Predicting Surgery Targets in Temporal Lobe Epilepsy Through Structural Connectome Based Simulations

RESEARCH ARTICLE
Predicting Surgery Targets in Temporal Lobe

Epilepsy through Structural Connectome
Based Simulations
Frances Hutchings1*, Cheol E. Han2,3, Simon S. Keller4, Bernd Weber5,6
Peter N. Taylor1,7‡, Marcus Kaiser1,7‡
1 Interdisciplinary Computing and Complex BioSystems, School of Computing Science, Newcastle
University, Newcastle upon Tyne, United Kingdom, 2 Department of Biomedical Engineering, Korea
University, Seoul, Republic of Korea, 3 Department of Brain Cognitive Sciences, Seoul National University,
a11111 Seoul, Republic of Korea, 4 Department of Molecular and Clinical Pharmacology, Institute of Translational
Medicine, University of Liverpool, Liverpool, United Kingdom, 5 Center for Economics and Neuroscience,
University of Bonn, Bonn, Germany, 6 Department of Epileptology, University of Bonn, Bonn, Germany,
7 Institute of Neuroscience, Newcastle University, Newcastle upon Tyne, United Kingdom
‡PNT and MK are joint senior authors.

* [email protected]
OPEN ACCESS
Citation: Hutchings F, Han CE, Keller SS, Weber B,

Taylor PN, Kaiser M (2015) Predicting Surgery
Abstract
Targets in Temporal Lobe Epilepsy through Structural Temporal lobe epilepsy (TLE) is a prevalent neurological disorder resulting in disruptive sei-
Connectome Based Simulations. PLoS Comput Biol
11(12): e1004642. doi:10.1371/journal.pcbi.1004642
zures. In the case of drug resistant epilepsy resective surgery is often considered. This is a
procedure hampered by unpredictable success rates, with many patients continuing to
Editor: Olaf Sporns, Indiana University, UNITED
STATES
have seizures even after surgery. In this study we apply a computational model of epilepsy
to patient specific structural connectivity derived from diffusion tensor imaging (DTI) of 22
Received: June 24, 2015
individuals with left TLE and 39 healthy controls. We validate the model by examining
Accepted: October 29, 2015 patient-control differences in simulated seizure onset time and network location. We then
Published: December 10, 2015 investigate the potential of the model for surgery prediction by performing in silico surgical
Copyright: © 2015 Hutchings et al. This is an open resections, removing nodes from patient networks and comparing seizure likelihood post-
access article distributed under the terms of the surgery to pre-surgery simulations. We find that, first, patients tend to transit from non-epi-
Creative Commons Attribution License, which permits leptic to epileptic states more often than controls in the model. Second, regions in the left
unrestricted use, distribution, and reproduction in any
medium, provided the original author and source are
hemisphere (particularly within temporal and subcortical regions) that are known to be
credited. involved in TLE are the most frequent starting points for seizures in patients in the model. In
Data Availability Statement: The data underlying
addition, our analysis also implicates regions in the contralateral and frontal locations which
the figures and conclusions of the manuscript are may play a role in seizure spreading or surgery resistance. Finally, the model predicts that
now available for public access at the Zenodo patient-specific surgery (resection areas chosen on an individual, model-prompted, basis
repository, DOI: 10.5281/zenodo.33639, URL: https://
and not following a predefined procedure) may lead to better outcomes than the currently
zenodo.org/record/33639#.VkN-WivY_l8.
used routine clinical procedure. Taken together this work provides a first step towards
Funding: PNT and MK were supported by the
patient specific computational modelling of epilepsy surgery in order to inform treatment
Engineering and Physical Sciences Research
Council of the United Kingdom (EP/K026992/1) as strategies in individuals.
part of the Human Brain Development Project (http://
www.greenbrainproject.org/). SSK was supported by
the Medical Research Council New Investigator
Research Grant (Grant Number: MR/K023152/1).
CEH was supported by National Research
PLOS Computational Biology | DOI:10.1371/journal.pcbi.1004642 December 10, 2015 1 / 24

Predicting Epilepsy Surgery Targets with Connectome Based Simulations
Foundation of South Korea funded by the Ministry of

Education, Science and Technology (R32-10142, Author Summary
MSIP-2010-0028631, MSIP-2013-004157, MSIP-
R1A1A1008173). BW was supported by a Temporal lobe epilepsy (TLE) is a disorder characterised by unpredictable seizures, where
Heisenberg Grant of the German Research Council surgical removal of brain tissue is often the final treatment option. In roughly 30% of cases
(DFG; We 4427/3-2). FH acknowledges support from surgery procedures are unsuccessful at preventing future seizures. This paper shows the
the Engineering and Physical Sciences Research application of a computational model which uses patient derived brain connectivity to pre-
Council of the United Kingdom. The funders had no
dict the success rates of surgery in people with TLE. We consider the brains of 22 patients
role in study design, data collection and analysis,
decision to publish, or preparation of the manuscript. as networks, with brain regions as nodes and the white matter connections between them
as edges. The brain network is unique to each subject and produced from brain imaging
Competing Interests: The authors have declared
scans of 22 patients and 39 controls. Seizures are simulated before and after surgery, where
that no competing interests exist.
surgery in the model is the removal of nodes from the network. The model successfully
identifies regions known to be involved in TLE, and its predicted success rates for surgery
are close to the results found in reality. The model additionally provides patient specific
recommendations for surgical procedures, which in simulations show improved results
compared to standard surgery in every case. This is a first step towards designing personal-
ised surgery procedures in order to improve surgery success rates.
Introduction
Epilepsy is a spectrum of disorders characterised by recurrent seizures originating in the brain.
Epileptic seizures are defined as spontaneous occurrences of signs and symptoms due to abnor-
mal neuronal activity in the brain [1, 2]. Roughly 50 million people suffer from epilepsy [3]
and in many cases it can be exceedingly debilitating. There are many types of epilepsy, classi-
fied by where in the brain seizures start or what the root cause is thought to be [4]. In this study
we focus on the most common medically intractable form of epilepsy, temporal lobe epilepsy
(TLE), characterised by a supposed focal origin of seizures in the temporal lobe.
Anti-convulsant drugs are the main method of treatment for epilepsy, however in about
30% of cases they are unsuccessful at preventing seizures. When drug therapies fail, surgery is
an alternative treatment option for many patients. In TLE this involves the resection of regions
of the temporal lobe assumed to be the source of the epileptic activity. Short term results show
53%-84% of patients achieving seizure freedom following surgery [5], however post-surgery
longitudinal studies report only around 47%—65% of patients become seizure free following
the resection of focal areas [6, 7].
Atrophy of focal brain areas in and around the temporal lobe is frequently found in people
with TLE [8–14] using magnetic resonance imaging (MRI). However, recent suggestions are
that TLE may involve areas far beyond the temoral lobe, forming a suggested epileptogenic net-
work [15]. Diffusion weighted MRI (DW-MRI) allows for the inference of hypothesised white
matter fibre tracts which connect brain areas [16, 17]. This can be used to create a patient spe-
cific network of structural brain connectivity between the aforementioned brain areas [18–23].
Epilepsy is a disorder which involves propagation of high amplitude oscillatory dynamics
over large areas of brain tissue during seizures. This means that modelling a disorder such as
epilepsy ideally requires a large scale model of the brain, with realistic connections through
which the activity can spread. To this end the brain can be modelled as a network, with struc-
turally distinct regions abstracted to single nodes and the connections between these discrete
regions mapped through imaging techniques such as DW-MRI. Using such simplified brain
networks, seizure initiation and spreading can be modelled using connectivity specific to indi-
vidual patients [2, 24–26].

Computational modelling of epilepsy has been accomplished in various forms, though very
few have used patient specific whole brain network data [27]. Most commonly neural mass
models—simulated populations of connected neurons—have been used to produce or explore
models of epileptogenesis [28–37]. Various mechanisms for seizure initiation have been pro-
posed, and one of the most commonly used is bistability (see [2, 38] for reviews). This mecha-
nism requires the system to have two stable states which coexist, a resting state (e.g. fixed
point) and a state with seizure-like dynamics, (e.g. a high amplitude limit cycle). While in a
parameter region where both these states are potentially available, the system is set to receive a
noisy input such that when the noise passes a certain threshold the system transitions from one
state to the other. A clear example of this in the context of epilepsy is the model described by
Kalitzin et al. [39] and refined by Benjamin et al. [40], where one parameter controls the attrac-
tion of the system to the two stable states. This has been applied to network models to investi-
gate the spread and likelihood of seizures dependent on connectivity [40, 41]. However, those
studies did not use structural connectivity derived from DTI.
We aim to develop a framework for predicting surgical outcomes in patients following
resective epilepsy surgery using a bistable model and DTI data obtained from patients with epi-
lepsy. Following previous reported surgical success rates of around 50%-80% [5–7] we
hypothesise that our model predictions shall also be in this range. Furthermore, we hypothesise
that subject-specific surgery will give greater seizure reduction than that of standard resection
procedures in our simulations.
Results
As detailed in the modelling section, we applied a mathematical model to DW-MRI acquired
patient specific connectivity networks and observed the results of simulations of regions of
interest (ROI), or ‘node’, activity over time. Each ROI can transition (‘escape’) into a seizure
state with a probability influenced by the region’s surface area, connectivity, and fluctuations in
background noise. Escape times are taken to be representative of seizure likelihood, as a fast
escape time means there would be more seizures on average for that node than for one with a
slower escape time, i.e. a greater likelihood of a seizure occurring. Our analysis includes the
simulations of 39 controls and 22 patients with left TLE.
Model validation
In reality controls have a significantly lower likelihood of having seizures than patients. Fig 1
shows this, on average, to be the case in the model. A comparison of the times taken for the
first three ROIs to escape from the resting state into the pseudo-seizure state for patients versus
controls is shown in the figure, with a clear difference in distribution evident. The time taken
for three nodes to escape was taken to be the seizure start time (‘escape time’) as we considered
this to be evidence of the beginning of spreading of the seizure activity. The simulation was
repeated 100 times with different noise seeds on each iteration, and the escape times were aver-
aged over the 100 runs. These results are significant (p < 0.001) checked using a Mann-Whit-
ney U test. These results also hold true for the first and second escape (S1 Fig).
Our second validation of the model examines where the initial escaping nodes are situated
in the brain. In left TLE a seizure origin in the left temporal lobe would be expected for the
model to match clinical observations. Fig 2a) shows the top ten escaping nodes, found by tak-
ing the mode for patient and control populations, in blue and red respectively. The five green
nodes indicate locations in the top ten of both populations. S2 Table shows the top three nodes
for every subject, from which the overall top ten were found, separately for patients and con-
trols. While the red control escape locations are scattered with no discernible pattern, the blue

Fig 1. Patient’s nodes tend to escape to the seizure state faster than control’s. Comparison of patients and controls initial transition times into the
seizure state. A seizure state was deemed to have occurred once three nodes had escaped, so the times here are when the third node hit the seizure state.
Asterisks represent a significant difference found at p < 0.001 using a Mann-Whitney U test. Each dot represents the mean escape time for one individual
subject over 100 iterations, grey bars reach the mean of each subject group, error bars show the standard error of the mean. Higher times indicate that the
individual took longer to reach a seizure state.
doi:10.1371/journal.pcbi.1004642.g001
patient escape locations are concentrated on the left hemisphere of the brain and predomi-
nantly in the temporal (temporal pole) or in subcortical (amygdala, putamen, thalamus) areas.
Furthermore, there was a greater consistency of escaping nodes in patients than controls, with
nodes in the top ten appearing 60.6% of the time. The left hemisphere amygdala was in the top
three in 54% of the patients when taking the modes for each patient over all iterations, and
occurred at least once in each patient top three over the 100 iterations. The consistency was
reduced in controls, where nodes in the top ten appeared 30.7% of the time, with no one node
occurring in more than 50% of controls over 100 iterations. Fig 2b) illustrates the difference
with a bar plot comparison between patients and controls. The model therefore also reproduces
a more stereotyped seizure pattern in patients, than in controls. This is in agreement with pre-
vious studies [25, 42].
Model underpinnings
In order to pinpoint why we see the pattern of escaping nodes found in Fig 2 a number of net-
work measures were tested. In the model the two major influences on node escape times are
surface area and connectivity. Surface area alone does not correlate strongly with the pattern of
fastest escaping nodes (see S3 Table) therefore we looked at the connectivity to see whether
there were any clear markers of seizure prone nodes from connectivity measures. One possibil-
ity was that hub nodes could prove to be more seizure prone, as there are ROIs with a large

Fig 2. Patient-control ten earliest escape locations compared. Part a) shows the locations of the top ten most ‘seizure-prone’ nodes. These were found
by taking the mode over the first three nodes to escape for each subject (split into patient and control groups) for all simulations without resections. The nodes
represent locations in the brain network, and are coloured according to the group they escaped first in—blue for patients, red for controls and green for where
the location was in the top ten to escape for both groups. Smaller grey circles show the locations of other nodes in the network which were not in either top
ten. The main figure has location labels for patient areas, and the inset figure (transverse plane) has a dashed line encircling the main group of nodes for
patients which are clustered in or around the temporal lobe. It is perhaps worth noting that while escape times are not apparent in this image controls nodes
did take longer to escape than patients for all locations. Part b) shows the consistency of the top ten nodes compared for patients and controls, where
patients show greater consistency for all the shared nodes and strikingly the left hemisphere amygdala escaped to the seizure state in every single patient at
least once over 100 iterations.
number of incoming connections and as such may receive large inputs pushing them closer to
the seizure threshold. Additionally the high number of connections means they are more likely
to receive input from a node which has already hit a seizure state and are then liable to be
pushed over the seizure threshold themselves by the pathological input. We predicted that the
nodes which appear as fast escapers for both patients and controls may be natural hubs,
explaining their appearance in Fig 2.
Fig 3 focuses on the five nodes which occurred in the top ten escaping nodes of both patients
and controls. The node measures for the shared five are displayed beneath a map of the distri-
bution of that node measure for all subjects. The results do support the idea that the shared
nodes are liable to be network hubs, showing a low clustering coefficient value but much higher
node strength and degree values than most ROIs. Fig 4 shows the remaining unshared nodes in
the top ten for patients and controls compared using a Mann-Whitney U test for the ROI in
patients against the same region in controls for given network measures. Because these differ
between groups we looked for evidence that there were differences in the node network

Fig 3. Commonly escaping nodes shared between patients and controls are network hubs. Five of the most commonly escaping nodes in patients and
controls occurred in both groups, and different node measures are compared for these nodes here. For each network measure in the figure the distribution for
all nodes in all subjects is shown in the top section of the plots. The dots in the lower section reveal where in this distribution the shared nodes occur for the
given measure. These locations are shown separately for the ROI in patients and controls (coloured green and red respectively) displaying the median value
for the groups. The high node strength, degree and eigenvector centrality values for the shared nodes with a low clustering coefficient support the idea that
these regions may be network hubs.
properties between the groups. However there were no clear trends to indicate that any one
network measure could account for these nodes being in the top ten.
In summary, the rapid escape time of some nodes can be attributed to shared properties of
being network hubs. However, the complex interplay of connectivity, surface area, variations in
streamline length (delays) and nonlinearities in the model, leads to highly nontrivial outcomes
in the simulations.

Fig 4. Commonly escaping nodes which differ between patients and controls compared. As these nodes have different escape time patterns in
patients and controls we checked for network differences between the two groups. Z scores were found using a Mann-Whitney U test comparing network
measures for the unshared nodes between patients and controls. A negative value means that the node score was lower in patients than in controls, and a
positive z score means the node score was higher in patients than in controls. There are no clear differences which implies more complex origins for the
different escape time results.

Fig 5. Simulated surgery results comparisons. For every patient three different methods of resection were used and then the escape times were
compared for each patient pre and post surgery (surgery simulation results compared to results from simulations where no surgery was carried out). Plot (a)
shows the mean percentage improvement in escape time for each patient by resection method, where escape time was taken to be the time by which 3
nodes had hit a seizure state. Grey bars reach the mean of each subject group, error bars show the standard error of the mean. The circled values are the
means for the patient shown in plot (b). A single patient representative example is shown in plot (b), with the percentage improvements found for each
simulation plotted and compared by resection technique. Stars represent the significance level, 1 star for significance at p < 0.05, 2 stars for p < 0.01 and 3
stars for p < 0.001. A Mann-Whitney U test with Bonferroni correction was carried out to find the p-values following significant results found from Kruskal-
Wallis tests.
Model prediction
Here we test the impact of removing 1) randomly selected adjacent nodes, 2) nodes routinely
removed clinically (amygdala, hippocampus and parahippocampal gyrus) and 3) optimised
patient specific nodes. It is worth noting here that the patient specific resections removed the
top three most seizure prone nodes for each individual patient. This did not account for
whether the chosen ROIs would be practical to remove in an actual surgery procedure. Fig 5(a)
shows a comparison of the simulation results when using different resection methods. The per-
centage improvements were found for each resection technique, and the mean improvement
for each patient is plotted in this figure. There is a clear increase in improvement from random
resections to clinical, and then again to the subject specific resections. Asterisks indicate that
the distributions of improvement times are significantly different between techniques. The
simulations were repeated over 100 iterations with differing noise seeds, and comparisons of
the escape times were patient by patient. Subject specific resections removed the top 3 fastest
escaping nodes for each patient. (S2 Table shows the top 3 escaping nodes for each patient).
Fig 5(b) shows a comparison of improvements in escape time for different resection meth-
ods for a single subject. In this case the simulated clinical resection was successful, producing
significant improvements. The subject specific resection was even more successful, significantly
reducing seizure likelihood compared to unresected and also the clinical resections. When
looking directly at escape times the difference between the unresected and random resection

Table 1. Percentage of subjects with significant improvements following surgery. A comparison of the
model generated success rate with reports in literature. In studies there have not been any patient specific
procedures reported to compare, so these reports only include widely used clinical procedures including the
amygdalohippocampectomy. The wide ranges in some reports are due to studies grouping patients by differ-
ent criteria such as lateralisation of atrophy, or in other cases due to longitudinal studies measuring improve-
ments at several different time points after the surgery. There is a general trend of decreased improvement
the longer the time period after surgery, reducing the reported successes.
Source Clinical (%) Patient Specific (%)

Model simulations 72.7 100
Spencer et al. 2008 [5] 53–84 -
Hemb et al. 2013 [6] 62–65 -
de Tisi et al. 2011 [7] 52–47 -
Arruda et al. 1996 [43] 50–93.6 -
doi:10.1371/journal.pcbi.1004642.t001
simulations was insignificant at p < 0.05, both for the patient singled out in Fig 5(b) and when
comparing all patient simulations together.
Finally Table 1 compares the percentage of patients who show significant improvements in
our model following resection of clinically removed areas with established literature on surgery
success. Simulation results were found using a paired t-test comparing before and after escape
times for each patient over all iterations, with a significance level of p < 0.05. The removal of
clinically resected regions in our model produced significant improvements in 72.7% of
patients, and the subject specific resections showed improvements in 100% of cases (also
included in the table). There is a wide range of reported success rates in literature, partially due
to the time scales of studies (ranging from 1 year after surgery to 18) and due to different defi-
nitions of significant improvement. Despite the variability the model results fit squarely in the
middle of the reported range, while the subject specific results show significant improvements
in every case.
Discussion
In this work we have presented three key findings. First, we have demonstrated how the combi-
nation of subject specific data and a nonlinear computational model can lead to successful
identification of TLE associated nodes in patients. Control subjects show few similarities in
identified nodes indicating clear differences in network organisation (Fig 2). Secondly, we have
shown in the model, in agreement with with clinical observations [6], that selective amygdalo-
hippocampectomy can result in reduced seizure likelihood in 72.7% of patients, which is within
the range reported in the literature for patients with similar conditions (Fig 5 and Table 1).
Third, we suggest that patient-specific operations (where different regions are selected for dif-
ferent patients) will result in a better outcome. Furthermore, to our knowledge, this is the first
study to combine computational modelling of brain dynamics with diffusion MRI based con-
nectivity data from patients with temporal lobe epilepsy.
Model validation
The model validation Figs 1 and 2 demonstrate results in line with clinical observations—a
higher propensity for seizures in patients and the spatial origin of seizure activity in the ipsilat-
eral temporal lobe. Indeed, validation of any model should always be the crucial first step. One
aspect the model does not accurately capture is that healthy controls still have seizures in the
model. Healthy controls can have a seizure—they have a propensity to seize, however, this

would typically be a result of an external factor such as trauma and therefore may be mechanis-
tically different. It is plausible to suggest that there may be other mechanisms and contributors
to seizure transition (besides those included here) which may be important to improving the
model, especially when considering healthy controls.
Another aspect that the model successfully validated was the spatial profile and consistency
(stereopy) of the ‘escaping’ nodes. Several areas located in the ipsilateral temporal lobe and sub-
cortical areas were involved in agreement with experimental / clinical observations [20, 44, 45].
Four areas not in the left hemisphere were involved—the amygdala, insula, thalamus and the
superior frontal region. Contralateral abnormalities have been noted in TLE before [20, 46],
particularly in temporal lobe areas [15]. It has been speculated that this could be indicative of
compensatory mechanisms of some form, as increased contralateral functional connectivity
has been previously noted in epilepsy patients [47]. If there is a structural correlate for this
potential compensatory connectivity increase then a larger number of connections may put
these regions more at risk of seizures in the model, for the same reasons as highly connected
hub nodes are vulnerable. A high number of inbound connections increases the likelihood of
receiving abnormal input from a seizing node, and increases the input received pushing the
node closer to the seizure threshold. The involvement of the amygdala, thalamus and insula in
TLE has long been known [48–50]. Interhemispheric connectivity between the mirrored
regions in both hemispheres may be the cause of the seizure vulnerability of the contralateral
amygdala, thalamus and insula. Abnormal activity from the ipsilateral hemispheric counter-
parts may spread through interhemispheric connections. The superior frontal region involve-
ment is less intuitive, and the lateral orbitofrontal region is also a rapidly escaping node located
in the frontal lobe. Abnormalities in frontal lobe areas have been documented in TLE [51–53],
with frontal regions frequently involved in seizure propagation, especially the orbitofrontal
cortex [54]. This may provide an explanation for the susceptibility of these nodes to seizures in
our model.
Model underpinnings
Network measures were tested for the most frequently escaping nodes in an attempt to distin-
guish further the influences behind the seizure susceptibility. Five of the top ten escaping nodes
are consistent in both patients and controls, and were found to exhibit properties of network
hubs, as shown in Fig 3. These were the thalamus (bilaterally), insula (bilaterally) and the con-
tralateral superior frontal area. The thalamus and insula are already known to be highly con-
nected and hub-like in the general population [55]. The high node strength and centrality
measures of these regions make them both good candidates for network hubs [56], and also
more likely to propagate abnormal activity. Having a greater number of incoming connections
puts hubs at greater risk of receiving input from a seizing area, and it follows that they would
also propagate the abnormal activity to a large number of regions they connect to. For the
remaining five nodes that differ for patients and controls seizure susceptibility is less clear.
There is no defining contribution from either topological (see Fig 4) or spatial (e.g. surface
area, fibre tract length—see S3 Table) features, although the lambda parameter derived from
surface area does have an effect on the patient nodes. Previous studies have looked for network
differences in TLE, finding changes in clustering coefficient in some areas, small world-ness
and network efficiency when comparing to controls [19–21], summarised in [27]. However a
recent study into the same dataset we use here found that spatial measures show far greater
changes in TLE than network measures [8]. Whilst it is possible that a feature may have been
missed, it seems that a combination of multiple interacting factors is most likely at play in
determining which ROIs are most seizure prone in the model.

Model prediction
The in silico resection experiments showed improvements for the clinically removed nodes to a
similar extent to results found in reality, with 72% of patients (16 out of 22) showing significant
improvements (p < 0.05) in seizure likelihood after simulated surgery. In this instance statisti-
cally significant improvements were taken to be analogous to successful surgery procedures. In
many surgery cases the patients are not completely seizure free but they do see some reduction
in seizure likelihood or intensity. This is a phenomenon which is reflected in some ways in our
results, with an overall skew of improvement, but not in all cases statistically significant.
The random resection experiments showed that the clinical resection results are not simply
due to the removal of any 3 nodes from the network. There was some improvement for the ran-
dom resections, although this was only to a statistically significant extent in one or two cases
(depending on the nodes removed), much less than for clinical resections.
The personalised resections showed by far the most success at reducing seizure likelihood.
These were based on the top three fastest escaping nodes for each patient. The ipsilateral hippo-
campus did not appear in any of the top escaping node vectors for patients. This is surprising
given the well established phenomenon of hippocampal sclerosis in TLE [57, 58], and the fact
that it is commonly removed in resection surgery. In one of the random node tests the ipsilat-
eral hippocampus was included in the 3 randomly selected regions to be removed, but this did
not show any greater improvement than other random node assortments. The reasons for this
unexpected result are unclear, and possibly due to the difficulty of parcelating subcortical struc-
tures in Freesurfer [59]. Although the outputs were checked visually [8] this may have
impacted results. If the hippocampus had a significant impact on ictal genesis then it would
seem logical that removing it would have a greater impact on seizure likelihood.
As well as the ipsilateral temporal pole, the thalamus and insula both appeared commonly
in the sets of earliest escaping nodes. The insula has been linked to TLE, with studies suggesting
that a number of patients with unsuccessful surgical outcomes may have an ictal focus in the
insula, hence the lack of results from temporal lobe resections [49, 60–62]. The thalamus may
similarly play a role in surgery resistent TLE [63, 64], and has been indicated to have an impor-
tant role in the propagation and generation of TLE seizures [50, 65]. The thalamus has been
found to exhibit abnormalities in TLE [66] with cell atrophy and reduced surface area [67].
However the thalamus is crucial for sensory-motor functions and therefore not an option for
resection. We speculate that the partial disconnection of the ipsilateral thalamus (rather than
the complete resection) may lead to improved clinical results.
Surgery prediction in the context of other works

This is not the first or only attempt to try and accomplish prediction of the affects of neurosur-
gical procedures. Imaging studies have been used to predicted how TLE surgery effects memory
and cognitive ability in individuals [68]. While computer modelling has been applied to mea-
sure the effects of in silico lesions on connectivity networks [69, 70], none have used DTI
derived networks of people with TLE. One study by Honey et al. [71] used computational
modelling of macaque connectivity with Kuramoto oscillators and a neural mass model to
examine how region synchrony and interaction differed after lesions. However, that study did
not look specifically at effects on seizure activity.
Recent modelling work has attempted to better characterise and identify epileptogenic areas
[28, 34] with mention of the potential effects of surgical intervention dependent on tissue char-
acteristics. Accurately identifying regions of seizure initiation is clearly an important consider-
ation in surgery planning and prediction. These studies did not use whole brain connectivity,
but rather a model representing a subset of cortical areas. Another recent work, incorporating

patient derived connectivity, found encouraging results in simulations of stimulation applied

to abate seizures [26]. This is an intriguing alternative to surgery and one which can also bene-
fit from subject-specific modelling work.
A study by Sinha et al [72] is closer to the aims and findings of this current project. Using
ECoG data to derive connectivity from functional synchrony the study applied a very similar
model to the one we have used here, without the incorporated surface area data. The model
predicted areas for surgery removal and tested in silico resections against in silico randomised
resection trials. The predicted regions for removal overlapped with clinically chosen resection
areas (identified by the ECoG testing among other clinical evaluations) in 5 out of 6 cases. The
model also predicted additional removal sites not identified by clinical tests. Those predictions
were not experimentally validated, but this is another strong example of how computational
modelling has the potential to be useful for surgery prediction, not limited to one method of
connectivity data acquisition.
Limitations of the study

The connectivity is assumed here to be a good and accurate representation of the white matter
tracts in the individual’s brains. However, there are restrictions of resolution and noise in this
imaging technique and the connectivity matrices produced may not always be fully accurate
[73]. A key limitation of diffusion inferred networks is that they are bi-directional. This
assumption of bi-directionality may be inaccurate, with one recent study into the Macaque
brain finding a higher incidence of unidirectional links than anticipated [74]. Directionality
has been shown to play a key role in the dynamics of a system and therefore should be consid-
ered a limitation of this and all diffusion based connectivity studies [75]. However the assump-
tion of bi-directionality has also been defended in other studies which claim that the majority
of brain connections do exhibit bi-directionality [76]. There is currently no better alternative
for imaging structural connectivity in vivo, but this is still a potential limitation. One possible
solution to this, which could especially be done in the case of epilepsy patients, would be to
map the directionality through active stimulation [77, 78].
Another limitation is the use of deterministic tractography in this study. Crossing fibres are
not well resolved in deterministic approaches, which can lead to inaccurate representations of
brain connectivity [79].
This is not an optimal protocol, and probabilistic algorithms would likely improve the accu-
racy of the connectivity matrices by resolving the crossing fibre problem [80]. However, there
have been studies which have found that deterministic tractography can succeed at capturing
key features of the connectome and thus can provide useful information [81–84]. Additionally
deterministic tractography can have greater success at characterising long range connections
[85], which are important when attempting to characterise seizure spreading patterns. Deter-
ministic and probabilistic tractography capture graph theory metrics similarly, share common
features [86], and can be remarkably reproducible [87]. That being said, as probabilistic tracto-
graphy has been found to be the superior method in terms of reproducibility and crossing fibre
resolution giving a more reliable connectome [80, 86, 88] it may therefore be beneficial to
rerun simulations using a probabilistic algorithm.
In addition to the innate issues with DTI, there is also the question of how to decide on the
brain atlas to use for parcellating into regions, and the choice of weighting used for edges. The
dataset used here split the brain into 82 different regions. There is no gold standard for choos-
ing the number of regions to use, and this choice can make a difference to the network’s prop-
erties [89]. Comparisons have found that network organizational principles seem to remain
independent of the parcellation scheme, but quantified measures do change [90]. A future

direction would be to replicate this study with an alternative, functionally defined atlas (e.g.
[91]). However, functional parcellations often don’t include subcortical areas and are based on
functional observations in healthy subjects. The possibility of changes in underlying processes
due to epileptic activity may make existing functional schemes inaccurate for patients. Apply-
ing alternative anatomical atlases would also be worthwhile further work. We anticipate,
though it should be tested, that results shown here would be reproducible with other atlases at
similar scales. This prediction is based on comparisons of our results with previous studies [92]
including work simulating epilepsy dynamics with human connectome data using different
atlases (AAL and Desikan) [24–26]. Choosing the weighting of edges is another factor to take
into account. In this instance the dataset weighted edges by the number of streamlines identi-
fied, however other options are available such as fractional anisotropy, which weights edges
according to the degree of anisotropy in voxels between regions, thought to give an idea of fibre
myelination or integrity [93, 94].
Another criticism of the model could be found in the level of abstraction used. Limitations
in the computational resources available inevitably leads to a trade off when abstracting a
model. There may be a danger of losing information from the system which is present in bio-
logical reality, for example meso- and micro-scale connectivity were not modelled here, yet
may be important [95]. As a model to assess the impact of global connectivity on TLE, having
too many other interacting factors would make it more difficult to identify the contribution of
heterogeneous networks and a danger of over-fitting the parameters may occur. However, as a
model for surgery prediction perhaps more biological detail would be useful to improve accu-
racy when looking to match in vivo results more closely. The compromise between simulation
speed and biological detail can be difficult. Even though this model is phenomenological, the
simple co-dependence on region surface area and connectivity makes it computationally feasi-
ble to simulate while producing encouraging results.
Future work and improvements

It is necessary to validate the model prediction of subject specific surgical success with patient
outcomes if the model will ever be useful clinically. Diffusion imaging is not routinely done
and it is extremely difficult to obtain data for a significant number of subjects with follow-up
metadata of surgical outcome. One of the key advantages of our study is that we have a fairly
homogeneous subject group (all patients with left TLE), however, statistical power becomes an
issue with a total sample size of only 22 subjects. We anticipate that in the future a multi-site
study with large cohorts will be required. With that we hope to enable better guidance for epi-
lepsy surgery.
Methods
Dataset
We collected data from 22 individuals with left mesial TLE showing unilateral hippocampal
sclerosis and 39 age-matched controls. Written informed consent was obtained, signed by all
participants, and conformed to local ethics requirements. IRB approval (032/08) was given by
the ethical review board of the medical faculty of Bonn. None of the individuals with mTLE
had undergone any previous neurosurgery. T1 weighted MRI scans and diffusion tensor imag-
ing (DTI) data were obtained using a 3 Tesla scanner, a Siemens MAGNETOM TrioTim syngo
(Erlangen, Germany). The T1 images were obtained using 1mm isovoxel, TR = 2500ms and
TE = 3.5ms. The DTI data used 2mm isovoxel, TR = 10,000ms, TE = 91ms and 64 diffusion
directions, b-factor 1000s mm−2 and 12 b0 images. In both cases FoV was 256mm.

Connectome creation from the acquired data is summarised in Fig 6, while Fig 7 shows
nodes and connections imaged in pseudo-3D space averaged over all subjects, and indicates
how this is then used to generate simulations of activity for each node. Firstly FreeSurfer [96]
was used to obtain surface meshes of gray and white matter boundries from the MRI data, and
to parcellate the brain into regions of interest (ROI) based on the Desikan atlas [97, 98]. There
were 82 ROI included, spanning cortical and subcortical regions (subcortical regions included
the Nucleus accumbens, Amygdala, Caudate, Hippocampus, Pallidum, Putamen and the Thal-
amus). Streamline tractography was obtained from DTI images using the Fiber Assignment by
Continuous Tracking (FACT) algorithm [99] through the Diffusion toolkit along with Track-
Vis [100]. First we performed eddy-correction of the image by applying an affine transform of
each diffusion volume to the b0 volume and rotating b-vectors using FSL toolbox (FSL, http://
www.fmrib.ox.ac.uk/fsl/). After the diffusion tensor and its eigenvector was estimated for every
voxel, we applied a deterministic tractography algorithm [99] initiating a single streamline
from the center of each voxel. Tracking was stopped when the angle change was too large (35
degree of angle threshold) or when tracking reached a voxel with a fractional anisotropy value
of less than 0.2 [84]. For further details see [81].
The centre coordinates of each voxel were the start of a single streamline, the total number
of streamlines never exceeded the number of seed voxels. The number of connecting stream-
lines were used to determine the connectivity matrix (S), as the streamline count has recently
been confirmed to provide a realistic estimate of white matter pathway projection strength
[101]. Distance matrices were also constructed using the mean fibre length of the streamlines
connecting each pair of ROIs. The surface area of each ROI was found using FreeSurfer for cor-
tical regions and for subcortical areas by computing the interface area to the white matter in T1
space [102].
Model
We apply a mathematical model to simulate activity of ROI in a network formed from acquired
DTI data. The network connectivity matrix, M, is determined by the patient specific weighted
connectivity produced in connectome creation as detailed in the previous section. The connec-
tion weights were derived from the streamline counts between ROIs and were normalised as
shown in Eq (2), where S is the matrix of streamline counts for an individual subject.
logðSÞ
M¼ ð1Þ
maxðlogðSÞÞ
The model used in this paper is adapted from Benjamin et al.’s 2012 paper [40] which is
derived from the work of Kalitzin [39] and has been previously used for simulating epilepsy
surgery [72]. To summarise, this model allows for noise dependent transitions from low to
high amplitude oscillations while in a bistable parameter region. The activation level of the ith
node, zi, is determined by Eq (2) below:
!
X
dzi ðtÞ ¼ f ðzi Þ þ b Mji ðzj ðt tij Þ zi ðtÞÞ dt þ adwi ðtÞ ð2Þ
j6¼i
where f(zi) is a function of zi described in Eq (3), wi(t) is a normally distributed noise term, α
scales the noise, β scales the connectivity matrix M. Mi,j is the connection strength between the

Fig 6. The creation of a connectome. A diagrammatic representation of the connectome creation process,
combining information from T1 MRI and DTI to create a network formed from brain regions connected by
white matter fibres. In this case parcellation of the brain into different regions split the grey matter into 82 ROI,
including 14 subcortical regions. The bottom images show the resulting connectivity matrix (left) of the
connectome and the connectome visualised as nodes and edges overlaid on a brain shadow (right), where
colours denote lobe areas.

Fig 7. Connectivity, node placement and how this relates to the mathematical model. Connectivity is shown where connections exist in 60% or more of
subjects, for visualisation only—all connections were included on a subject specific basis for simulations. Nodes and connections are overlaid on a 3D brain
image to show the rough positioning of the nodes and the distance of connections between them. The nodes are the positions of ROI, with colours
corresponding to different lobes, and the black lines show the connections between different regions. The zoomed in top left graphic shows how connections
between nodes are weighted, and the blue box shows how the activity change of one node is found using the model. The bottom right figure shows a time
series of the model simulation. The model begins in the background state with low amplitude oscillations and transits to a high amplitude seizure-like state at
approximately t = 14, indicated by the dashed line. In this exemplary case the escape time would be taken as 14 for the node in this simulation.
ith and jth nodes, and τi,j is the delay between them. f(zi) is defined by:
f ðzi Þ ¼ ðli 1 þ îoÞzi þ 2zi jzi j zi jzi j

2 4
ð3Þ
Where z is complex, λi controls basin of attraction to the background and seizure-like states,
ω controls frequency of oscillations and î is imaginary.
Delays between areas (τi,j) are incorporated into the model as proportional to the mean
streamline length, grouped into bins, and use a biologically plausible propagation speed of 7
metres per second [25, 103]. S2 Fig explores a comparison between mean and median fibre
lengths as a justification for this approach.
The parameter λi is derived from the difference (Di) of the specific subject’s region surface
area from the control group distribution of region surface areas for the region i, as described in

Eq (4):
Di
li ¼ þ 0:5 ð4Þ
c
The difference (Di) is a measure of how many standard deviations away from the control group
distribution the subject is, found through standard scores with tie-adjustment and continuity cor-
rection. In our case a very negative difference indicates a reduced surface area in region i com-
pared to the control group, and vice-versa. In the controls a leave-one-out method was used to
quantify how far each control was from the remaining controls. The ψ is set to 15, giving a wide
spread for λ whilst ensuring 0 < λ < 1, i.e the bifurcations to monostability for all subjects.
The ω value controls the frequency of oscillations, and in all results detailed here this was
set to a value of 15 to produce an oscillation frequency of 3Hz in the seizure state, which is typi-
cally observed in the intracranial electrocorticographic recordings of many patients with TLE
[104]. The randomised noise w(t) is a complex Wiener process approximated by the Matlab
function randn, generating normally distributed random numbers; this noise is scaled by α.
Matlab was used to implement the model, and a fixed step Euler-Maruyama solver was
employed to find solutions to the equations. Parameter β was fixed at a value of 0.01 for all sub-
jects, and delays were found using the mean fibre lengths. The parameter α, which scales the
amplitude of the noise, was set to a fixed value of 0.05 in order to allow a fair comparison of
escape times without biasing by external factors, i.e. the only parameters influencing compari-
sons are those derived from the subject specific MRI (M, λ, τ).
In line with previous studies [40, 72] we use ‘escape time’ as a measure of epileptogenecity.
Simulations are performed with the initial conditions placing all nodes in the background state,
then upon simulation we measure how long it takes for each node to transit to the seizure state.
This is determined by measuring the Euclidean distance Ei from the fixed point (0,0), over
time. Only when Ei > 1 is the system considered to be in a seizure state. Thus, the escape time
of node i is min(t(Ei > 1)). This is illustrated in Fig 7 (lower right panel).
Validation of the model

To test the performance of the model we ran simulations for both controls and patients over 100
iterations with changing noise seeds. The two groups were compared to identify any differences
in either the node escape times or in the locations of commonly escaping nodes. In addition to
comparing the time taken for the first node to escape we also ran simulations until three nodes
had escaped in succession, taking this as an indicator that the activity had begun to propagate.
Simulation of surgery
We simulated three different conditions of surgery: 1) random resection, 2) clinical resection,
and 3) patient-specific resection. Simulating surgery was achieved by setting inputs and outputs
to and from a resected node to zero values, effectively isolating the node from the network.
1. Random resections removed three nodes from the network, the first chosen at random and
then its two nearest neighbours with connection weights greater than the average were also
removed, in order to remove a set of nodes with similar characteristics to those removed in
clinical resections. These resections serve as a benchmark to which other resective tech-
niques can be compared.

2. Clinical resections removed three regions most commonly resected in an amygdalohippo-

campectomy procedure (ipsilateral hippocampus, amygdala and parahippocampal gyrus).
3. For subject specific optimised resections, the top three fastest escaping nodes were found for
every patient and subsequently removed (see S2 Table for a list of the top three nodes for
each patient). These top escaping nodes were different for every patient, and in many cases
the vector of resected nodes included one or more of the clinically removed areas.
Supporting Information
S1 Fig. Patient and Control escape time comparison, first and second nodes to escape. Simi-
lar to Fig 1, controls and patients show different distributions of escape times. Part a) shows
the times for the first nodes to escape, and part b) plots the times taken for the second nodes to
escape in each subject.
(EPS)
S2 Fig. Mean streamline length is similar to median streamline length between regions.
Exemplary connections between brain areas demonstrating similarity of streamline length
measures. a) Streamlines connecting the left (purple) and right (red) superior frontal areas.
Streamlines are coloured according to their direction. b) Histogram of the lengths of the
streamlines shown in a). c) Streamlines connecting the precentral (yellow) and postcentral
(red) gyri in the right hemisphere. Transparency of the gyri enables visualisation of the stream-
lines. d) Histogram of the lengths of the streamlines shown in c).
(EPS)
S1 Table. The numerical codes for different regions. This table shows which numbers corre-
spond to which brain region. ‘lh’ is short for left hemisphere, (nodes 1 to 41) while ‘rh’ is short
for right hemisphere (nodes 42 to 82).
(PDF)
S2 Table. Top escaping nodes for all subjects. This table shows a list of the 3 nodes which
were consistently the earliest to escape to a seizure state for each individual in patients and con-
trols, ranked in the order of number of appearances. It additionally includes patient age and
gender information, where M indicates a male and F female. The numerical code for the node
regions is explained in the previous table. Where there is a NaN (‘Not a Number’) label instead
of a number there were only one or two nodes which were consistently the earliest to escape.
(PDF)
S3 Table. A table of further node measures for the fastest escaping nodes of patients and
controls. Fibre length measures and surface areas were checked and documented to look for
any evidence of influence. The table shows z scores found from mean fibre lengths normalised
by maximum fibre length and z scores from surface area data. These standard scores show the
deviation from the main distribution and were found for the mean lengths of fibres connected
to the fastest nodes, as well as for the surface areas of the fastest nodes. These measures were
checked for all subjects and also within subject groupings of patients and controls.
(PDF)
S4 Table. Further patient information. Additional information recorded about the patients
included in this study. FS in the Early insult column is short for Febrile Seizures, and the sei-
zure types recorded are simple partial seizures (SPS), complex partial seizures (CPS) and sec-
ondary generalised tonic clonic seizures (SGTCS).
(PDF)

Author Contributions
Conceived and designed the experiments: PNT MK FH. Performed the experiments: FH. Ana-
lyzed the data: FH PNT MK SSK. Contributed reagents/materials/analysis tools: BW CEH
PNT. Wrote the paper: FH PNT MK SSK CEH BW.
References
1. Fisher RS, Boas WvE, Blume W, Elger C, Genton P, Lee P, et al. Epileptic seizures and epilepsy: defi-
nitions proposed by the International League Against Epilepsy (ILAE) and the International Bureau for
Epilepsy (IBE) [Journal Article]. Epilepsia. 2005; 46(4):470–472. doi: 10.1111/j.0013-9580.2005.
66104.x PMID: 15816939
2. Baier G, Goodfellow M, Taylor PN, Wang Y, Garry DJ. The importance of modeling epileptic seizure
dynamics as spatio-temporal patterns [Journal Article]. Frontiers in Physiology. 2012; 3. doi: 10.3389/
fphys.2012.00281 PMID: 22934035
3. WHO. World Health Organisation Epilepsy fact sheet [Web Page]. WHO; 2012. http://www.who.int/
mediacentre/factsheets/fs999/en/
4. Berg AT, Berkovic SF, Brodie MJ, Buchhalter J, Cross JH, Van Emde Boas W, et al. Revised terminol-
ogy and concepts for organization of seizures and epilepsies: report of the ILAE Commission on Clas-
sification and Terminology, 2005–2009. Epilepsia. 2010; 51(4):676–685. doi: 10.1111/j.1528-1167.
2010.02522.x PMID: 20196795
5. Spencer S, Huh L. Outcomes of epilepsy surgery in adults and children. The Lancet Neurology. 2008;
7(6):525–537. doi: 10.1016/S1474-4422(08)70109-1 PMID: 18485316
6. Hemb M, Palmini A, Paglioli E, Paglioli EB, da Costa JC, Azambuja N, et al. An 18-year follow-up of
seizure outcome after surgery for temporal lobe epilepsy and hippocampal sclerosis [Journal Article].
Journal of Neurology, Neurosurgery & Psychiatry. 2013; 84(7):800–805. doi: 10.1136/jnnp-2012-
304038
7. de Tisi J, Bell GS, Peacock JL, McEvoy AW, Harkness WF, Sander JW, et al. The long-term outcome
of adult epilepsy surgery, patterns of seizure remission, and relapse: a cohort study. The Lancet.
2011; 378(9800):1388–1395. doi: 10.1016/S0140-6736(11)60890-8
8. Taylor PN, Han CE, Schoene-Bake JC, Weber B, Kaiser M. Structural connectivity changes in tempo-
ral lobe epilepsy: Spatial features contribute more than topological measures. NeuroImage: Clinical.
2015; 8:322–328. doi: 10.1016/j.nicl.2015.02.004
9. Bonilha L, Elm JJ, Edwards JC, Morgan PS, Hicks C, Lozar C, et al. How common is brain atrophy in
patients with medial temporal lobe epilepsy? Epilepsia. 2010; 51(9):1774–1779. doi: 10.1111/j.1528-
1167.2010.02576.x PMID: 20412283
10. Bonilha L, Kobayashi E, Rorden C, Cendes F, Li L. Medial temporal lobe atrophy in patients with
refractory temporal lobe epilepsy. Journal of Neurology, Neurosurgery & Psychiatry. 2003; 74
(12):1627–1630. doi: 10.1136/jnnp.74.12.1627
11. Bonilha L, Rorden C, Castellano G, Pereira F, Rio PA, Cendes F, et al. Voxel-based morphometry
reveals gray matter network atrophy in refractory medial temporal lobe epilepsy. Archives of neurol-
ogy. 2004; 61(9):1379–1384. doi: 10.1001/archneur.61.9.1379 PMID: 15364683
12. Bonilha L, Rorden C, Castellano G, Cendes F, Li LM. Voxel-based morphometry of the thalamus in
patients with refractory medial temporal lobe epilepsy [Journal Article]. NeuroImage. 2005; 25
(3):1016–1021. doi: 10.1016/j.neuroimage.2004.11.050 PMID: 15809001
13. Pitkänen A, Tuunanen J, Kälviäinen R, Partanen K, Salmenperä T. Amygdala damage in experimen-
tal and human temporal lobe epilepsy. Epilepsy research. 1998; 32(1):233–253. PMID: 9761324
14. Bernasconi N, Duchesne S, Janke A, Lerch J, Collins D, Bernasconi A. Whole-brain voxel-based sta-
tistical analysis of gray matter and white matter in temporal lobe epilepsy. NeuroImage. 2004; 23
15. Thivard L, Lehéricy S, Krainik A, Adam C, Dormont D, Chiras J, et al. Diffusion tensor imaging in
medial temporal lobe epilepsy with hippocampal sclerosis. NeuroImage. 2005; 28(3):682–690. doi:
10.1016/j.neuroimage.2005.06.045 PMID: 16084113
16. Assaf Y, Pasternak O. Diffusion tensor imaging (DTI)-based white matter mapping in brain research: a
review [Journal Article]. Journal of Molecular Neuroscience. 2008; 34(1):51–61. doi: 10.1007/s12031-
007-0029-0 PMID: 18157658
17. Mori S, Zhang J. Principles of diffusion tensor imaging and its applications to basic neuroscience
research [Journal Article]. Neuron. 2006; 51(5):527–539. doi: 10.1016/j.neuron.2006.08.012

18. Besson P, Dinkelacker V, Valabregue R, Thivard L, Leclerc X, Baulac M, et al. Structural connectivity
differences in left and right temporal lobe epilepsy. NeuroImage. 2014; 100:135–144. doi: 10.1016/j.
neuroimage.2014.04.071 PMID: 24814212
19. DeSalvo MN, Douw L, Tanaka N, Reinsberger C, Stufflebeam SM. Altered Structural Connectome in
Temporal Lobe Epilepsy [Journal Article]. Radiology. 2014; 270(3):842–848. doi: 10.1148/radiol.
13131044 PMID: 24475828
20. Bonilha L, Helpern JA, Sainju R, Nesland T, Edwards JC, Glazier SS, et al. Presurgical connectome
and postsurgical seizure control in temporal lobe epilepsy [Journal Article]. Neurology. 2013; 81
(19):1704–1710. doi: 10.1212/01.wnl.0000435306.95271.5f PMID: 24107863
21. Bonilha L, Nesland T, Martz GU, Joseph JE, Spampinato MV, Edwards JC, et al. Medial temporal
lobe epilepsy is associated with neuronal fibre loss and paradoxical increase in structural connectivity
of limbic structures [Journal Article]. Journal of Neurology, Neurosurgery & Psychiatry. 2012; 83
(9):903–909. doi: 10.1136/jnnp-2012-302476
22. Liao W, Zhang Z, Pan Z, Mantini D, Ding J, Duan X, et al. Default mode network abnormalities in
mesial temporal lobe epilepsy: a study combining fMRI and DTI [Journal Article]. Human Brain Map-
ping. 2011; 32(6):883–895. doi: 10.1002/hbm.21076 PMID: 20533558
23. Zhang Z, Liao W, Chen H, Mantini D, Ding JR, Xu Q, et al. Altered functional–structural coupling of
large-scale brain networks in idiopathic generalized epilepsy. Brain. 2011; 134(10):2912–2928. doi:
10.1093/brain/awr223 PMID: 21975588
24. Yan B, Li P. The emergence of abnormal hypersynchronization in the anatomical structural network of
human brain. NeuroImage. 2013; 65:34–51. doi: 10.1016/j.neuroimage.2012.09.031 PMID:
23000784
25. Taylor PN, Goodfellow M, Wang Y, Baier G. Towards a large-scale model of patient-specific epileptic
spike-wave discharges. Biological Cybernetics. 2013; 107(1):83–94. doi: 10.1007/s00422-012-0534-
2 PMID: 23132433
26. Taylor PN, Thomas J, Sinha N, Dauwels J, Kaiser M, Thesen T, Ruths J. Optimal control based sei-
zure abatement using patient derived connectivity. Frontiers in Neuroscience. 2015; 9:202 doi: 10.
3389/fnins.2015.00202 PMID: 26089775
27. Taylor PN, Kaiser M, Dauwels J. Structural connectivity based whole brain modelling in epilepsy. Jour-
nal of Neuroscience Methods. 2014; 236(0):51–57. Available from: http://www.sciencedirect.com/
science/article/pii/S0165027014002945 doi: 10.1016/j.jneumeth.2014.08.010
28. Jirsa VK, Stacey WC, Quilichini PP, Ivanov AI, Bernard C. On the nature of seizure dynamics. Brain.
2014; 137(8):2210–2230. doi: 10.1093/brain/awu133
29. Wang Y, Goodfellow M, Taylor PN, Baier G. Phase space approach for modeling of epileptic dynam-
ics. Physical Review E. 2012; 85(6):061918. doi: 10.1103/PhysRevE.85.061918
30. Taylor PN, Baier G. A Spatially Extended Model for Macroscopic Spike-Wave Discharges. Journal of
Computational Neuroscience. 2011; 31(3):679–684. doi: 10.1007/s10827-011-0332-1 PMID:
21556886
31. Kramer MA, Szeri AJ, Sleigh JW, Kirsch HE. Mechanisms of seizure propagation in a cortical model.
Journal of Computational Neuroscience. 2007; 22(1):63–80. doi: 10.1007/s10827-006-9508-5
32. Marten F, Rodrigues S, Suffczynski P, Richardson MP, Terry JR. Derivation and analysis of an ordi-
nary differential equation mean-field model for studying clinically recorded epilepsy dynamics. Physi-
cal Review E. 2009; 79(2):021911. doi: 10.1103/PhysRevE.79.021911
33. Breakspear M, Roberts JA, Terry JR, Rodrigues S, Mahant N, Robinson PA. A unifying explanation of
primary generalized seizures through nonlinear brain modeling and bifurcation analysis. Cerebral Cor-
tex. 2006; 16(9):1296. doi: 10.1093/cercor/bhj072
34. Proix T, Bartolomei F, Chauvel P, Bernard C, Jirsa VK. Permittivity coupling across brain regions
determines seizure recruitment in partial epilepsy. The Journal of Neuroscience. 2014; 34(45),
15009–15021. doi: 10.1523/JNEUROSCI.1570-14.2014 PMID: 25378166
35. Wendling F, Benquet P, Bartolomei F, Jirsa V. Computational models of epileptiform activity. Journal
of neuroscience methods. 2015; doi: 10.1016/j.jneumeth.2015.03.027
36. da Silva FHL, Blanes W, Kalitzin SN, Parra J, Suffczynski P, Velis DN. Dynamical diseases of brain
systems: different routes to epileptic seizures [Journal Article]. Biomedical Engineering, IEEE Trans-
actions on. 2003; 50(5):540–548. doi: 10.1109/TBME.2003.810703
37. Wendling F, Bartolomei F, Bellanger J, Chauvel P. Epileptic fast activity can be explained by a model
of impaired GABAergic dendritic inhibition. European Journal of Neuroscience. 2002; 15(9):1499–
1508. doi: 10.1046/j.1460-9568.2002.01985.x PMID: 12028360

38. Da Silva FL, Blanes W, Kalitzin SN, Parra J, Suffczynski P, Velis DN. Epilepsies as dynamical dis-
eases of brain systems: basic models of the transition between normal and epileptic activity. Epilepsia.
2003; 44(s12):72–83. doi: 10.1111/j.0013-9580.2003.12005.x
39. Kalitzin SN, Velis DN, Lopes da Silva FH. Stimulation-based anticipation and control of state transitions
in the epileptic brain. Epilepsy & Behavior. 2010; 17(3):310–323. doi: 10.1016/j.yebeh.2009.12.023
40. Benjamin O, Fitzgerald T, Ashwin P, Tsaneva-Atanasova K, Chowdhury F, Richardson MP, et al. A
phenomenological model of seizure initiation suggests network structure may explain seizure fre-
quency in idiopathic generalised epilepsy [Journal Article]. J Math Neurosci. 2012; 2(1):1. doi: 10.
1186/2190-8567-2-1 PMID: 22657571
41. Terry JR, Benjamin O, Richardson MP. Seizure generation: the role of nodes and networks [Journal
Article]. Epilepsia. 2012; 53(9):e166–e169. doi: 10.1111/j.1528-1167.2012.03560.x PMID: 22709380
42. Schindler K, Ghast H, Stieglitz L, Stibal A, Hauf M, Weist R, Mariani L, Rummel C. Forbidden ordinal
patterns of periictal intra-cranial EEG indicate deterministic dynamics in human epileptic seizures. Epi-
lepsia 2011; 52:1771–1780 doi: 10.1111/j.1528-1167.2011.03202.x PMID: 21838792
43. Arruda F, Cendes F, Andermann F, Dubeau F, Villemure J, Jones-Gotman M, et al. Mesial atrophy
and outcome after amygdalohippocampectomy or temporal lobe removal. Annals of neurology. 1996;
40(3):446–450. doi: 10.1002/ana.410400314 PMID: 8797534
44. Richardson MP. Large scale brain models of epilepsy: dynamics meets connectomics. Journal of
Neurology, Neurosurgery and Psychiatry. 2012; 83(12):1238–1248. doi: 10.1136/jnnp-2011-301944
PMID: 22917671
45. Keller SS, Roberts N. Voxel-based morphometry of temporal lobe epilepsy: An introduction and
review of the literature. Epilepsia. 2008; 49(5):741–757. doi: 10.1111/j.1528-1167.2007.01485.x
PMID: 18177358
46. Seidenberg M, Kelly KG, Parrish J, Geary E, Dow C, Rutecki P, et al. Ipsilateral and contralateral MRI
volumetric abnormalities in chronic unilateral temporal lobe epilepsy and their clinical correlates. Epi-
lepsia. 2005; 46(3):420–430. doi: 10.1111/j.0013-9580.2005.27004.x PMID: 15730540
47. Bettus G, Guedj E, Joyeux F, Confort-Gouny S, Soulier E, Laguitton V, et al. Decreased basal fMRI
functional connectivity in epileptogenic networks and contralateral compensatory mechanisms.
Human Brain Mapping. 2009; 30(5):1580–1591. doi: 10.1002/hbm.20625 PMID: 18661506
48. Kullmann DM. What’s wrong with the amygdala in temporal lobe epilepsy? Brain. 2011; 134
(10):2800–2801. doi: 10.1093/brain/awr246 PMID: 21921018
49. Isnard J, Guénot M, Ostrowsky K, Sindou M, Mauguière F. The role of the insular cortex in temporal
lobe epilepsy. Annals of Neurology. 2000; 48(4):614–623. doi: 10.1002/1531-8249(200010)48:4%
3C614::AID-ANA8%3E3.0.CO;2-S PMID: 11026445
50. Guye M, Régis J, Tamura M, Wendling F, Mc Gonigal A, Chauvel P, et al. The role of corticothalamic
coupling in human temporal lobe epilepsy. Brain. 2006; 129(7):1917–1928. doi: 10.1093/brain/awl151
PMID: 16760199
51. McDonald CR, Hagler DJ, Ahmadi ME, Tecoma E, Iragui V, Gharapetian L, et al. Regional neocortical
thinning in mesial temporal lobe epilepsy. Epilepsia. 2008; 49(5):794–803. doi: 10.1111/j.1528-1167.
2008.01539.x PMID: 18266751
52. Riley JD, Moore S, Cramer SC, Lin JJ. Caudate atrophy and impaired frontostriatal connections are
linked to executive dysfunction in temporal lobe epilepsy. Epilepsy & Behavior. 2011; 21(1):80–87.
doi: 10.1016/j.yebeh.2011.03.013
53. Bernhardt BC, Worsley K, Kim H, Evans A, Bernasconi A, Bernasconi N. Longitudinal and cross-sec-
tional analysis of atrophy in pharmacoresistant temporal lobe epilepsy. Neurology. 2009; 72
(20):1747–1754. doi: 10.1212/01.wnl.0000345969.57574.f5 PMID: 19246420
54. Lieb JP, Dasheiff RM, Engel J, et al. Role of the frontal lobes in the propagation of mesial temporal
lobe seizures. Epilepsia. 1991; 32(6):822–837. doi: 10.1111/j.1528-1157.1991.tb05539.x
55. Tomasi D, Volkow ND. Association between functional connectivity hubs and brain networks. Cere-
bral Cortex. 2011; 21(9):2003–2013. doi: 10.1093/cercor/bhq268 PMID: 21282318
56. Sporns O, Honey CJ, Kötter R. Identification and classification of hubs in brain networks. PLoS One.
2007; 2(10):e1049–e1049. doi: 10.1371/journal.pone.0001049 PMID: 17940613
57. Engel J, Pedley TA. Epilepsy: A Comprehensive Textbook, Second Edition. 530 Walnut Street, Phila-
delphia, PA 19106 USA: Lippincott Williams and Wilkins, A Wolters Kluwer business; 2008.
58. Jackson GD, Briellmann RS, Kuzniecky RI. Temporal lobe epilepsy. Elsevier Academic Press, Bur-
lington, MA; 2005.
59. Babalola KO, Patenaude B, Aljabar P, Schnabel J, Kennedy D, Crum W, et al. An evaluation of four
automatic methods of segmenting the subcortical structures in the brain. NeuroImage. 2009; 47

60. Isnard J, Mauguiere F. The insula in partial epilepsy. Revue neurologique. 2005; 161(1):17–26. doi:
10.1016/S0035-3787(05)84970-7 PMID: 15677998
61. Ostrowsky K, Isnard J, Ryvlin P, Guénot M, Fischer C, Mauguière F. Functional mapping of the insular
cortex: clinical implication in temporal lobe epilepsy. Epilepsia. 2000; 41(6):681–686. doi: 10.1111/j.
1528-1157.2000.tb00228.x PMID: 10840399
62. Ryvlin P, Kahane P. The hidden causes of surgery-resistant temporal lobe epilepsy: extratemporal or
temporal plus? editorial review. Current opinion in neurology. 2005; 18(2):125–127. doi: 10.1097/01.
wco.0000162852.22026.6f PMID: 15791141
63. Keller SS, Richardson MP, Schoene-Bake JC, O’Muircheartaigh J, Elkommos S, Kreilkamp B, et al.
Thalamotemporal alteration and postoperative seizures in temporal lobe epilepsy. Annals of neurol-
ogy. 2015; 77(5):760–774. doi: 10.1002/ana.24376 PMID: 25627477
64. Keller SS, Richardson MP, O’Muircheartaigh J, Schoene-Bake JC, Elger C, Weber B. Morphometric
MRI alterations and postoperative seizure control in refractory temporal lobe epilepsy. Human Brain
Mapping. 2015; 36(5):1637–1647. doi: 10.1002/hbm.22722 PMID: 25704244
65. Rosenberg DS, Mauguière F, Demarquay G, Ryvlin P, Isnard J, Fischer C, et al. Involvement of
medial pulvinar thalamic nucleus in human temporal lobe seizures. Epilepsia. 2006; 47(1):98–107.
doi: 10.1111/j.1528-1167.2006.00375.x PMID: 16417537
66. Keller SS, O’Muircheartaigh J, Traynor C, Towgood K, Barker GJ, Richardson MP. Thalamotemporal
impairment in temporal lobe epilepsy: A combined MRI analysis of structure, integrity, and connectiv-
ity. Epilepsia. 2014; 55(2):306–315. doi: 10.1111/epi.12520 PMID: 24447099
67. Seidenberg M, Hermann B, Pulsipher D, Morton J, Parrish J, Geary E, et al. Thalamic atrophy and
cognition in unilateral temporal lobe epilepsy. Journal of the International Neuropsychological Society.
2008; 14(03):384–393. doi: 10.1017/S1355617708080399 PMID: 18419837
68. Bonelli SB, Powell RH, Yogarajah M, Samson RS, Symms MR, Thompson PJ, et al. Imaging memory
in temporal lobe epilepsy: predicting the effects of temporal lobe resection. Brain. 2010;p. awq006.
69. Alstott J, Breakspear M, Hagmann P, Cammoun L, Sporns O. Modeling the impact of lesions in the
human brain [Journal Article]. PLoS Computational Biology. 2009; 5(6):e1000408. doi: 10.1371/
journal.pcbi.1000408 PMID: 19521503
70. Kaiser M, Martin R, Andras P, Young MP. Simulation of robustness against lesions of cortical net-
works. European Journal of Neuroscience. 2007; 25(10):3185–3192. doi: 10.1111/j.1460-9568.2007.
05574.x PMID: 17561832
71. Honey CJ, Sporns O. Dynamical consequences of lesions in cortical networks [Journal Article].
Human Brain Mapping. 2008; 29(7):802–809. doi: 10.1002/hbm.20579 PMID: 18438885
72. Sinha N, Dauwels J, Wang Y, Cash SS, Taylor PN. An in silico approach for pre-surgical evaluation of
an epileptic cortex. In: Engineering in Medicine and Biology Society (EMBC), 2014 36th Annual Inter-
national Conference of the IEEE. IEEE; 2014. p. 4884–4887.
73. Jbabdi S, Johansen-Berg H. Tractography: where do we go from here? Brain Connectivity. 2011; 1
(3):169–183. doi: 10.1089/brain.2011.0033 PMID: 22433046
74. Markov N, Ercsey-Ravasz M, Gomes AR, Lamy C, Magrou L, Vezoli J, et al. A weighted and directed
interareal connectivity matrix for macaque cerebral cortex. Cerebral Cortex. 2012;p. bhs270.
75. David O, Friston KJ. A neural mass model for MEG/EEG:: coupling and neuronal dynamics. Neuro-
Image. 2003; 20(3):1743–1755. doi: 10.1016/j.neuroimage.2003.07.015 PMID: 14642484
76. Felleman DJ, Van Essen DC. Distributed hierarchical processing in the primate cerebral cortex. Cere-
bral Cortex. 1991; 1(1):1–47. doi: 10.1093/cercor/1.1.1 PMID: 1822724
77. Diehl B, Piao Z, Tkach J, Busch RM, LaPresto E, Najm I, et al. Cortical stimulation for language map-
ping in focal epilepsy: correlations with tractography of the arcuate fasciculus. Epilepsia. 2010; 51
(4):639–646. doi: 10.1111/j.1528-1167.2009.02421.x PMID: 20002151
78. Keller CJ, Honey CJ, Mégevand P, Entz L, Ulbert I, Mehta AD. Mapping human brain networks with
cortico-cortical evoked potentials. Philosophical Transactions of the Royal Society B: Biological Sci-
ences. 2014; 369(1653):20130528. doi: 10.1098/rstb.2013.0528
79. Bucci M, Mandelli ML, Berman JI, Amirbekian B, Nguyen C, Berger MS, Henry RG. Quantifying diffu-
sion MRI tractography of the corticospinal tract in brain tumors with deterministic and probabilistic
methods. NeuroImage: Clinical, 2013; 3:361–368. doi: 10.1016/j.nicl.2013.08.008
80. Tournier JD, Calamante F, Connelly A. MRtrix: Diffusion tractography in crossing fiber regions. Inter-
national Journal of Imaging Systems and Technology. 2012; 22:53–66. doi: 10.1002/ima.22005
81. Lim S, Han CE, Uhlhaas PJ, Kaiser M. Preferential detachment during human brain development:
Age-and sex-specific structural connectivity in diffusion tensor imaging (DTI) data. Cerebral Cortex.
2015; 25(6):1477–1489. doi: 10.1093/cercor/bht333 PMID: 24343892

82. Avecillas-Chasin JM, Alonso-Frech F, Parras O, del-Prado N, Barcia JA. Assessment of a method to
determine deep brain stimulation targets using deterministic tractography in a navigation system. Neu-
rosurgical review, 2015; 38(4):739–751. doi: 10.1007/s10143-015-0643-1 PMID: 25962557
83. Catani M, Jones DK, Donato R. Occipito temporal connections in the human brain. Brain, 2003; 126
(9):2093–2107. doi: 10.1093/brain/awg203
84. Mori S, Van-Zijl PC. Fiber tracking: principles and strategies-a technical review. NMR in Biomedicine.
2002; 15(7–8):468–480. doi: 10.1002/nbm.781
85. Khalsa S, Mayhew SD, Chechlacz M, Bagary M, Bagshaw AP. The structural and functional connec-
tivity of the posterior cingulate cortex: Comparison between deterministic and probabilistic tractogra-
phy for the investigation of structure–function relationships. NeuroImage. 2014; 102:118–127. doi: 10.
1016/j.neuroimage.2013.12.022
86. Bonilha L, Gleichgerrcht E, Fridriksson J, Rorden C, Breedlove JL, Nesland T, Paulus W, Helms G,
Focke NK. Reproducibility of the Structural Brain Connectome Derived from Diffusion Tensor Imaging.
PLoS One 2015; 10(9): e0135247. doi: 10.1371/journal.pone.0135247 PMID: 26332788
87. Girard G, Whittingstall K, Deriche R, Descoteaux M. Structural connectivity reproducibility through
multiple acquisitions. Organization for Human Brain Mapping. 2015;Conference paper
88. Lilja Y, Ljungberg M, Starck G, Malmgren K, Rydenhag B, Nilsson DT. Visualizing Meyer’s loop: A
comparison of deterministic and probabilistic tractography. Epilepsy research, 2014; 108(3):481–490.
doi: 10.1016/j.eplepsyres.2014.01.017 PMID: 24559840
89. Zalesky A, Fornito A, Harding IH, Cocchi L, Yücel M, Pantelis C, et al. Whole-brain anatomical net-
works: does the choice of nodes matter? NeuroImage. 2010; 50(3):970–983. doi: 10.1016/j.
neuroimage.2009.12.027 PMID: 20035887
90. de Reus MA, Van den Heuvel MP. The parcellation-based connectome: limitations and extensions.
NeuroImage, 2013; 80:397–404. doi: 10.1016/j.neuroimage.2013.03.053 PMID: 23558097
91. Gordon EM, Laumann TO, Adeyemo B, Huckins JF, Kelley WM, Petersen SE. Generation and evalu-
ation of a cortical area parcellation from resting-state correlations. Cerebral Cortex. 2014;. doi: 10.
1093/cercor/bhu239
92. Cabral J, Hugues E, Kringelbach ML, Deco G. Modeling the outcome of structural disconnection on
resting-state functional connectivity. NeuroImage. 2012; 62(3): 1342–1353. doi: 10.1016/j.
neuroimage.2012.06.007 PMID: 22705375
93. Kochunov P, Williamson DE, Lancaster J, Fox P, Cornell J, Blangero J, Glahn DC. Fractional anisot-
ropy of water diffusion in cerebral white matter across the lifespan. Neurobiology of aging 2012; 33(1):
9–20. doi: 10.1016/j.neurobiolaging.2010.01.014 PMID: 20122755
94. Wijtenburg SA, McGuire SA, Rowland LM, Sherman PM, Lancaster JL, Tate DF, Hardies LJ, Patel B,
Glahn DC, Hong LE, Fox PT, Kochunov P. Relationship between fractional anisotropy of cerebral
white matter and metabolite concentrations measured using 1 H magnetic resonance spectroscopy in
healthy adults. NeuroImage. 2013; 66:161–168. doi: 10.1016/j.neuroimage.2012.10.014 PMID:
23073233
95. Wang Y, Goodfellow M, Taylor P, Baier G. Dynamic mechanisms of neocortical focal seizure onset.
PLoS Computational Biology. 2014; 10(8):e1003787. doi: 10.1371/journal.pcbi.1003787 PMID:
25122455
96. FreeSurfer;. http://surfer.nmr.mgh.harvard.edu/. Available from: http://surfer.nmr.mgh.harvard.edu/
97. Fischl B, van der Kouwe A, Destrieux C, Halgren E, Ségonne F, Salat DH, et al. Automatically parcel-
lating the human cerebral cortex. Cerebral Cortex. 2004; 14(1):11–22. doi: 10.1093/cercor/bhg087
PMID: 14654453
98. Desikan RS, Ségonne F, Fischl B, Quinn BT, Dickerson BC, Blacker D, et al. An automated labeling
system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest.
NeuroImage. 2006; 31(3):968–980. doi: 10.1016/j.neuroimage.2006.01.021 PMID: 16530430
99. Mori S, Barker PB. Diffusion magnetic resonance imaging: its principle and applications. The Anatomi-
cal Record. 1999; 257(3):102–109. doi: 10.1002/(SICI)1097-0185(19990615)257:3%3C102::AID-
AR7%3E3.3.CO;2-Y PMID: 10397783
100. Wang R, Benner T, Sorensen A, Wedeen V. Diffusion toolkit: a software package for diffusion imaging
data processing and tractography. In: Proceedings of the International Society for Magnetic Reso-
nance in Medicine. vol. 15; 2007. p. 3720.
101. van den Heuvel MP, de Reus MA, Feldman Barrett L, Scholtens LH, Coopmans FM, Schmidt R, Pre-
uss TM, Rilling JK, Li L. Comparison of diffusion tractography and tract tracing measures of connectiv-
ity strength in rhesus macaque connectome. Human Brain Mapping. 2015; 36(8):3064–3075 doi: 10.
1002/hbm.22828 PMID: 26058702

102. Liao W, Zhang Z, Pan Z, Mantini D, Ding J, Duan X, et al. Altered functional connectivity and small-
world in mesial temporal lobe epilepsy. PloS One. 2010; 5(1):e8525. doi: 10.1371/journal.pone.
0008525 PMID: 20072616
103. Bojak I, Oostendorp TF, Reid AT, Kötter R. Connecting mean field models of neural activity to EEG
and fMRI data. Brain topography. 2010; 23(2):139–149. doi: 10.1007/s10548-010-0140-3 PMID:
20364434
104. Quiroga RQ, Garcia H, Rabinowicz A. Frequency evolution during tonic-clonic seizures. Electromyog-
raphy and clinical neurophysiology. 2002; 42(6):323–332.

Predicting Surgery Targets in Temporal Lobe Epilepsy Through Structural Connectome Based Simulations

Uploaded by

Copyright:

Available Formats

Predicting Surgery Targets in Temporal Lobe Epilepsy Through Structural Connectome Based Simulations

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Predicting Surgery Targets in Temporal Lobe Epilepsy Through Structural Connectome Based Simulations

Uploaded by

Copyright:

Available Formats

RESEARCH ARTICLE

Predicting Surgery Targets in Temporal Lobe

‡PNT and MK are joint senior authors.

Citation: Hutchings F, Han CE, Keller SS, Weber B,

PLOS Computational Biology | DOI:10.1371/journal.pcbi.1004642 December 10, 2015 1 / 24

Foundation of South Korea funded by the Ministry of

PLOS Computational Biology | DOI:10.1371/journal.pcbi.1004642 December 10, 2015 2 / 24

PLOS Computational Biology | DOI:10.1371/journal.pcbi.1004642 December 10, 2015 3 / 24

PLOS Computational Biology | DOI:10.1371/journal.pcbi.1004642 December 10, 2015 4 / 24

PLOS Computational Biology | DOI:10.1371/journal.pcbi.1004642 December 10, 2015 5 / 24

PLOS Computational Biology | DOI:10.1371/journal.pcbi.1004642 December 10, 2015 6 / 24

PLOS Computational Biology | DOI:10.1371/journal.pcbi.1004642 December 10, 2015 7 / 24

PLOS Computational Biology | DOI:10.1371/journal.pcbi.1004642 December 10, 2015 8 / 24

Source Clinical (%) Patient Speciﬁc (%)

PLOS Computational Biology | DOI:10.1371/journal.pcbi.1004642 December 10, 2015 9 / 24

PLOS Computational Biology | DOI:10.1371/journal.pcbi.1004642 December 10, 2015 10 / 24

Surgery prediction in the context of other works

PLOS Computational Biology | DOI:10.1371/journal.pcbi.1004642 December 10, 2015 11 / 24

patient derived connectivity, found encouraging results in simulations of stimulation applied

Limitations of the study

PLOS Computational Biology | DOI:10.1371/journal.pcbi.1004642 December 10, 2015 12 / 24

Future work and improvements

PLOS Computational Biology | DOI:10.1371/journal.pcbi.1004642 December 10, 2015 13 / 24

PLOS Computational Biology | DOI:10.1371/journal.pcbi.1004642 December 10, 2015 14 / 24

PLOS Computational Biology | DOI:10.1371/journal.pcbi.1004642 December 10, 2015 15 / 24

f ðzi Þ ¼ ðli 1 þ ^ioÞzi þ 2zi jzi j zi jzi j

PLOS Computational Biology | DOI:10.1371/journal.pcbi.1004642 December 10, 2015 16 / 24

Validation of the model

PLOS Computational Biology | DOI:10.1371/journal.pcbi.1004642 December 10, 2015 17 / 24

2. Clinical resections removed three regions most commonly resected in an amygdalohippo-

PLOS Computational Biology | DOI:10.1371/journal.pcbi.1004642 December 10, 2015 18 / 24

PLOS Computational Biology | DOI:10.1371/journal.pcbi.1004642 December 10, 2015 19 / 24

PLOS Computational Biology | DOI:10.1371/journal.pcbi.1004642 December 10, 2015 20 / 24

PLOS Computational Biology | DOI:10.1371/journal.pcbi.1004642 December 10, 2015 21 / 24

PLOS Computational Biology | DOI:10.1371/journal.pcbi.1004642 December 10, 2015 22 / 24

PLOS Computational Biology | DOI:10.1371/journal.pcbi.1004642 December 10, 2015 23 / 24

PLOS Computational Biology | DOI:10.1371/journal.pcbi.1004642 December 10, 2015 24 / 24

You might also like