Computational 

Neuroscience II: 
Applications
Gustavo Patow
IMAE / UdG

1
 Applications: 
Epilepsy and BCI

2
Epilepsy

3
 INTRODUCTION: From epileptic seizure…
An epileptic seizure is a transient occurrence of signs
and/or symptoms due to abnormal excessive or
synchronous neuronal activity in the brain.

- low-voltage fast discharges (LVFD)(#)

Two generic patterns can be distinguished:
- and spike-and-wave discharges (SWD)( )
 INTRODUCTION: …to Epilepsy
• Drug resistant focal epilepsies are 
severe diseases affecting young people 
(1% of the world population, high 
morbidity and mortality)
• Epilepsy surgery is the only potential 
curative treatment and consists in the 
removal of the most epileptogenic brain 
Jean Talairach & Jean Bancaud
Paris in 1960s regions
• Includes depth electrode implantation 
to record the intracerebral EEG signals 
(stereoelectroencephalography, SEEG)
• Failures (complete or partial) in about 
40% (particularly in extra temporal lobe 
epilepsy)
• No significant improvement of surgery  
success rate in 50 years
SEEG RECORDINGS

cough, touch the  tapping nurse’s 
nose, cross the legs hands,  "bella, bella!"
start singing, rhythmic basin,  
balancement, joviality
How to use whole‐brain 
models to model epilepsy 
spread?
 How to model epilepsy? Jirsa et al., Brain (2014)  
Jirsa et al., Neuroimage (2017)
Proix et al., Brain (2017)
Workflow of technology Nonlinear dynamic model
Standard pipelines:
Not physiologically detailed/realistic  
FSL, MrTrix, FreeSurfer,… Functional Personalized
Dynamically detailed
Brain Network model

Noninvasive brain Construction of Functional region

imaging brain avatar model selection

Patient specific
interventions Develop Finger Refine
patient printing of network
charts patient pathology

Parameter space  Data fitting
exploration
How to model epilepsy? Exemplary 
patient
How to model epilepsy? Exemplary 
[Jirsa et al. Neuroimage (2017)]
patient
 Jirsa et al IEEE 2002

How to model epilepsy? Ghosh et al. PLoS CB 2008

Deco, Jirsa, McIntosh Nat Rev Neurosci 2011  
Deco, Jirsa Journ Neurosci 2012

Large‐scale network model Deco, Jirsa, McIntosh TINS 2013
Ritter et al Brain Connectivity 2013  
Jirsa et al. Neuroimage 2017 
Petkoski et al. Plos CB 2018
 Sanz-Leon et al. Front. Neuroinform
How to model epilepsy? 2013, Neuroimage 2015

Building a large‐scale network model
connectivity from DTI

in put
topography from MRI

Weight matrix Tract length

matrix
 Sanz-Leon et al. Front. Neuroinform
How to model epilepsy? 2013, Neuroimage 2015

Building a large‐scale network model
connectivity from DTI
neural mass

in put in put
topography from MRI

Weight matrix Tract length

matrix
 How to model epilepsy? Jirsa et al. Brain 2014; 
Saggio et al. J Math Neurosci 2017

Modeling seizures propagation: Epileptor
[seizure recording from A. Ikeda]
onset bifurcation offset bifurcation
Phenomenological neural mass model for reproducing
seizure temporal dynamics
fast discharges

Taxonomy of seizures
slow (metabolic) processes

spike and wave

Parameter controls the tissue excitability/epileptogenicity:

healthy epileptogenic
 How to model epilepsy? Jirsa et al. Brain 2014; 
Saggio et al. J Math Neurosci 2017

Modeling seizures propagation: Epileptor
[seizure recording from A. Ikeda]
onset bifurcation offset bifurcation
Phenomenological neural mass model for reproducing
seizure temporal dynamics
fast discharges

Taxonomy of seizures
slow (metabolic) processes

spike and wave

Epileptor Time Series

Parameter controls the tissue excitability/epileptogenicity:

healthy epileptogenic
 How to model epilepsy? Jirsa et al. Brain 2014; 
Saggio et al. J Math Neurosci 2017

Modeling seizures propagation: Epileptor
[seizure recording from A. Ikeda]
onset bifurcation offset bifurcation
Phenomenological neural mass model for reproducing
seizure temporal dynamics
slow-variable
Taxonomy of seizures

The permittivity coupling enables seizure propagation:

Parameter controls the tissue excitability/epileptogenicity:

healthy epileptogenic
How to model epilepsy? [Jirsa et al. Neuroimage (2017)]

Building a virtual epileptic patient
epileptor in put

clinical hypothesis

connectivity from DTI

topography from MRI

in put
How to model epilepsy? [Jirsa et al. Neuroimage (2017)]

Building a virtual epileptic patient
• MRI anomalies
• definition of the structural anomaly as new
seed ROI by delineation in T1 and DWI
• and integration in the model via a
modification of the local connectivity of
the ROI (here the HH)
How to model epilepsy? [Jirsa et al. Neuroimage (2017)]

Building a virtual epileptic patient
Epileptogenic Networks
 How to model epilepsy? [Jirsa et al. Neuroimage (2017)]

Building a virtual epileptic patient
in put
epileptor
o utput

clinical hypothesis Spatial organization of EZ and PZ matches

empirical data

connectivity from DTI

topography from MRI in put

 How to model epilepsy? [Jirsa et al. Neuroimage (2017)]

Exemplary patient
Patient-specific virtual brain

VEP model captures both simple and complex seizure spread

as well as responses towards stimulation
From virtual epilepsy model to 
patient’s bed

Some use‐cases
 VEP USE‐CASE 1 [Proix et al. Brain 2017]

Predicting seizure propagation
VEP: validation for 2
cohorts (> 50patients)
?

Scoring of predicted PZ Engel score for epilepsy surgery

binary score PZ predicEon Engel score

Connectome-based virtualization **

Discrepancy with TVB

predict PZ

Green: discrepancy with TVB

Yellow: Epileptogenic Zone
Red: Propagation zone

connectomes
 [Olmi et al., Plos CB 2019]
VEP USE‐CASE 2
Controlling seizure propagation: Resection vs. Lesion
Background

Proix et al.
Brain, 2017

resection

simulations

• gamma knifes, laser surgery…

 [Olmi et al., Plos CB 2019]
VEP USE‐CASE 2
Controlling seizure propagation: Resection vs. Lesion
normal resection minimally invasive resection

Entire EZ
200 Random Lesions
Selected Lesions
Strongest weights EZ
Number of Lesions

Strongest weights EZ+PZ

150

100

50

0 cj cm fb pg ac et gc il ml cv fbo js pc sf rb
VEP USE‐CASE 3 [Hashemi et al., 2020]

The Bayesian Virtual Epileptic Patient
fast
4

discharges
3.8

3.6

SEEG data 3.4

rest
preprocessing:
3.2

band pass filter -2 0

10-120Hz
- 1.5
-1 -5
- 0.5
- 10
0
0.5 - 15
1
1.5 - 20

Data feature for

fitting:
Energy
dissipation in
space and time
 VEP USE‐CASE 3 [Hashemi et al., 2020]

The Bayesian Virtual Epileptic Patient
Main contents • A world’s first clinical trial of
of the clinical report personalized brain modelling in epilepsy.
• 400 patients in 12 hospitals across France.
 [Courtiol et al., 2020]
VEP USE‐CASE 4
The Resting Virtual Epileptic Patient
pathologic/epileptic
Extended Epileptor model neurons pop

Epileptor
B

Hopf
C

physiologic/healthy
Nonlinear dynamical model
neurons pop
Not physiologically detailed/realistic
Dynamically detailed
 [Courtiol et al., 2020]
VEP USE‐CASE 4
The Resting Virtual Epileptic Patient
Empirical data for 15 patients 
and 5 controls.
Interictal discharges do not 
change  overall connectivity.

*
Global optimal 
working point  
b2 = 1
shifted towards 
lower excitability in 
b2 =3.25 epileptic patients.
1s
Brain‐Computer Interfaces 
(BCI)

30
 The Brain‐Computer Interface
• A brain‐computer interface (BCI) is a computer‐based system 
that acquires brain signals, analyzes them, and translates them 
into commands that are relayed to an output device to carry 
out a desired action.
• In principle, any type of brain signal could be used to control a 
BCI system
• The most commonly studied signals are electrical signals from brain 
activity measured from electrodes on the scalp, on the cortical 
surface, or in the cortex…

31
What is a Brain‐Computer Interface?
• Our brains are composed of neurons that interact
electrically and chemically
• The activity of one neuron cannot be measured on the scalp.
• When a large number of neurons (100,000's) in a small
region of the outer layer (cerebral cortex) are active
together, the sum of their electric currents becomes
strong enough to be measured on the scalp.
• Electroencephalography (EEG) is the measurement of
weak electric potentials on the scalp ~ 10-50
microvolts

32
 The Brain‐Computer Interface
• A BCI system consists of 4 sequential components: 
1. signal acquisition
2. feature extraction
3. feature translation
4. device output. 
• They are controlled by an operating protocol that defines the 
onset and timing of operation, the details of signal processing, 
the nature of the device commands, and the oversight of 
performance

33
 What is a 
BCI?

[Shih et al., 2012]
 Signal Acquisition
Electroencephalography (EEG)

Deep EEG
35
 Signal Acquisition
Magnetoencephalography (MEG)

36
EEG at Different Scalp Locations
EEG activity changes quickly.
EEG is measured at electrodes
placed at multiple scalp locations one second of EEG
Standard 10-20 System for
electrode positions

EEG at multiple electrodes

positive negative

http://butler.cc.tut.fi/~malmivuo/bem/bembook/13/13.htm
Morris and Maisto, Psychology: An Introduction
 Stimulus Driven

Visual stimulus is rare but

expected

For example, looking for an A:

DEFXAJLTA...

When A appears, EEG signal

will contain a P300 wave, a stimulus
Positive deflection 300 ms after
stimulus

Must average the EEG over

multiple trials to detect the
P300 wave
300 ms
Winterer, et al., Arch Gen Psychiatry, 2003;60:1158-1167.
 Typing using the P300
In 1988, Donchin developed use of P300 for
typing application.

Row or column chosen randomly and

highlighted about 5 times a second.

Subject looks at desired letter.

When row or column containing the letter

lights up, P300 wave occurs.

Reliable detection requires averaging over

15 trials

Remains the most common brain-computer

interface

1988: Farwell and Donchin, Electroencephalogr. Clin. Neurophysiol., vol. 70, pp. 510–523 .
http://ida.first.fraunhofer.de/projects/bci/competition_ii/albany_desc/albany_desc_ii.html
P300 Spelling Accuracy in Lab and Home

• Single letter appears

• Count a target letter
• b, p, d
• 20 - each target letters
• 20 - non-target letters
 Our Approach:
BCI in Real-Life Situations

Adults with no Adults with severe motor

impairments impairments
Data collected in the lab Data collected in homes
Accuracy of Selecting Neuro-Pulse G.Tec Biosemi
Correct Letter 1 76 % 63 84

2 84 80 82

3 96 68 88

Subjects with 4 68 75 75
no impairments 5 66 74 89
in lab 83 77 62
6
7 75 70 79

8 73 70 91

9 78 73 70

Averag 76 72 80
e
10 64 63

11 85 65

12 90 92 89
Patients in
13 45 73
homes
14 74 60 71

15 73 70 90

16 63 76 69

Averag 68 77 74
e
Replacing Senses

43
Electrical Interference from
Medical Equipment?

Power wheelchair

Hospital bed

Respirator
 Signal Processing
• Azimo can be controlled using a device that measures brain signals
• The device uses and EEG and a Near‐InfraRed Spectroscopy (NIRS) sensor
• 90% accuracy rate

45
Assignment
• Remember: you can do all the exercises in the 
assignment
• Hand‐in: 4 weeks after last theoretical class
• Assignment @ Moodle
• Just jupyter notebook
• And run the ipynb file!
• Problems? Do not hesitate to ask!

46
Bibliography: Epilepsy
• V.K. Jirsa, W.C. Stacey, P.P. Quilichini, A.I. Ivanov, C. Bernard, On the nature of seizure dynamics, Brain, 137 (8) 
(2014), pp. 2210‐2230
• P. Sanz Leon, S. Knock, M. Woodman, L. Domide, J. Mersmann, A. McIntosh, V. Jirsa, The Virtual Brain: a 
simulator of primate brain network dynamics, Front. Neuroinf., 7 (2013), p. 10
• P. Sanz‐Leon, S.A. Knock, A. Spiegler, V.K. Jirsa, Mathematical framework for large‐scale brain network modeling 
in the virtual brain, Neuroimage, 111 (2015), pp. 385‐430
• V. Jirsa, T. Proix, D. Perdikis, M. Woodman, H. Wang, J. Gonzalez‐Martinez, C. Bernard, C. Benar, M. Guye, P. 
Chauvel, F. Bartolomei, The virtual epileptic patient: individualized whole‐brain models of epilepsy spread, 
NeuroImage, 145 (2017), pp. 377‐388
• T. Proix, F. Bartolomei, M. Guye, V.K. Jirsa, Individual brain structure and modelling predict seizure propagation, 
Brain, 140 (3) (2017), pp. 641‐654
• Maria Luisa Saggio, Andreas Spiegler, Christophe Bernard & Viktor K. Jirsa, Fast–Slow Bursters in the Unfolding 
of a High Codimension Singularity and the Ultra‐slow Transitions of Classes, The Journal of Mathematical 
Neuroscience volume 7, Article number: 7 (2017) 
• S. Olmi, S. Petkoski, M. Guye, F. Bartolomei, V. Jirsa, Controlling seizure propagation in large‐scale brain 
networks, PLoS Comput. Biol., 15 (2) (2019), pp. 1‐23
• Julie Courtiol, Maxime Guye, Fabrice Bartolomei, Spase Petkoski and Viktor K. Jirsa, Dynamical Mechanisms of 
Interictal Resting‐State Functional Connectivity in Epilepsy, Journal of Neuroscience 15 July 2020, 40 (29) 5572‐
5588
• M. Hashemi, A.N. Vattikonda, V. Sip, M. Guye, F. Bartolomei, M.M. Woodman, V.K. Jirsa, The Bayesian Virtual 
Epileptic Patient: A probabilistic framework designed to infer the spatial map of epileptogenicity in a 
personalized large‐scale brain model of epilepsy spread, NeuroImage, Volume 217, 2020, 116839, ISSN 1053‐
8119 47
Bibliography: BCI
• “Biological Psychology, Introduction to behavioural, cognitive and 
clinical neuroscience”  Breedlove, Watson and Rosenweig 2010
• “How BCI’s work”
http://computer.howstuffworks.com/brain‐computer‐
interface2.htm
• Okada, Y. (1983): Neurogenesis of evoked magnetic fields. In: 
Williamson, S.H.; Romani, G.L.; Kaufman, L.; Modena, I., editors. 
Biomagnetism: an Interdisciplinary Approach. New York: Plenum 
Press, pp 399‐408
• “Disruptions: Brain Computer Interfaces Inch Closer to Mainstream”
http://bits.blogs.nytimes.com/2013/04/28/disruptions‐no‐words‐
no‐gestures‐just‐your‐brain‐as‐a‐control‐pad/
Group D 48
Computational Neuroscience II: 
Applications
Thanks!

49