Decision Making

NeuroImage 60 (2012) 21912205
Contents lists available at SciVerse ScienceDirect
NeuroImage
journal homepage: www.elsevier.com/locate/ynimg
Full Length Article
Decision-making under risk: A graph-based network analysis using functional MRI

Ludovico Minati a, b,, Marina Grisoli c, Anil K. Seth d, e, Hugo D. Critchley b, e, f
a
Scientic Department Fondazione IRCCS Istituto Neurologico Carlo Besta, Milano, Italy
Psychiatry, Clinical Imaging Sciences Centre (CISC), Brighton & Sussex Medical School (BSMS), Falmer, UK
c
Neuroradiology, Fondazione IRCCS Istituto Neurologico Carlo Besta, Milano, Italy
d
Informatics, University of Sussex, Falmer, UK
e
Sackler Centre for Consciousness Science, University of Sussex, Falmer, UK
f
Sussex Partnership NHS Foundation Trust, Brighton, UK
b
a r t i c l e
i n f o
Article history:
Accepted 15 February 2012
Available online 24 February 2012
Keywords:
Decision making
Functional MRI (fMRI)
Psychophysiological interaction (PPI)
Dynamic causal modeling (DCM)
Networks
Graph theory
Prefrontal cortex
a b s t r a c t
Adaptive behavior requires choosing effectively between options involving risks and potential rewards. Existing studies implicate lateral and medial prefrontal areas, striatum, insula, amygdala and parietal regions in
specic aspects of decision-making. However, limited attention is given to how brain networks encode economic parameters in patterns of inter-regional interactions. Here, healthy participants underwent functional
MRI while evaluating mixed gambles presenting potential gains, losses and associated outcome probabilities. Connectivity graphs were constructed from analyses of psychophysiological interactions across a comprehensive atlas of brain regions. Expected value correlated positively with activity within medial
prefrontal and occipital cortices, and modulated effective connectivity across a network that extended
substantially beyond these nodes. Value-sensitive effective connections were found to be arranged as a unitary, small world network in which medial and anteriorlateral prefrontal areas featured as hubs, characterized by dense connectivity and high shortest-path centrality. Further analyses revealed that the observed
effective connectivity effects were more pertinent to dichotomous gain/loss comparisons than to continuous
value determination. Factoring expected value into its constituent components, potential loss modulated
connectivity across a subset of the value-sensitive network, whereas potential gain and outcome probability
were not signicantly embodied in functional interactions. Regional response non-linearity was excluded as
an artifactual basis to the observed effects, and directionality inferences were conrmed by comparison of
dynamic causal models. Our ndings extend current literature demonstrating that the representation of
value is dependent on distributed processing taking across a widespread network which feeds information
into a limited set of integrative prefrontal nodes. This study also has more general paradigmatic implications
for neuroeconomics, demonstrating the value of explicit modeling of inter-regional interactions for understanding the neural substrates of decisional processes.
2012 Elsevier Inc. All rights reserved.
Introduction
Many aspects of everyday life such as crossing a road, choosing to
eat, spending or saving money, confronting wild animals in hunter
gatherer societies, require the ability to take decisions effectively.
Typical decisional situations are characterized by multiple options,
each carrying potential rewards, risks and associated outcome
probabilities. Financial gambles provide a powerful tool to investigate
systematically how motivational decisions are reached in the presence of risk. For example, a basic gambling game can be proposed
where individual gambles differ in values for potential loss (kLOSE),
potential win (kWIN) and outcome probability (pWIN). If one has to
Corresponding author at: Scientic Department, Fondazione IRCCS Istituto Neurologico Carlo Besta, Via Celoria 11, 20133 Milano MI, Italy.
E-mail address: lminati@ieee.org (L. Minati).
1053-8119/$ see front matter 2012 Elsevier Inc. All rights reserved.
doi:10.1016/j.neuroimage.2012.02.048
choose among such mixed gambles, or decide whether or not to

take an individual gamble, the locally optimal strategy would be to
calculate the expected value (EV = kWIN pWIN kLOSE (1 pWIN))
for each option and systematically choose the ones with the highest
EV (Machina, 1982; von Neumann and Morgenstern, 1944). However,
in the real world biases such as risk-aversion (i.e. potential losses are
overrepresented relative to gains) and loss-aversion (i.e. large potential losses are preferentially avoided irrespective of overall value)
exert powerful inuences. Prospect theory is a prominent account of
realistic decision making, in which deviations from purely utilitarian
behavior such as risk aversion are represented as non-linear value
functions. Particularly, prospect theory postulates that choosing
under risk involves an initial editing step, wherein the apparent
economic parameters (e.g., kWIN, kLOSE and pWIN) are integrated into
abstract value representations (e.g., EV), followed by an evaluation
step that compares the EV of each option to a reference point
2192
L. Minati et al. / NeuroImage 60 (2012) 21912205
determined by the EVs of all options available in a given situation

(Kahneman and Tversky, 1979; Tversky and Kahneman, 1992;
Wilkinson, 2008).
Published neuroeconomic studies of motivational decisionmaking indicate that both the evaluation of different choice options,
and the process of choosing between them, involve an interplay between associative regions, mainly within prefrontal cortex, and
phylogenetically-older structures including striatum, insula, cingulate
cortex and amygdala (Alexander and Brown, 2011; Andersen and Cui,
2009; Clark, 2010; Kable and Glimcher, 2009; Levine, 2009; Platt and
Huettel, 2008; Rangel and Hare, 2010; Rushworth and Behrens, 2008;
Trepel et al., 2005; Wallis and Kennerley, 2010; Wang, 2008). To a
certain extent, it is possible to link subcomponents of decisionmaking processes to specic brain regions. In current views, the
dorsal and ventral lateral prefrontal cortices (DLPFC, VLPFC) and parietal lobe are implicated in integrative computation of EV, whereas abstract value representation involves primarily ventromedial and
dorsomedial prefrontal cortices (VMPFC, DMPFC) and adjacent orbitofrontal cortex (OFC). Ascending dopamine pathways, projecting to
the striatum and medial PFC, are associated with the representation
of reward through prediction error signals that support motivational
learning. In parallel, the anterior insular and cingulate cortices,
which are linked to generation and representation of emotional bodily arousal, are involved in the representation of risk and uncertainty
(Alexander and Brown, 2011; Andersen and Cui, 2009; Clark, 2010;
Critchley, 2005; Kable and Glimcher, 2009; Levine, 2009; Platt and
Huettel, 2008; Rangel and Hare, 2010; Rushworth and Behrens,
2008; Trepel et al., 2005; Wallis and Kennerley, 2010; Wang, 2008).
Most published functional neuroimaging studies in this area are
reliant on univariate analytic techniques to establish direct relationships between specic economic variables and regional brain activity.
The approach provides valuable insights, but it treats each region in
isolation rather than as a node in a dynamical network. In contrast,
a recent alternative perspective is that observable behavior arises as
an emergent property of multiple areas interacting as a network
(e.g., Sporns, 2009; Stephan, 2004).
Animal neuro-histological studies, and more recent human in-vivo
neuroimaging investigations of structural and functional connectivity,
suggest that topological features inherent in the organization of cortical networks underpin the brain's capacity to combine efciently
functional specialization and integration (Buckner et al., 2009;
Hagmann et al., 2008; Sporns, 2009; Sporns and Tononi, 2001;
Sporns et al., 2000; Sporns et al., 2004). Specically, the density of
node connections (referred to as degree in the context of graph
theory) is distributed very inhomogeneously across the cortex, with
a small number of well-dened, discrete areas displaying a disproportionately large number of links (Hagmann et al., 2008; Sporns, 2009;
Wang et al., 2010; Zemanov et al., 2006). Such key areas are hypothesized to act as hubs, with the dual purpose of non-linearly integrating information from multiple afferent connections and of
relaying signals, thus optimizing the wiring and metabolic costs of
long-range interactions (Bassett and Bullmore, 2006; Bullmore and
Bassett, 2011; Guimer et al., 2005; Zamora-Lpez et al., 2010). Individual cortical hubs represent convergencedivergence zones,
which exchange information bi-directionally with earlier sensory
and association cortices to encode specic combinatorial arrangements of knowledge (e.g., representing an object or an abstract situation; Meyer and Damasio, 2009). The collective action of cortical hubs
leads to the emergence of segregated large-scale networks that correspond to behavioral abilities in specic neuropsychological domains
(Mesulam, 1998).
Univariate analysis alone is unable to provide a comprehensive neural account of risky decision-making, which involves complex interactions between multiple cognitive processes (e.g., working memory,
strategic and probabilistic reasoning, and the abstract representation
of relative magnitudes) integrated with affective and physiological
response. Within a realistic neuroeconomic framework, relevant economic parameters are encoded not only as the intensity of activity within specic regions but also at network level, as modulations of dynamic
connectivity between cortical hubs.
Among multivariate techniques, psychophysiological interaction
(PPI) tests for the changes in the (physiological) inuence of one
source region on the activity of a second target region as a result
of modulation by an external (psychological) factor, such as an economic parameter in a decisional task. If such an interaction is
shown, the inference is that target region functionally integrates
neural information from the source region with the taskdetermined psychological context (Friston et al., 1997). Another technique for examining communication across brain regions is dynamic
causal modeling (DCM), whereby brain networks are treated as dynamic systems that are subject to external perturbation. DCM is
used to gain quantitative estimates of the magnitude and direction
of causal interaction between brain regions, by representing neuronal
activity as a hidden variable within a biophysical model of observable
hemodynamic responses. One can use Bayesian comparisons of prespecied DCMs to achieve the best model of parameter effects on
inter-regional effective connectivity (Friston et al., 2003; Stephan et
al., 2007). To date there are only a few published studies applying
connectivity analyses to nancial decision-making: During reward
processing activity in the ventral striatum is closely mirrored by
responses across the insula, amygdala, hippocampus and orbitofrontal cortex (Camara et al., 2008). During a study of charitable giving,
VMPFC activity integrated inputs from the anterior insula and temporal cortex to reect the amount donated (Hare et al., 2010). Altered
value-dependent coupling between sub-genual cingulate and amygdala is observed when monetary rewards are linked to physical pain
(Park et al., 2011). Each of these studies used a small number of
seed regions of interest, that were dened either on the basis of
results from univariate analyses or a priori hypotheses.
A complementary approach is to parcellate all brain areas potentially involved in task performance and perform PPI analyses across
all of them to identify networks of connections modulated by relevant
economic parameters. This approach has potential advantages in that
it embeds minimal assumptions and yields whole-brain effective connectivity graphs that can be evaluated using established topological
metrics, supporting inferences on the integrative role of each region
and enabling the discovery of the cortical hubs involved in decisionmaking (Bassett and Bullmore, 2006; Camara et al., 2008; Camara et
al., 2009; Salvador et al., 2005; Zamora-Lpez et al., 2010). For example, this type of graph-based network analysis applied to functional
resting-state data highlights abnormalities in large scale dynamics
in early-stage dementia and schizophrenia (see Wang et al., 2010
for a review). Graph theory provides a wide range of topological metrics characterizing individual nodes, three parameters being particularly relevant in this context: i) degree, representing the number
of connections for a given node, ii) betweenness centrality, representing the number of shortest paths passing through a given node
and iii) clustering coefcient, representing the density of
connections across the network neighbors of a given node. On
graphs of effective connectivity, nodes exhibiting comparatively
large values of degree and betweenness centrality and small clustering coefcients are likely to act as hubs integrating information
from a representation distributed over multiples areas (Bassett and
Bullmore, 2006; Bullmore and Bassett, 2011; Newman, 2010;
Sporns, 2009; Wang et al., 2010).
Here, we sought to gain more comprehensive understanding of the
neural bases to risky decision-making by considering in detail the
processing of gambles from a network perspective. Healthy participants performed a functional MRI study during which they evaluated
and decided whether to accept or reject mixed gambles in the
absence of feedback. Effective connectivity graphs were constructed
from PPI analyses with the aim of identifying key integrative structures
supporting the representation of the economic parameters characterizing the task. Additionally, we tested i) if the main inferences drawn
through PPI analysis on effective connectivity directionality are
concordant with Bayesian comparisons of DCMs, ii) if the observed
PPI effects are, in fact, artifactual consequences of different response
non-linearities across brain regions, iii) if approximating EV by its
sign function leads to results similar to those obtained for continuous
EV, iv) if the observed PPI effects are also apparent when choicerelated variance is removed and v) if similar PPI effects are observed
for accepted and rejected gambles separately. For succinctness, these
additional analyses were only performed for EV, as the strongest driver
of decisional behavior (Kahneman and Tversky, 1979; Wilkinson,
2008).
Methods
Participants
Nineteen healthy subjects (8 females, age 36 7 years, formal
education 16 3 years) participated in the experiment, which was
conducted at the Fondazione IRCCS Istituto Neurologico Carlo Besta
(FINCB) and formally approved by the local institutional review
board (project no. fMRI-DM) and by the research governance and
ethics committee of the Brighton & Sussex Medical School (BSMS;
PhD project no. 10/056/MIN). After being informed of the study purpose, all participants provided written informed consent. No money
or material compensation was provided, and in accordance with
local regulations participants were not playing with real currency.
The dataset used for this study was originally acquired in the context
of a univariate correlational investigation published separately
(Minati et al., 2012); due to data storage problems (loss), data from
3of the initial 22 participants could not be used. All participants
were naive to the experiment, right-handed according to the
Edinburgh Handedness Inventory (Oldeld, 1971) and without a
history of neurological or psychiatric disorders. None was taking
medication affecting the central nervous system.
Gambling task
Participants viewed a back-projection screen covering approximately one third of the visual eld through a mirror mounted on
the head coil. Each gamble consisted of three parameters, presented
2193
visually as white numbers (Arial font) on a black background: potential loss (thereafter referred to as kLOSE, top left of the screen), potential win (kWIN, top right) and probability of winning (pWIN, bottom
central; Fig. 1a). From these parameters, expected value was derived
according to EV = kWIN pWIN kLOSE (1 pWIN); the average
amount at stake, referred to with kWIN,kLOSE, was also considered
to analyze the fMRI data. The task was written in Matlab (Mathworks
Inc., Natick MA, USA) using the Cogent 2000 toolkit (Wellcome Laboratory of Neurobiology, UCL, London, UK).
Participants were requested to evaluate the gambles in terms of
gut feeling without attempting any mathematical computation,
giving equal consideration to the three parameters. They were
asked to accept or reject each gamble rapidly, within a maximum of
6 s, and informed that no feedback would be provided but the computer would keep track of their responses in a virtual bank-account,
starting from a balance of zero and updated with the amount won
or lost each time. They were also made aware that the outcomes
were not pre-determined, but would be decided by the computer
each time the response button was pressed. They were encouraged
to earn as much money as possible by deciding as though they were
playing with real currency. Prior to scanning, all participants
practiced on ten sample gambles.
Responses were made using an MRI-compatible keyboard
according to a four-level scale, i.e. condent reject (left middle),
unsure reject (left index), unsure accept (right index) and condent accept (right middle). Participants were asked to express their
judgment using all four levels but told that there was no difference in
the consequence of condent and unsure responses. This fourlevel scale was used to promote reection on each gamble, rather
than use of a xed rule (Tom et al., 2007). Upon response, each gamble
disappeared immediately and was replaced by a white xation cross.
The gambles presented during each session were randomly
extracted from a pool whose characteristics are shown in Fig. 1be.
The range for kWIN and kLOSE was between 0 and 300 , while pWIN
was between 0.1 and 0.9. The distributions of kWIN and kLOSE were
comparable (MannWhitney test p = 0.5) and the linear correlation
coefcient with respect to EV was similar across kWIN (r = 0.29),
kLOSE (r = 0.31) and pWIN (r = 0.24). This balancing ensured that the
properties of the stimuli would not bias the evaluation toward one
parameter over the others.
Each scanning session comprised four task blocks, each lasting
9 min. and followed by a pause of 3 min during which structural
Fig. 1. Stimuli and behavioral responses. a) Gamble presentation format. be) Distributions of kWIN, kLOSE, pWIN and EV in the gamble pool. f) Gamble acceptance probability (w) as a
function of EV; the dotted lines denote 1 SD over participants.
2194
imaging sequences were performed. Across the four blocks, 152 gambles were presented with a constant inter-trial interval of 14.4 s. The
rst 4 gambles after start and each pause were considered as warmup trials and discarded.
Data acquisition and preprocessing
Imaging was performed at the neuroradiology department of the
FINCB by means of a 1.5 T MR scanner (Magnetom Avanto, Siemens
AG, Erlangen, DE), equipped with an 8-channel head matrix coil.
Functional images sensitized to bloodoxygen level-dependant
(BOLD) contrast were obtained using T2-weighted echo-planar sequences with TR = 2400 ms, TE = 45 ms, FoV 315 210 mm, matrix
90 60, 35 slices, 3.5 mm, no gap (isotropic voxel). For each task
block, 240 volumes were acquired, yielding a total of 960 time points
per scanning session. Corresponding phase/magnitude eld-maps
were acquired with a gradient-echo sequence having TR = 560 ms,
TE = 4.7/9.5 ms. Anatomical scans were obtained using a volumetric
T1-weighted magnetization-prepared rapid-acquisition gradientecho sequence, with TR = 1640 ms, TE = 2 ms, FoV 256 256 mm,
matrix 256 256, 160 slices, 1 mm, no gap.
The functional imaging data were pre-processed using SPM8
(Wellcome Trust Centre for Neuroimaging, UCL, London, UK) running
under Matlab 7. All processing steps were completed on a Sun Blade
2500 workstation (Sun Microsystems Inc., Palo Alto CA, USA). After
slice-timing correction, functional volumes were realigned within
each session and un-warped using the corresponding eld-maps.
The average functional volume was co-registered with the individual
anatomical scan utilizing an elastic transformation. The anatomical
scan was thereafter segmented to obtain normalization matrices,
which were subsequently applied to transform the functional
volumes into MNI space. Spatial smoothing was then performed
using an 8 mm FWHM kernel.
Subsequent processing steps were performed using software developed in-house. The time-courses of individual MNI-space voxels
were ltered by removing the variance related to head translation
and rotation using linear regression and by removing long-term
uctuations tting the baseline with a third-degree polynomial and
subtracting it.
prefrontal cortices (VMPFC and DMPFC), orbitofrontal cortex (OFC),

caudate nucleus (CAUD), lenticular nucleus (LENT), ventral striatum
(VS), thalamus (THAL), amygdala (AMYG), hippocampus and parahippocampal gyrus (HPP).
These ROIs were initially dened on the basis of the automated
anatomical labeling atlas (Tzourio-Mazoyer et al., 2002; Fig. 2) and
manually rened to closely match anatomical boundaries on the
high-resolution canonical brain provided in SPM8 by two experienced
operators (LM and MG) using the MRIcro software (GSU/GT Center for
Advanced Brain Imaging, Atlanta GA, USA). The Niewenhuys atlas was
utilized as principal reference on gyral anatomy (Niewenhuys, 1996).
Cingulate cortex was subdivided according to the areas dened by
Vogt (2005). The prefrontal cortex was parcellated considering the criteria described in Rajkowska and Goldman-Rakic (1995), Fuster (1997)
and Barbey and Grafman (2011).
Response extraction and univariate analysis
Regional time-courses were extracted by averaging over all voxels
within each ROI having relative gray matter content exceeding 50%,
as determined by the segmentation at pre-processing. The timecourses were epoched between 2.4 s (1 TR) before and 14.4 s
(6 TRs) after each gamble onset, and the relative blood oxygenation
level dependent signal percent change (BOLD%) derived with respect to the pre-stimulus baseline. For each gamble, the response amplitude was measured by averaging the BOLD% between 4.8 s and
9.6 s post-stimulus, corresponding to the peak of the hemodynamic
response (Fig. 3). The response amplitudes were z-normalized across
trials, for each region in each participant and the effect of variation in
RTs was removed by linear regression.
As a rst analysis, we sought to determine the direct effect of the
economic parameters (kWIN, kLOSE, pWIN, EV and kWIN,kLOSE) on
regional responses and performed linear regressions over the pooled
z-normalized response amplitudes of all participants, for each combination of parameter and region. We also determined whether,
irrespective of the economic parameters, regional responses were
signicantly different from zero by means of one-sample t-tests. We
accounted for multiple comparisons over the 42 regions controlling
the family-wise error (FWE) by applying BonferroniHolm's correction (Holm, 1979).
Behavioral response analysis

Psychophysiological interaction (PPI) modeling
The reaction times (RTs) were z-normalized at individual level,
and trials for which they fell outside 3 SDs from the average or
exceeded 6 s were removed. For each participant, the proportion of
positive-EV gambles accepted and negative-EV gambles rejected,
subsequently referred to as EV-discrimination, was calculated. The
total amount earned, dened as the sum of the EVs of the accepted
gambles, was also determined. To obtain a measure of risk aversion,
two separate logistic regressions were performed modeling gamble
acceptance as a function of kWIN and kLOSE, and the regression coefcient ratio = LOSE / WIN was calculated (as in Tom et al., 2007).
Further, individual decision-weighting curves representing gamble
acceptance probability w as a function of EV were generated.
Regions of interest (ROIs)
On the basis of literature on the functional neuroanatomy of nancial decision-making (e.g. Kable and Glimcher, 2009; Platt and Huettel,
2008; Rangel and Hare, 2010; Trepel et al., 2005), the following regions
were considered a priori to be of particular interest: primary motor
cortex (limited to hand knob, MI), supplementary motor area (SMA),
angular gyri (ANG), precuneus (PREC), early associative visual areas
(OCC), anterior, middle and posterior (ACC, MCC and PCC) cingulate
cortices, anterior insula (AINS), ventral-, dorsal- and antero-lateral prefrontal cortices (VLPFC, DLPFC and ALPFC), ventral- and dorsal-medial
We next applied PPI analysis to search for connections between pair

of regions modulated by the economic parameters. In this analysis, a
separate multilinear regression was performed for each parameter
and pair of putative source and target regions, according to
y = 0 + 1 k + 2 x + 3 x k, where x and y represent the
z-transformed response amplitudes of the source and target regions
respectively, 0 is a constant term, 1 corresponds to the direct effect of
the economic parameter k, 2 corresponds to the direct effect of activity
in the source region (i.e., intrinsic functional connectivity) and 3
corresponds to the interaction between the parameter k and activity
in the source region (i.e., effective connectivity).
Adjacency matrices for the connections modulated by each economic parameter (kWIN, kLOSE, pWIN, EV and kWIN,kLOSE) were constructed by thresholding the signicance of the 3 term at p b 0.05
FWE-corrected with BonferroniHolm's correction over the 1722
(i.e., 42 (42 1)) possible directed region pairs. To reduce connection clutter for graph visualization purposes, adjacency matrices
were also generated for p b 0.01 FWE.
Directed graphs of effective connectivity for each parameter were
drawn from the adjacency matrices using the Force Atlas algorithm
implemented in the Gephi 0.8 program (Bastian et al., 2009). For each
node, the degree (i.e., total number of connections), indegree (i.e.,
number of afferent connections) and outdegree (i.e., number of
2195
Fig. 2. Regions-of-interest (ROIs) used for signal extraction. The co-ordinates are given in MNI space. See Methods section for list of abbreviations.
efferent connections) were computed (e.g., Bullmore and Bassett,

2011; Newman, 2010). The betweenness centrality, dened as the
number of shortest paths passing though a node as given by Dijkstra's
algorithm, was also calculated, taking into account connection
directionality (Brandes, 2001). Further, the clustering coefcient
was obtained by triangle computations, taking into account
connection directionality (Latapy, 2008).
EV was observed to be the strongest driver of behavior and the
parameter for which the largest number of effective connections
was observed. We therefore tested by means of a one-sample t-test

whether the corresponding effective connectivity network for EV
as a whole had small-world features, i.e. larger average clustering
coefcient than 20 corresponding ErdsRnyi random graphs,
matched for number of connections, nodes and average path length.
The small-worldness coefcient S was computed to quantify
the topological effect of clustering with respect to random
connections (Humphries and Gurney, 2008). We also tested for signicant modularity, i.e. presence of distinct network sub-clusters,
Fig. 3. BOLD% responses evoked by gamble presentation, averaged across all participants and stimuli. The error bars denote 1 SD over participants.
2196
using algorithms described in detail elsewhere (Blondel et al.,

2008).
To deal with the possible emergence of artifactual PPIs from different response non-linearities across regions, we repeated the PPI
analysis removing its direct effect on responses in source and
target regions up to the third degree. In other words, we considered
0
2
3
y^ 0 2 x^3 x^k , where x^ x 0 1 EV 2 EV 2 3 EV 3 and
similarly for y^ . Further, given that gamble acceptance increases very
rapidly in the EV 0 region (see Results section), we hypothesized
that the effect of EV could be similar to that of a step transfer
function. We therefore repeated the PPI analysis approximating EV
with its sign function and compared the resulting matrices of the
F-values for the 3 term across pairs of regions, representing these
as heatmaps.
To investigate to what extent the observed PPI effects for EV are
related to value computation and representation rather than choice,
we repeated the PPI analysis removing, for each targetsource
pair, the direct effect of choice on the target region and its interaction with the response of the source region. Further, we repeated
the PPI analysis considering separately accepted and rejected
gambles.
Dynamic causal modeling (DCM)

An additional analysis was performed to conrm the concordance
of the directionality inferences drawn through PPI with respect to
DCM10 (Wellcome Trust Centre for Neuroimaging, UCL, London,
UK). Separate DCM analyses were undertaken for the ve most
densely interconnected regions on effective connectivity for EV
(bilateral ALPFC, bilateral DMPFC, right VMPFC; see Results section),
modeling directed links with their neighbors within the prefrontal
cortex. For each ROI, the rst eigenvariate of the time-course was
extracted and individually adjusted for the main effect of stimulus
presentation as well as for the nuisance movement regressors. Individual bilinear, deterministic one-state DCMs (Stephan et al., 2007)
were then specied for each target region. Two models were constructed and contrasted for each region: the rst modeled afferent
connections in the direction inferred from PPI analysis and the second
modeled the connections in the opposite direction. For both models,
EV was entered as modulatory parameter for all connections and directly inuencing the target region. For this part, gamble presentation
was entered as mini-blocks with duration corresponding to the measured RTs. After estimation at the level of individual participants, the
models were entered in a random-effects Bayesian model selection
analysis, which yielded group-level model exceedance probabilities

for each pair of alternative models (Stephan et al., 2009).
Results
Behavioral responses
Participants discriminated positive and negative EV beyond
chance (78 8% vs. 50%, one-sample t-test p b 0.001) and all earned
a positive amount of virtual money (1650 540 vs. 0 , onesample t-test p b 0.001). The reaction times were 2.5 0.9 s, and the
6 s limit was never exceeded. The regression coefcient ratio indicated signicant risk-aversion (1.3 0.5 vs. 1, one-sample t-test
p = 0.02). Gamble acceptance probability increased monotonically
with EV (Fig. 1f)
Univariate analysis
The task evoked positive responses bilaterally across the majority
of regions, i.e. MI, SMA, PREC, OCC, MCC, VLPFC, ALPFC, DLPFC, OFC,
DMPFC, AINS, ANG, CAUD, LENT, THAL and VS. In the ACC, positive responses were detected only on the left hemisphere, in the VMPFC signicant negative responses were detected bilaterally. In the PCC,
AMYG and HPP there was no signicant response (Fig. 3).
As expected, strong motor response-related effects were expressed
in MI bilaterally as correlations with kWIN, kLOSE and EV (Fig. 4).
Expected value (EV) was positively correlated with response amplitude
in right OCC (F = 18.8), left ACC (F = 10.6), VMPFC bilaterally (F = 9 for
right, F = 9.9 for left) and right DMPFC (F = 9.6); no negative
correlations were observed for this parameter. Potential loss (kLOSE)
was negatively correlated with response amplitude across left LENT
(F = 18.0), VMPFC bilaterally (F = 17.7 for right, F = 19.4 for
left), left VS (F = 13.5), right DLPFC (F = 12.3), ACC bilaterally
(F = 9.1 for right, F = 12.2 for left), left MCC (F = 10.8), right
ANG (F = 10.0), left ALPFC (F = 9.8), right OCC (F = 9.7), right
PREC (F = 9) and right PCC (F = 8.8); no positive correlations
were observed for this parameter. No signicant correlations were observed for kWIN and pWIN beyond MI. Average amount (kWIN,kLOSE)
was negatively correlated with response amplitude in right ANG (F =
10.1) and right DLPFC (F = 10.4).
Connections modulated by EV
An effective connectivity network was identied for the representation of expected value (EV), in which the right VMPFC, right DMPFC,
Fig. 4. Color-map for the direct effects of kWIN, kLOSE, pWIN, EV and kWIN,kLOSE on regional BOLD responses (blue: negative, red: positive). In this map, negative F-values indicate a
negative relationship and threshold of p(FWE) b 0.05 corresponds to |F| > 8.0. (For interpretation of the references to color in this gure legend, the reader is referred to the web version of this article.)
2197
Fig. 5. Directed inter-regional connections modulated by EV at p(FWE) b 0.01. Red and blue edges respectively denote positive and negative sign of the PPI interaction term (see
Methods section). Node diameter represents degree (i.e., overall number of node connections), and node color (blue through red) represents indegree (i.e., number of afferent
connections). (For interpretation of the references to color in this gure legend, the reader is referred to the web version of this article.)
right ALPFC, left ALPFC and left DMPFC appeared as prominent hubs
(Fig. 5 and Fig. 6). At the FWE-corrected threshold of p b 0.05
(adopted for all network metrics calculations), the completeness
was 9.7%, i.e. 168 directed connections out of 1722 were signicantly
modulated by EV; at p b 0.01 (adopted for display purposes only to
reduce connection clutter), the completeness was 7.1%, corresponding to 122 connections. At both thresholds, 84% of modulated
connections were negative, i.e. had 3 b 0.
The regions with highest degree (Fig. 7), i.e. above the 90th percentile, were the right VMPFC (30 connections), right DMPFC (30),
right ALPFC (29), left ALPFC (21) and left DMPF C (21). The spread
observed for indegree was considerably larger than for outdegree
(SD 6.6 vs. 3.2), i.e. all regions had between 5 and 10 efferent
connections, whereas a small number had a substantially larger number of afferent connections, up to 25. Specically, the regions with
highest indegree, above the 90th percentile, were the right DMPFC
(25 afferent connections), right VMPFC (22), right ALPFC (21), left
DMPFC (17) and left ALPFC (14).
Betweenness centrality was highest, i.e. above the 90th percentile,
for the right ALPFC (141 shortest paths passing through), right
DMPFC (85), right VMPFC (78), left SMA (47) and left VMPFC (44).
Excluding regions with zero clustering coefcients, the lowest clustering values, i.e. below the 10th percentile, were observed in the
right ALPFC (0.13 proportion of connections among neighbors),
right VMPFC (0.16), left OCC (0.17), right DMPFC (0.19) and left
ALPFC (0.19).
Overall, the small-worldness coefcient S was 6.2, indicating that
the graph strongly exhibited the topological features of a small
world network. Accordingly, the average clustering coefcient was

substantially higher than that of comparable ErdsRnyi random
networks (0.37 vs. 0.06 0.03, p b 0.001) matched for mean path
length (1.93 vs. 1.98 0.21, p = 0.4). The modularity coefcient was
0.14, indicating that the network was not signicantly subdivided in
subclusters (with respect to the accepted cut-off of 0.4; see Blondel
et al., 2008).
There were strong correlations between the traditional PPI analysis, the analysis performed removing non-linear EV effects (r = 0.7,
p b 0.001) and the analysis performed approximating EV with its
sign function (r = 0.89, p b 0.001) when comparing the F-values for
the modulation term 3. Accordingly, the distribution of F-values
across regions appeared to follow a very similar pattern for the
three analyses (Fig. 8).
Further, there were strong correlations between the traditional
PPI analysis, the analysis performed removing the direct and
interaction effects of choice (r = 0.68, p b 0.001) and the analysis
limited to rejected gambles (r = 0.45, p b 0.001), even though the corresponding interaction F-values were diffusely smaller. Contrariwise,
there was no signicant correlation with the interaction F-values
computed over accepted gambles, the majority of which were in the
vicinity of zero, except for degenerate values along the diagonal
(Fig. 9).
DCM revealed that, for each of the ve hub regions under
consideration (Fig. 10), the exceedance probability was signicantly
higher for the model assuming afferent connections as discovered
by the PPI analysis than for the model in which the direction of
these connections was inverted, i.e. 0.88 vs. 0.12 for left ALPFC, 0.88
2198
Fig. 6. Breakout of the connections modulated by EV (shown in Fig. 5), highlighting separately the connections of the most highly interconnected nodes: right and left ALPFC (a and
b), right VMPFC (c), right and left DMPFC (d and e).
vs. 0.12 for right VMPFC, 0.80 vs. 0.20 for left DMPFC, 0.77 vs. 0.23 for
right DMPFC and 0.90 vs. 0.10 for right ALPFC.
Connections modulated by the other economic parameters
A smaller effective connectivity network was discovered for potential loss (kLOSE, Fig. 11), in which the majority of connections overlapped corresponding the EV-sensitive network (31 out of 38, 82%). In
this simpler network, the right ALPFC and right VMPFC remained as
hubs. At the FWE-corrected threshold of p b 0.05, the completeness
was 2.2%, i.e. 38 directed connections out of 1722 were signicantly
modulated by kLOSE; at p b 0.01, the completeness was 1.0%, corresponding to 17 connections. At both thresholds, the majority of modulated connections were positive, i.e. had 3 > 0 (86% at p b 0.05, 94%
at p b 0.01). The regions with highest degree, i.e. above the 90th percentile, were the right ALPFC (14 connections), left ALPFC (10),
right DLPFC (6) and right VMPFC (6). The regions with highest indegree were the right ALPFC (11 afferent connections), left ALPFC (8),
right VMPF C (6) and left ACC (3). Since the path length and neighbor
connectivity are not representative for such small networks (e.g.,
Newman, 2010), other network metrics were not computed.
No signicant effective connectivity was observed for potential
win (kWIN), probability of winning (pWIN) or average amount magnitude (kWIN,kLOSE) was observed.
Discussion
The present experiment was motivated to extend our understanding of the processing of individual mixed gambles, i.e. choice options presenting a potential reward associated with a risk, beyond
the representation of individual economic parameters in regional responses to the level of network dynamics. From a neuroeconomics
perspective, detailed analysis of the processing of simple, individual

monetary gambles is highly relevant, as it enables the isolation of primary neural correlates of adaptive action selection behaviors from
complicating factors such as uncertainty (dealing with incomplete information) and ambiguity (ignorance of outcome probabilities), reward prediction error signaling and learning (such as found in the
presence of outcome feedback), and bargaining and social behavioral
(such as found when playing with an opponent). Further, extending
neuroeconomic investigations beyond a purely localizationalist approach is necessary from a theoretical viewpoint, considering that
decision-making behaviors are highly abstract and multifactorial,
and therefore unlikely to be adequately explicable in terms of activity
taking place with isolated regions. Graph-based analysis of effective
connectivity networks explicitly addresses two related issues; i)
whether the neural representation of value and other economic parameters is widely distributed or limited to interactions among a restricted set of regions and; ii) whether the regions where stimulus
value is reported to correlate strongly with response amplitude (e.g.
ventromedial PFC and striatum) also stand out as informationintegrating hubs during a decisional task.
Our mixed-gambling task successfully reproduced the key behavioral economic observation of risk aversion: participants overemphasized potential losses relative to equivalent gains. The effect was
smaller in comparison with scenarios where there is substantial economic involvement (i.e., logistic regression coefcient ratio = 1.3
when compared to ~ 2 in Tversky and Kahneman, 1992;
Abdellaoui et al., 2007; Tom et al., 2007). However the reproducibility
of risk aversion was reassuring in that participants were not simply
approximating a mathematical computation, but applying similar
decisional criteria to those that lead to the adaptive biases observed
in real-world decision-making (Kahneman and Tversky, 1979;
Wilkinson, 2008). Indeed, previous behavioral economic studies
2199
Fig. 7. Network statistics for inter-regional connections modulated by EV, calculated at p(FWE) b 0.05. Indegree: number of afferent connections; outdegree: number of efferent
connections; degree: overall number of connections; betweenness centrality: number of shortest paths among all possible nodes that pass through a given node; clustering coefcient: density of connections among the neighbors of a given node.
demonstrate that signicant deviations from locally optimal behavior

can be obtained in the absence of real monetary endowment, and that
the availability of small rewards may even elicit paradoxical effects
(Gneezy and Rustichini, 2000).
Consistent with an earlier traditional voxel-wise appraisal of this
dataset using univariate Gaussian-eld inference methods (Minati
et al., 2012), the univariate analysis of average ROI responses presented here revealed representation of expected value (EV) in the
amplitude of regional responses within early associative visual areas
and ventral and dorsal medial prefrontal cortices (Fig. 4). Our ndings
here add to the evidence for prefrontal gain brain circuits that track
potential gains, and in which medial prefrontal cortex harbors abstracted representations of value and potential reward across classes
of actions and stimuli (e.g., Alexander and Brown, 2011; Chib et al.,
2009; Glscher et al., 2009; Smith et al., 2010; Wunderlich et al.,
2009). Signicant correlations between economic parameters and
visual activity are noteworthy, and plausibly reect back-projected
effects of salience from prefrontal, cingulate, amygdalar and brainstem regions (Critchley, 2005; Serences, 2008; Vuilleumier et al.,
2004; Wendt et al., 2011).
Fig. 8. Comparison of F-values for the modulatory effect of EV under i) traditional PPI analysis, ii) PPI analysis in which the effects of EV on activity in the source and target regions
have been removed up to the third degree (i.e., x^ x 0 1 EV 2 EV 2 3 EV 3 ), iii) PPI analysis where continuous EV was replaced by its sign function. Blue denotes negative,
red positive sign (range 50). (For interpretation of the references to color in this gure legend, the reader is referred to the web version of this article.)
2200
Fig. 9. Comparison of F-values for the modulatory effect of EV under i) traditional PPI analysis, ii) PPI analysis in which the direct and interaction effects of choice (accept or reject)
on activity in the target region have been removed, iii) traditional PPI analyses performed separately for accepted and rejected gambles. Blue denotes negative, red positive sign
(range 50). (For interpretation of the references to color in this gure legend, the reader is referred to the web version of this article.)
Our network-discovery analysis for effective connectivity revealed

a network of striking richness (Fig. 5). While only ve regions
displayed correlation between EV and response amplitude, EV
modulated effective connectivity between 36 regions (when
stringently correcting for multiple comparisons over the 1722 directed connections rather than the 42 nodes). One immediate inference is
that the representation of value is deeply embedded within dynamical interactions among brain regions, rather than only in the intensity
of local activity. Since effective connectivity is conceptually distinct
and mathematically independent from direct parametric correlation
(Friston et al., 1997; Friston et al., 2003; Stephan et al., 2007), it is
interesting to identify regions expressing their contributions to the
encoding of EV differentially, i.e. where there is uncoupling of
regional activity and network contribution. Anteriorlateral prefrontal cortex exemplies this: direct correlations between response
amplitude and EV were clearly absent (p = 0.7 for right, p = 0.4 for
left) yet the number of signicant EV-modulated connections was
very large (29 for right, 21 for left, both above the 90th percentile of
the degree distribution).
In recent years, it has been consistently observed that selforganized biological, social and technological networks display
small-world topology. This suggests that small-worldness is a pervasive ecological plausible feature of dynamical collective behaviors
(Humphries and Gurney, 2008). Our nding of small-world topology

within brain systems supporting adaptive decision-making extends
such observations to the level of the neurobiological mechanisms underlying the motivational, microeconomic behavior of the individual.
Small-worldness generally represents an optimum trade-off between
local and global connectivities, providing efcient information transfer with limited connectivity complexity (Bassett and Bullmore,
2006; Bullmore and Bassett, 2011; Ginestet et al., 2011; Newman,
2010; Watts and Strogatz, 1998). More specically, simulations of
neuronal systems demonstrate that small-world networks display
dynamical properties that are very different from corresponding random assemblies, notably drastically increased long-large synchronizability of activity and the emergence of specic attractors (Barahona
and Pecora, 2002; Bullmore and Bassett, 2011; Ginestet et al., 2011;
Newman, 2010; Watts and Strogatz, 1998). We speculate that, in
this context, small-worldness emerged from the evolutionary optimization of information transfer across regions involved in adaptive
survival-related behavior. This view is strengthened by the relative
lack of modularity, suggesting that the distributed representation of
value is not built around specic sub-modules implementing
segregable operations.
For each node, we computed multiple metrics representing independent features representing its local connectivity (degree, indegree
2201
Fig. 10. Alternative dynamic causal models (DCMs) for the directionality of inter-regional connections modulated by EV involving the most highly interconnected nodes: right and
left ALPFC (d and a), right VMPFC (c), right and left DMPFC (b and e). The bar charts denote model exceedance probability for the DCMs on the left and right.
and outdegree), its position in the directed network (betweenness

centrality) and the structure of connections across its neighbors
(clustering coefcient). These node parameters implicate the dorsomedial prefrontal cortex, right ventro-medial prefrontal cortex and
the anteriorlateral prefrontal cortex as key network hubs (Figs. 6
and 7). In the context of structural and functional brain connectivity,
large degree, large betweenness centrality and small clustering coefcient are the hallmarks of integrative cortical nodes (Buckner et al.,
2009; Hagmann et al., 2008; Sporns, 2009; Sporns and Tononi,
2001; Sporns et al., 2000, 2004). In the context of international air
trafc, nodes displaying these features are referred to as connector
hubs and are highly critical to sustain efcient long-haul transfer of
goods and passengers (Guimer et al., 2005). Our ndings provide
Fig. 11. Directed inter-regional connections modulated by kLOSE at p(FWE) b 0.01. Red
and blue edges respectively denote positive and negative sign of the PPI interaction
term (see Methods section). Node diameter represents degree (i.e., overall number of
node connections), and node color (blue through red) represents indegree (i.e., number of afferent connections). (For interpretation of the references to color in this gure legend, the reader is referred to the web version of this article.)
novel evidence that advances our understanding of the neural substrates of risky decision-making by providing the rst explicit demonstration of the integrative role of these regions at the level of effective
connectivity. A substantial body of literature reporting parametric
correlation between monetary and subjective value and responses
in these areas already existed (e.g., Alexander and Brown, 2011;
Andersen and Cui, 2009; Clark, 2010; Kable and Glimcher, 2009;
Levine, 2009; Platt and Huettel, 2008; Rangel and Hare, 2010;
Rushworth and Behrens, 2008; Trepel et al., 2005; Wallis and
Kennerley, 2010; Wang, 2008), but remained open to the criticism
that the observation of a local correlation could reect a distal or a distributed process, rather than a primary causal relationship between regional activity and value representation processes (Ramsey et al.,
2010). In other words, it could not be completely ruled out that
correlations with medial prefrontal cortex activity could at least partly
arise as a secondary expression of processing taking place elsewhere
(Philiastides et al., 2010). Our results demonstrate that the medial
and anterior lateral prefrontal cortices have a key integrative role,
displaying converging effective connectivity involving nodes that do
not even display signicant activity correlations upon univariate
analysis.
Hubs in dynamical networks may serve diverse functions, in spite
of having similar topological properties of their connections. Notably,
one can distinguish between hubs that mainly provide communication facilities while performing limited local computation, and those
that also contain signicant processing and memory infrastructure
(Barrat et al., 2008; Bullmore and Bassett, 2011; Newman, 2010;
Sporns, 2009). Here, we are unable to denitively distinguish
between these roles since no information is available about the
temporal dynamics of inter-regional interactions, however we can
hypothesize that processing nodes could also exhibit a direct
2202
correlation with EV due to their direct computational engagement,

whereas purely communication nodes would be characterized by
less functionally-specic activity and therefore not exhibit direct
activity correlations. From this perspective, at the chosen signicance
thresholds the bilateral ventro-medial and right dorso-medial
prefrontal cortices would be classied as processing nodes, whereas
the left dorso-medial prefrontal cortex and the anteriorlateral
pre-frontal cortex bilaterally would be classied as communication
nodes. While this distinction must remain speculative for our data,
particularly as it depends on the choice of signicance thresholds, future research should explore whether the relationship between topological parameters and univariate task effects can identify different
types of network hubs in fMRI analysis. As demonstrated by ex-vivo
tracer studies, the ventral- and dorsal-medial prefrontal cortices
have dense bi-directional connections with the orbitofrontal cortex,
dorsal- and ventral lateral prefrontal cortex and anteriorlateral prefrontal cortex, and are also directly interconnected with the cingulate,
hippocampus, amygdala and striatum (Barbas, 2000; Brog et al.,
1993; Ongr and Price, 2000; Ongr et al., 2003). Anteriorlateral
prefrontal cortex is densely interconnected with both dorsallateral
and ventrolateral prefrontal cortices, as well as with subcortical structures (Barbey et al., 2009). Recent meta-analyses of functional imaging studies propose that this region is organized in a hierarchical
form, with the more anterior portion supporting more abstract integrative functions (Badre, 2008; Botvinick, 2008). Since the dorsallateral prefrontal cortex and ventrallateral prefrontal cortex (e.g.,
Sakagami and Pan, 2007; Trepel et al., 2005) are involved in integrating multisensory information into abstract variables that subserve
value computation, the hypothesis of a key integrative role of the mesial and anterior prefrontal cortices accords with the organization of
structural connections.
It is noteworthy that the dorsal and ventral striatum were in a
marginal topological location in the effective connectivity network
for expected value, i.e. were associated with low degree, zero indegree and zero betweenness centrality. This nding suggests that the
striatum does not play a major integrative role in the evaluative
aspects of risky decision-making embodied within this task. It is noteworthy also that a recent behavioral study performed with the same
task stimuli demonstrated normal performance in patients with subcortical neurodegeneration (Minati et al., 2011). Our results therefore
robustly associate primary value discrimination during choice-option
evaluation with medial prefrontal cortical function. However, this
notion is at odds with a recent study where correlations with EV
and individual risk aversion were observed in the striatum (Tom et
al., 2007). We speculate that this discrepancy could be due to the
task design: in our study participants had to weigh potential losses
and gain with their respective outcome probabilities, whereas in the
study by Tom et al. (2007) the outcome probability was xed at 50%
and the decisional problem could therefore be reduced to a simple
comparison between the two magnitudes. The need for more abstract
processing here may have led to a shift toward predominantly cortical representation, reducing the engagement of basal ganglia circuitry. Also, it cannot be ruled out that the correlations observed by
Tom et al. (2007) could have been secondary to reward expectation,
since participants were remunerated with a fraction of their earning,
whereas here no material compensation was available.
Lesion studies have convergently demonstrated that damage to
the ventral (Bechara et al., 1999; Clark et al., 2008) as well as dorsal
(Manes et al., 2002) mesial prefrontal cortex reduces the ability to
discriminate between risk-advantageous and disadvantageous situations. Discrete damage to the orbitofrontal (Manes et al., 2002) and
lateral prefrontal cortices (Clark et al., 2003; Manes et al., 2002) has
less obvious effect on decision-making in the absence of uncertainty,
i.e. when outcome probabilities are explicitly represented. Our results
are in accord with this literature, indicating that the medial prefrontal
cortex harbors key integrative hubs for the representation of
expected value. We provide the additional insight that the anterior

lateral prefrontal cortex may also implement critical integrative
functions, as indicated by the high density of connections. Further
lesions studies are therefore motivated to disentangle the effect of
focal caudal and rostral damage to the lateral prefrontal cortex, and
to test the hypothesis that for a given volume rostral lesions will
have a substantially larger impact on value representation. It is also
important to consider that signicant effects have also been observed
for damage to the anterior insula and amygdala, which appeared
rather peripheral in our reconstruction of the effective connectivity
network (De Martino et al., 2010; Weller et al., 2009).
Importantly, throughout this study all PPI analyses were
conducted on the basis of the directly-measured overall amplitude
of individual hemodynamic responses, rather than using regressors
convolved with the canonical function to disentangle overlapping activations; while signicantly increasing task duration, this approach
has the substantial advantage of removing assumptions of timecourse consistency and linear summation, which have the potential
to generate artifactual interactions in fast-paced designs where the
responses to individual gambles overlap considerably (Duann et al.,
2002; Serences, 2004; Wager et al., 2005). Further, our inferences
on effective connectivity were drawn applying family wise error
through an implementation of Bonferroni's correction that very robustly controls for the probability of type I error (Holm, 1979).
While we cannot exclude outright that our choice of ROIs could
have introduced correlations between neighboring regions caused
by overlapping activity, our network-level inferences are robust
with respect to this potential source of error since the majority of
connections were observed between clearly anatomically-separate
areas. Since some of the regions for which we observed the most
dense effective connectivity did not even display any signicant
parametric correlation between the economic parameters and
regional response, generating the ROI set on the basis of a preliminary
voxel-based univariate analysis would likely have concealed important network elements.
One cannot exclude a priori that non-linear correlations between
EV (or generally any continuous task parameter) and response
amplitude may be present, and non-linear effects are indeed
well-documented in neuroeconomics (e.g., Hsu et al., 2009). This represents a potential problem for any kind of PPI analysis, because if a
target and a source region exhibit activity correlating with EV
with different degrees of non-linearity, the bilinear (3) term of the
PPI equation may become artifactually signicant without a real
underlying functional interaction. We addressed this problem by
performing an additional analysis in which the direct effects of EV
were removed up to the third degree separately from both source
and target regions before modeling the functional interaction. The
results endorsed those of the initial analysis, strengthening condence in the observed patterns of effective connectivity (Fig. 8).
Notably, this is the rst PPI study where the possibility of spurious
interactions due to non-linear effects is explicitly addressed in this
way.
We also gained further insight with respect to the coding of value
by showing that the underlying effective connectivity pattern was
mostly unaltered when the continuous EV was replaced with a sign
function (positive or negative). This indicates that the network is
principally engaged in the discrimination of risk advantageousness
or disadvantageousness, reducing the continuum of abstract value
to a more basic heuristic, i.e. comparing the weighted potential gain
and loss (i.e., kWIN p and kLOSE (1 pWIN)) rather than summing
them continuously (Fig. 8). Accordingly, we did not nd any
parametric modulation in response amplitude or connectivity for
average amount magnitude (kWIN,kLOSE). A dichotomous comparison mechanism is also apparent within the behavioral data, where
the acceptance probability function w(EV) rises steeply in the vicinity
of EV = 0 and is essentially at elsewhere (Fig. 1f). Our results
therefore strongly endorse the view that the mesial prefrontal cortex
acts as fundamentally a decisional comparator (Venkatraman et al.,
2009; Wunderlich et al., 2009).
Inferences on directional connectivity can only be drawn indirectly from fMRI data, i.e. by considering the statistical relationships
among the time series. Specically, the directionality of effective connectivity in PPI analyses is extractable since the bilinear term for a
connection between two regions is different from the same connection modeled in the opposite direction (Friston et al., 1997, 2003).
DCM shares some conceptual similarities with PPI analysis, but is
not equivalent as it embeds an explicit biophysical model of neurovascular coupling and relies on a Bayesian approach for comparing
competing models (Stephan, 2004). We applied DCM to a subset of
frontal regions for the purpose of conrming our inferences based
on PPI analysis. Modeling afferent connections into the medial and
anteriorlateral prefrontal cortices carried much stronger evidence
compared to models assuming the opposite directionality, reassuring
on the validity of our PPI observations (Fig. 10). An alternative
method for conrming PPI network discovery results would be to
use Granger causality, a data-driven method based on the relative
predictability of time-series in the context of linear autoregressive
modeling. While this framework is powerful and makes fewer
assumptions on underlying causal structure as compared to DCM, it
is best suited to steady-state data generated in the absence of
modulating variables (such as EV) rather than event-related data incorporating modulating factors; furthermore, its application to fMRI
is still controversial due to several factors including the potentially
confounding effect of hemodynamic responses and the slow sampling
rate (TR) (Bressler and Seth, 2011; Valdes-Sosa et al., 2011). Although
for these reasons we did not apply Granger causality in the present
study, in future work we will examine whether Granger causality
can reveal additional features of causal connectivity in our data.
Univariate correlations and effective connectivity analyses
indicated that potential gain (kWIN) and outcome probability (pWIN)
are not signicantly represented in BOLD activity as individual
parameters (Fig. 4). However, since both are necessary for the calculation of EV, they must be reected in neural activity at some level; a
plausible interpretation of our nding is that their coding is subtler in
comparison to potential loss (kLOSE) and does not generate sufcient
uctuations in the intensity of post-synaptic activity to reect on
the detectable BOLD signal. By contrast, signicant negative
correlations between kLOSE and response amplitude were observed
across the medial and lateral prefrontal cortices, cingulate cortex, occipitalparietal cortex and basal ganglia. This effect goes along the
gain brain notion that potential losses are encoded as diffuse deactivations but, remarkably, in comparison with the previous study by
Tom et al. (2007) here the mismatch in the representation of potential losses and gains was substantially larger, with no positive correlation detected for potential gain in any region. We hypothesize that
this result, which is in keeping with the previous voxel-based univariate analysis of this dataset (Minati et al., 2012), is due to the fact that
our task demanded abstract integration of three economic parameters, rather than simple comparison of explicit potential loss and
gain with equal probability. We further observed that potential loss
modulated connectivity across a network largely representing a subset of the effective connectivity for EV, centered around the
bilateral anteriorlateral prefrontal cortex and the right ventromedial
prefrontal cortex (Fig. 11). This nding substantiates the view that
the brain invests far more resources in the encoding of loss than potential gains, likely because in natural competitive games potential
losses may prove fatal even when presented in the context of overall
risk-advantageous choice options (Kahneman and Tversky, 1979;
Tversky and Kahneman, 1992; Wilkinson, 2008). An important issue
with the design of the present study is that EV was strongly coupled
with choice, and the observed connectivity modulations may therefore reect deliberative behavior rather than primarily value
2203
encoding. We addressed this issue by performing an additional

analysis in which choice-related effects and interactions were
removed prior to inferring EV-sensitive connections. While the
corresponding F-values were diffusely reduced, likely also as a consequence of removing a substantial part of EV-related variance that also
correlated with choice, the overall connectivity pattern was signicantly preserved (Fig. 9). To further explore the relationship between
choice and EV representation, we also re-analyzed the data separating
the accepted and rejected gambles, revealing that, albeit weaker, the
connectivity pattern inferred with the overall PPI analysis is present
over rejected gambles but clearly absent for accepted gambles. On
one hand, the presence of EV-modulated connections in rejected
gambles conrms that our initial, overall PPI inferences were not
purely incidental to choice-related neural differences between
accepted and rejected gambles. On the other, the absence of such
effects in accepted gambles highlights a fundamental difference in
value processing, which could speculatively be interpreted as
evidence of deeper value encoding under risk-disadvantageous conditions. Due to the coupling between EV and choice, this effect cannot
be fully resolved on the basis of our data alone. Further experiments
should be conducted requesting participants to determine riskadvantageousness mentally without making any decision, and loosening the coupling between EV and choice for example by means of
introducing trials where participants are requested to decide in a
highly risk-averse or risk-prone manner (i.e., generally boosting or
depleting the proportion of accepted gambles irrespective of EV).
Conclusion
This study provides the rst comprehensive evaluation of effective
connectivity during risky decision-making based on a networkdiscovery approach. Our ndings reveal that expected value modulates
effective connectivity across a network that includes and extends
substantially beyond those nodes that express direct relationships
between value and activity. This network is characterized by smallworld topology, unitary architecture and prominently features medial
and anteriorlateral prefrontal areas as key hubs, characterized by
dense connectivity and high betweenness centrality. The differential
expression of univariate task effects across these hubs suggests that
the ventro-medial prefrontal cortex may particularly represent a processing hub, whereas the anteriorlateral prefrontal cortex may subserve primarily communication functions. Our ndings thus extend
existing literature by robustly demonstrating that the representation
of value is dependent on distributed processing taking place across a
widespread network, feeding information into a restricted set of integrative prefrontal regions. Additionally, dichotomizing the gambles
into risk advantageous or disadvantageous revealed that the operation
underlying the observed connectivity is more pertinent to comparison
of weighted gains and losses than to summative determination of
continuous value. Further analyses revealed that, albeit smaller, an
effect of continuous value was present across rejected gambles, but
clearly absent for accepted gambles. Our ndings also demonstrate
that potential loss is signicantly over-represented in comparison
with potential gain and outcome probability at the level of regional
activity as well as in terms of effective connectivity. Since mapping
effective connectivity revealed strongly signicant effects involving
regions for which univariate effects were clearly absent, the results of
the present study have broad paradigmatic implications for neuroeconomics in general, demonstrating the value of multivariate analysis in
obtaining a comprehensive account of the neural substrates of
decisional processes.
Acknowledgments
LM was wholly funded and employed by the Fondazione IRCCS
Istituto Neurologico Carlo Besta (FINCB) during the core period of
2204
this study. This research also forms part of a BSMS doctoral dissertation (10/056/MIN) and is based on preparatory work conducted
while LM was previously employed by BSMS. Data acquisition was
conducted at and supported by the FINCB. The authors are grateful
to Francesca Epifani for excellent operational assistance during data
acquisition and to Sylvie Piacentini for useful discussions. The results
of this study have in part been presented at the Mapping Functional
Networks for Brain Surgery ISMRM workshop (Milan, IT, Sep 2011).
AKS and HC gratefully acknowledge the support of the Dr. Mortimer
and Theresa Sackler Foundation. AKS is also supported by the EPSRC
fellowship EP/G007543/1. The authors are grateful to an anonymous
reviewer for insightful feedback on the interpretation of network
analysis results.
References
Abdellaoui, M., Bleichrodt, H., Paraschiv, C., 2007. Loss aversion under prospect theory:
a parameter-free measurement. J. Manage. Sci. 53, 16591674.
Alexander, W.H., Brown, J.W., 2011. Medial prefrontal cortex as an action-outcome
predictor. Nat. Neurosci. 14 (10), 13381344.
Andersen, R.A., Cui, H., 2009. Intention, action planning, and decision making in
parietalfrontal circuits. Neuron 63, 568583.
Badre, D., 2008. Cognitive control, hierarchy, and the rostro-caudal organization of the
frontal lobes. Trends Cogn. Sci. 12 (5), 193200.
Barahona, M., Pecora, L.M., 2002. Synchronization in small-world systems. Phys. Rev.
Lett. 89, 054101.
Barbas, H., 2000. Connections underlying the synthesis of cognition, memory, and
emotion in primate prefrontal cortices. Brain Res. Bull. 52, 319330.
Barbey, A.K., Grafman, J., 2011. An integrative cognitive neuroscience theory of social
reasoning and moral judgment. WIREs Cogn. Sci. 2, 5567.
Barbey, A.K., Krueger, F., Grafman, J., 2009. An evolutionarily adaptive neural architecture
for social reasoning. Trends Neurosci. 32 (12), 603610.
Barrat, A., Barthelemy, M., Vespignani, A., 2008. Dynamical Processes on Complex
Networks. Cambridge University Press, Cambridge, UK.
Bassett, D.S., Bullmore, E., 2006. Small-world brain networks. Neuroscientist 12, 512523.
Bastian, M., Heymann, S., Jacomy, M., 2009. Gephi: an open source software for exploring and manipulating networks. Int. AAAI Conf. Weblogs and Social Media.
Bechara, A., Damasio, H., Damasio, A.R., Lee, G.P., 1999. Different contributions of the
human amygdala and ventromedial prefrontal cortex to decision-making. J.
Neurosci. 19, 54735481.
Blondel, V.D., Guillaume, J.L., Lambiotte, R., Lefebvre, E., 2008. Fast unfolding of communities in large networks. J. Stat. Mech. 10, 1000.
Botvinick, M.M., 2008. Hierarchical models of behavior and prefrontal function. Trends
Cogn. Sci. 12 (5), 201208.
Brandes, U., 2001. A faster algorithm for betweenness centrality. J. Math. Soc. 25,
163177.
Bressler, S.L., Seth, A.K., 2011. WienerGranger causality: a well established methodology.
NeuroImage 58 (2), 323329.
Brog, J.S., Salyapongse, A., Deutch, A.Y., Zahm, D.S., 1993. The patterns of afferent
innervation of the core and shell in the accumbens part of the rat ventral striatum: immunohistochemical detection of retrogradely transported uoro-gold. J.
Comp. Neurol. 338 (2), 255278.
Buckner, R.L., Sepulcre, J., Talukdar, T., Krienen, F.M., Liu, H., Hedden, T., Andrews-Hanna,
J.R., Sperling, R.A., Johnson, K.A., 2009. Cortical hubs revealed by intrinsic functional
connectivity: mapping, assessment of stability, and relation to Alzheimer's disease.
J. Neurosci. 29, 18601873.
Bullmore, E.T., Bassett, D.S., 2011. Brain graphs: graphical models of the human brain
connectome. Annu. Rev. Clin. Psychol. 7, 113140.
Camara, E., Rodriguez-Fornells, A., Mnte, T.F., 2008. Functional connectivity of reward
processing in the brain. Front. Hum. Neurosci. 2, 19.
Camara, E., Rodriguez-Fornells, A., Ye, Z., Mnte, T.F., 2009. Reward networks in the
brain as captured by connectivity measures. Front Neurosci. 3 (3), 350362 Dec 15.
Chib, V.S., Rangel, A., Shimojo, S., O'Doherty, J.P., 2009. Evidence for a common representation of decision values for dissimilar goods in human ventromedial prefrontal
cortex. J. Neurosci. 29, 1231512320.
Clark, L., 2010. Decision-making during gambling: an integration of cognitive and
psychobiological approaches. Philos. Trans. R. Soc. Lond. B Biol. Sci. 365, 319330.
Clark, L., Manes, F., Antoun, N., Sahakian, B.J., Robbins, T.W., 2003. The contributions of
lesion laterality and lesion volume to decision-making impairment following
frontal lobe damage. Neuropsychologia 41, 14741483.
Clark, L., Bechara, A., Damasio, H., Aitken, M.R., Sahakian, B.J., Robbins, T.W., 2008.
Differential effects of insular and ventromedial prefrontal cortex lesions on risky
decision-making. Brain 131, 13111322.
Critchley, H.D., 2005. Neural mechanisms of autonomic, affective, and cognitive
integration. J. Comp. Neurol. 493, 154166.
De Martino, B., Camerer, C.F., Adolphs, R., 2010. Amygdala damage eliminates
monetary loss aversion. Proc. Natl. Acad. Sci. U.S.A. 107 (8), 37883792.
Duann, J.R., Jung, T.P., Kuo, W.J., Yeh, T.C., Makeig, S., Hsieh, J.C., Sejnowski, T.J., 2002.
Single-trial variability in event-related BOLD signals. NeuroImage 15 (4), 823835.
Friston, K.J., Buechel, C., Fink, G.R., Morris, J., Rolls, E., Dolan, R.J., 1997. Psychophysiological and modulatory interactions in neuroimaging. NeuroImage 6, 218229.
Friston, K.J., Harrison, L., Penny, W., 2003. Dynamic causal modelling. NeuroImage 19,
12731302.
Fuster, J.M., 1997. The prefrontal cortex. Anatomy, Physiology, and Neuropsychology of
the Frontal Lobe. Lippincott-Raven.
Ginestet, C.E., Nichols, T.E., Bullmore, E.T., Simmons, A., 2011. Brain network analysis:
separating cost from topology using cost-integration. PLoS One 6 (7), e21570.
Glscher, J., Hampton, A.N., O'Doherty, J.P., 2009. Determining a role for ventromedial
prefrontal cortex in encoding action-based value signals during reward-related
decision making. Cereb. Cortex. 19, 483495.
Gneezy, U., Rustichini, A., 2000. Pay enough or don't pay at all. Q. J. Econ. 115, 791810.
Guimer, R., Mossa, S., Turtschi, A., Amaral, L.A., 2005. The worldwide air transportation
network: anomalous centrality, community structure, and cities' global roles. Proc.
Natl. Acad. Sci. U.S.A. 102 (22), 77947799.
Hagmann, P., Cammoun, L., Gigandet, X., Meuli, R., Honey, C.J., Wedeen, V.J., Sporns, O.,
2008. Mapping the structural core of the human cerebral cortex. PLoS Biol. 6, e159.
Hare, T.A., Camerer, C.F., Knoepe, D.T., Rangel, A., 2010. Value computations in ventral
medial prefrontal cortex during charitable decision making incorporate input from
regions involved in social cognition. J. Neurosci. 30, 583590.
Holm, S., 1979. A simple sequentially rejective multiple test procedure. Scand. J. Stat. 6,
6570.
Hsu, M., Krajbich, I., Zhao, C., Camerer, C.F., 2009. Neural response to reward anticipation
under risk is nonlinear in probabilities. J. Neurosci. 29 (7), 22312237.
Humphries, M.D., Gurney, K., 2008. Network small-world-ness: a quantitative method
for determining canonical network equivalence. PLoS One 3, e0002051.
Kable, J.W., Glimcher, P.W., 2009. The neurobiology of decision: consensus and controversy. Neuron 63, 733745.
Kahneman, D., Tversky, A., 1979. Prospect theory: an analysis of decision under risk.
Econometrica 4, 263291.
Latapy, M., 2008. Main-memory triangle computations for very large (sparse power
law) graphs. Theor. Comp. Sci. 407, 458473.
Levine, D.S., 2009. Brain pathways for cognitiveemotional decision making in the
human animal. Neural Netw. 22, 286293.
Machina, M.J., 1982. Expected utility analysis without the independence axiom.
Econometrica 50, 277323.
Manes, F., Sahakian, B., Clark, L., Rogers, R., Antoun, N., Aitken, M., Robbins, T., 2002.
Decision-making processes following damage to the prefrontal cortex. Brain 125,
624639.
Mesulam, M.M., 1998. From sensation to cognition. Brain 121, 10131052.
Meyer, K., Damasio, A., 2009. Convergence and divergence in a neural architecture for
recognition and memory. Trends Neurosci. 32, 376382.
Minati, L., Piacentini, S., Ferr, F., Nanetti, L., Romito, L., Mariotti, C., Grisoli, M., Medford,
N., Critchley, H.D., Albanese, A., 2011. Choice-option evaluation is preserved in
early Huntington and Parkinson's disease. NeuroReport 22, 753757.
Minati, L., Grisoli, M., Franceschetti, S., Epifani, F., Granvillano, A., Medford, N., Harrison,
N.A., Piacentini, S., Critchley, H.D., 2012. Neural signatures of economic parameters
during decision-making: a functional MRI (FMRI), electroencephalography (EEG)
and autonomic monitoring study. Brain Topogr. 25, 7396.
Newman, M., 2010. Networks: An Introduction. Oxford University Press, Oxford, UK.
Niewenhuys, R., 1996. The Human Central Nervous System: A Synopsis and Atlas.
Springer-Verlag, New York.
Oldeld, R.C., 1971. The assessment and analysis of handedness: the Edinburgh inventory. Neuropsychologia 9, 97113.
Ongr, D., Price, J.L., 2000. The organization of networks within the orbital and medial
prefrontal cortex of rats, monkeys and humans. Cereb Cortex. 10 (3), 206219.
Ongr, D., Ferry, A.T., Price, J.L., 2003. Architectonic subdivision of the human orbital
and medial prefrontal cortex. J. Comp. Neurol. 460 (3), 425449.
Philiastides, M.G., Biele, G., Heekeren, H.R., 2010. A mechanistic account of value computation in the human brain. Proc. Natl. Acad. Sci. U.S.A. 107, 94309435.
Park, S.Q., Kahnt, T., Rieskamp, J., Heekeren, H.R., 2011. Neurobiology of value integration:
when value impacts valuation. J. Neurosci. 31, 93079314.
Platt, M.L., Huettel, S.A., 2008. Risky business: the neuroeconomics of decision making
under uncertainty. Nat. Neurosci. 11, 398403.
Rajkowska, G., Goldman-Rakic, P.S., 1995. Cytoarchitectonic denition of prefrontal
areas in the normal human cortex: II. Variability in locations of areas 9 and 46
and relationship to the Talairach coordinate system. Cereb Cortex 5, 323337.
Ramsey, J.D., Hanson, S.J., Hanson, C., Halchenko, Y.O., Poldrack, R.A., Glymour, C., 2010.
Six problems for causal inference from fMRI. NeuroImage 49 (2), 15451558.
Rangel, A., Hare, T., 2010. Neural computations associated with goal-directed choice.
Curr. Opin. Neurobiol. 20, 262270.
Rushworth, M.F., Behrens, T.E., 2008. Choice, uncertainty and value in prefrontal and
cingulate cortex. Nat. Neurosci. 11, 389397.
Sakagami, M., Pan, X., 2007. Functional role of the ventrolateral prefrontal cortex in
decision making. Curr. Opin. Neurobiol. 17 (2), 228233.
Salvador, R., Suckling, J., Coleman, M.R., Pickard, J.D., Menon, D., Bullmore, E., 2005.
Neurophysiological architecture of functional magnetic resonance images of
human brain. Cereb Cortex. 15, 13321342.
Serences, J.T., 2004. A comparison of methods for characterizing the event-related
BOLD time series in rapid fMRI. NeuroImage 21 (4), 16901700.
Serences, J.T., 2008. Value-based modulations in human visual cortex. Neuron 60,
11691181.
Smith, D.V., Hayden, B.Y., Truong, T.K., Song, A.W., Platt, M.L., Huettel, S.A., 2010.
Distinct value signals in anterior and posterior ventromedial prefrontal cortex. J.
Neurosci. 30, 24902495.
Sporns, O., 2009. Networks of the Brain. The MIT Press, Cambridge MA, USA.
Sporns, O., Tononi, G.M., 2001. Classes of network connectivity and dynamics.
Complexity 7, 2838.

Sporns, O., Tononi, G., Edelman, G.M., 2000. Theoretical neuroanatomy: relating
anatomical and functional connectivity in graphs and cortical connection matrices.
Cereb Cortex 10, 127141.
Sporns, O., Chialvo, D.R., Kaiser, M., Hilgetag, C.C., 2004. Organization, development and
function of complex brain networks. Trends Cogn. Sci. 8, 418425.
Stephan, K.E., 2004. On the role of general system theory for functional neuroimaging.
J. Anat. 205, 443470.
Stephan, K.E., Harrison, L.M., Kiebel, S.J., David, O., Penny, W.D., Friston, K.J., 2007.
Dynamic causal models of neural system dynamics: current state and future
extensions. J. Biosci. 32, 129144.
Stephan, K.E., Penny, W.D., Daunizeau, J., Moran, R.J., Friston, K.J., 2009. Bayesian model
selection for group studies. NeuroImage 46, 10041017.
Tom, S.M., Fox, C.R., Trepel, C., Poldrack, R.A., 2007. The neural basis of loss aversion in
decision-making under risk. Science 315, 515518.
Trepel, C., Fox, C.R., Poldrack, R.A., 2005. Prospect theory on the brain? Toward a
cognitive neuroscience of decision under risk. Cogn. Brain Res. 23, 3450.
Tversky, A., Kahneman, D., 1992. Advances in prospect theory cumulative representation of uncertainty. J Risk Uncertain 5, 297323.
Tzourio-Mazoyer, N., Landeau, B., Papathanassiou, D., Crivello, F., Etard, O., Delcroix, N.,
Mazoyer, B., Joliot, M., 2002. Automated anatomical labeling of activations in SPM
using a macroscopic anatomical parcellation of the MNI MRI single-subject brain.
NeuroImage 15, 273289.
Valdes-Sosa, P.A., Roebroeck, A., Daunizeau, J., Friston, K., 2011. Effective connectivity:
inuence, causality and biophysical modeling. NeuroImage 58 (2), 339361.
Venkatraman, V., Payne, J.W., Bettman, J.R., Luce, M.F., Huettel, S.A., 2009. Separate
neural mechanisms underlie choices and strategic preferences in risky decision
making. Neuron 62, 593602.
Vogt, B.A., 2005. Pain and emotion interactions in subregions of the cingulate gyrus.
Nat. Rev. Neurosci. 6, 533544.
2205
von Neumann, J., Morgenstern, O., 1944. Theory of Games and Economic Behavior.
Princeton, USA, Princeton UP.
Vuilleumier, P., Richardson, M.P., Armony, J.L., Driver, J., Dolan, R.J., 2004. Distant
inuences of amygdala lesion on visual cortical activation during emotional face
processing. Nat. Neurosci. 7, 12711278.
Wager, T.D., Vazquez, A., Hernandez, L., Noll, D.C., 2005. Accounting for nonlinear BOLD
effects in fMRI: parameter estimates and a model for prediction in rapid
event-related studies. NeuroImage 25 (1), 206218.
Wallis, J.D., Kennerley, S.W., 2010. Heterogeneous reward signals in prefrontal cortex.
Curr. Opin. Neurobiol. 20, 191198.
Wang, X.J., 2008. Decision making in recurrent neuronal circuits. Neuron 60, 215234.
Wang, J., Zuo, X., He, Y., 2010. Graph-based network analysis of resting-state functional
MRI. Front. Syst. Neurosci. 4, 16.
Watts, D.J., Strogatz, S.H., 1998. Collective dynamics of small-world networks. Nature
393, 440442.
Weller, J.A., Levin, I.P., Shiv, B., Bechara, A., 2009. The effects of insula damage on
decision-making for risky gains and losses. Soc. Neurosci. 4, 347358.
Wendt, J., Weike, A.I., Lotze, M., Hamm, A.O., 2011. The functional connectivity
between amygdala and extrastriate visual cortex activity during emotional picture
processing depends on stimulus novelty. Biol. Psychol. 86, 203209.
Wilkinson, N., 2008. An Introduction to Behavioural Economics. Palgrave-Macmillan,
Basingstoke, UK.
Wunderlich, K., Rangel, A., O'Doherty, J.P., 2009. Neural computations underlying
action-based decision making in the human brain. Proc Natl Acad Sci USA 106,
1719917204.
Zamora-Lpez, G., Zhou, C., Kurths, J., 2010. Cortical hubs form a module for multisensory
integration on top of the hierarchy of cortical networks. Front Neuroinform. 4, 1.
Zemanov, L., Zhou, C.S., Kurths, J., 2006. Structural and functional clusters of complex
brain networks. Phy. D 224, 202212.

Decision Making

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Decision Making

Hochgeladen von

Copyright:

Verfügbare Formate

NeuroImage 60 (2012) 21912205

Contents lists available at SciVerse ScienceDirect

Full Length Article

Decision-making under risk: A graph-based network analysis using functional MRI

choose among such mixed gambles, or decide whether or not to

L. Minati et al. / NeuroImage 60 (2012) 21912205

determined by the EVs of all options available in a given situation

L. Minati et al. / NeuroImage 60 (2012) 21912205

L. Minati et al. / NeuroImage 60 (2012) 21912205

prefrontal cortices (VMPFC and DMPFC), orbitofrontal cortex (OFC),

Behavioral response analysis

We next applied PPI analysis to search for connections between pair

L. Minati et al. / NeuroImage 60 (2012) 21912205

efferent connections) were computed (e.g., Bullmore and Bassett,

was observed. We therefore tested by means of a one-sample t-test

L. Minati et al. / NeuroImage 60 (2012) 21912205

using algorithms described in detail elsewhere (Blondel et al.,

Dynamic causal modeling (DCM)

analysis, which yielded group-level model exceedance probabilities

L. Minati et al. / NeuroImage 60 (2012) 21912205

world network. Accordingly, the average clustering coefcient was

L. Minati et al. / NeuroImage 60 (2012) 21912205

perspective, detailed analysis of the processing of simple, individual

L. Minati et al. / NeuroImage 60 (2012) 21912205

demonstrate that signicant deviations from locally optimal behavior

L. Minati et al. / NeuroImage 60 (2012) 21912205

Our network-discovery analysis for effective connectivity revealed

(Humphries and Gurney, 2008). Our nding of small-world topology

L. Minati et al. / NeuroImage 60 (2012) 21912205

and outdegree), its position in the directed network (betweenness

L. Minati et al. / NeuroImage 60 (2012) 21912205

correlation with EV due to their direct computational engagement,

expected value. We provide the additional insight that the anterior

L. Minati et al. / NeuroImage 60 (2012) 21912205

encoding. We addressed this issue by performing an additional

L. Minati et al. / NeuroImage 60 (2012) 21912205

L. Minati et al. / NeuroImage 60 (2012) 21912205

Das könnte Ihnen auch gefallen