunya /2009BE-2 00239) to Dr.
Josep M Serra-Grabulosa.
REFERENCES
Adan A, Serra-Grabulosa JM ., 2010. Effects of caffeine
and glucose, alone and combined, on cognitive
performance. Human Psychopharmacology clinical
and experimental, 25 (4), 310 – 317.
Ashby, F., 2011. Statistical analysis of fMRI Data.
Cambridge, MA: MIT Press
Bush G., Luu P., Posner M. I., 2000. Cognitive and
emotional influences in anterior cingulate cortex.
Trends Cogn Sci., 4 (6), 215 – 222.
Calhoun, V. D., Adali, T., Pearlson, G. D. and Pekar, J. J.,
2001,”A Method forMaking Group Inferences From
Functional MRI Data Using Independent Component
Analysis”, HBM, 14, 140 – 151.
Calhoun, V. D., Adali, T., Pearlson, G. D., 2004,
”Independent component analysis applied to fMRI
data: a generative model for validating results”, J.
VLSI Signal Process, 37, 281 – 291.
Cornblatt, B. A., Lezenweger, M. F., Erlenmeyer-Kimling,
L., 1989. The Continuous Performance Test, Identical
Pairs Version: II. Contrasting attentional profiles in
schizophrenic and depressed patients. Psychiatry
Research, 29, 65 – 85.
Gusnard D. A., Raichle M. E., 2001. Searching for a
baseline: Functional imaging and the resting human
brain. Nature Neuroscience Reviews, 2, 685 – 694.
D’Esposito, M., Zarahn, E., Aguirre, G. K., 1999. Event-
Related functional MRI: implications for cognitive
Psychology. Psychological Bulletin, 125, 155 – 64.
Hyvärinen, A., Oja E., 2000. Independent component
analysis: Algorithms and applications. Neural
Networks, 13, 411 – 430.
Hyvärinen A., Oja E., Karhunen J., 2001. Independent
component analysis, John Wiley & Sons.
McKiernan K. A., Kaufman J. N., Kucera-Thompson J.,
Binder J. R., 2003. A parametric manipulation of
factors affecting task-induced deactivation in
functional neuroimaging. Journal of Cognitive
Neuroscience, 15, 394 – 408.
Nielsen F. A., Balslev D., Hansen L. K., 2005. Mining the
posterior cingulate: segregation between memory and
pain components. Neuroimage, 27 (3), 520 – 532.
Nieuwenhuis S., Ridderinkhof K. R., Blom J., Band G.P.,
Kok A., 2001. Error-related brain potentials are
differentially related to awareness of response errors:
evidence from an antisaccade task. Psychophysiology,
38 (5), 752 – 60.
Posner M. I., DiGirolamo G. J., 1998. Executive attention:
Conflict, target detection, and cognitive control. In
Parasuraman R. The attentive brain. Cambridge, Mass:
MIT Press.
Scott A. Huettel, Allen W. Song, and Gregory McCarthy.,
2004. Functional magnetic resonance imaging.
Sunderland, MA: Sinauer Associates
Serra-Grabulosa J. M, Adan A, Falcon C , Bargallo N,
2010a Glucose and caffeine effects on sustained
attention: an exploratory fMRI study. Human
Psychopharmacology clinical and experimental 25 (7-
8), 543 – 552
Serra-Grabulosa J. M., Adan A., Falcón C., Bargalló N.,
Solé-Casals J., 2010b. Cerebral correlates of the
continous performance test-identical pairs version: An
fMRI study. In Proceedings of the Third Internationa
Conference on Bio-inspired Systems and Signal
Processing – BIOSIGNALS.
Stoeckel C., Gough .P. M., Watkins K. E., Devlin J. T.,
2009. Supramarginal gyrus involvement in visual
word recognition. Cortex, 45 (9),1091 – 1096.
Tzourio-Mazoyer N, Landeau B, Papathanassiou D,
Crivello F, Etard O, Delcroix N. et al. 2002.
Automated anatomical labeling of activations in SPM
using a macroscopic anatomical parcellation of the
MNI MRI single-subject brain. Neuroimage 15, 273 –
289.
Volz K. G., Schubotz R. I., von Cramon D. Y., 2005.
Variants of uncertainty in decision-making and their
neural correlates. Brain Res. Bull, 67 (5), 403 – 12.
Weissman D. H., Gopalakrishnan A., Hazlett C. J,
Woldorff M. G., 2005. Dorsal Anterior Cingulate
Cortex Resolves Conflict from Distracting Stimuli by
Boosting Attention toward Relevant Events. Cerebral
Cortex, 15, 229 – 237.
NCTA 2011 - International Conference on Neural Computation Theory and Applications
436