Functional brain Imaging :from measurement to cognition

Line GarneroLaboratoire de Neurosciences Cognitives & Imagerie Cérébrale

CNRS UPR640

Centre de Magnétoencéphalographie

Hôpital La Salpêtrière

Brain Imaging

Principle

• Brain in action

•Rely «  brain state «  with a behaviour (motor, cognitive on normal and pathological subjects)

• Non invasive : available to human brain

• Large scale observation (106 neurones)

Properties

Required Models

- Observable quantities for activation : neurobiological and hemodynamical model

-Recording devices : Physical model : from biology to physics

-Experimental protocols :Psychological / cognitive models of processing

- Data AnalysisBrain processing model

-Interpretation :Based on both models

SUMMARY

-Neurophysiological bases

-Recording devices and physical models

-Exprimental protocols

-Data analysis : segregation or integration

-«Brain Reading » and Brain Computer Interface

Two imaging modalities

• Electrical neuronal activity• Non invasive Imaging• High temporal resolution• Problem for localisation

EEG/MEG

• Hemodynamical activity• Non invasive • Spatial resolution ~1mm• Limited temporal resolution

Functional MRI

Neurobiological principles

Bases physiologiques de l’ I. C. F.

MetabolismeATP synthesis

Consumption ofGlucose et O2

EEG - MEG

PPSE - PPSIIntracellular ou

extracellular

HemodynamicsDeoxygenation

Blood Flow Increase

TEP

TEP

IRMf

NeuronalAction Potential

D’après B. Mazoyer

Recordings principlesInstrumentation

Recorded currents

Quelques mm2

Conduction currents (extracellular)

Current dipole intracellular

Cortical Macrocolumn 105 - 106 neuronesQ =I x d ~10 à 100 nAm

I

Physical modelisation of the EM fields

Js : source currentJc : conduction current

Jc = E = - grad(V) tissue conductivities

div(Js +Jc) = 0 ====> div(Js) = div[grad(V)]

dv'rr

'rr)JJ(

4)r(B

3cs0

Biot et Savart Law

Maxwell equations in quasistatic approximation (PPSE : 10 ms)

Electroencephalography (EEG) et Magnétoencéphalography (MEG)

:

MEG : magnetic field measurement.Scale : 10-13 teslaSQUID sensors1st MEG : 1969

EEG : Electric Potential measurements.Electrodes on the head surface

Scale : few microvolts1er EEG : 1929

MEG INSTRUMENTATION

Detection :

Flux detection : coils+

Flux transformer : squids

SQUIDS : Low Temperature Supra Josephson Junctions

MEG EEG

functional MRI : principle

Capillaires

i i

e

BeB

e Be

Bi Bi

BOLD effectBOLD effect Oxyhémoglobine : diamagnetic. Désoxyhémoglobine : paramagnétique.

• local variation of magnetic susceptibility due to the variation of concentration of désoxyhémoglobine (intrisic contrast agent)

• variation of the RMN signal amplitude

Hemodynamical responseHemodynamical response

Response to a stimulus : slow variation (15 s).

Average on many repetitions

Functional MRI : principle

Outside B

Inside B

M

ResonanceRadiofrequency B1

B1

B0

Time relaxationsBloch

equations

MRI Physical principle

Proton spins

IRM screen projecteur

PC

plaque à filtre

MRI acquisition

temps

FMRI sequence

Data : Structural and functional MRI

IRMfsignal IRMf

IRMa

Transversal relaxation time T2*

Longitudinal relaxation time T1

Experimental protocols

Experimental protocols

• Objectives– Reveals cerebral activation linked to a mental process

(cognitive, sensori-moteur...) • Test the activation of a specific area

• Localize the areas activated in a given process

• Find the dynamics of a mental process (chronometry, netwrk dynamics...)

• Localize the origin of the measured fields source reconstruction

?IRMfIRMf

MEEGMEEG

[N. George et A.L. Paradis]

•Repetition of different tasks and stimuli

•Contrasts between tasks and conditions

• Study on group of subjects

• comparison between different groups

Experimental protocols

Cognitive substraction

• Equalization of all required processes Except process of interest

• Hypothesis

– The differential cerebral activity of the contrasts TEST - CONTROL reveal only the process of interest

CONTROLProcessOf interest CONTRÔLE _ =

ProcessOf interest

TEST

• constant task and variable stimulus• constant stimulus and variable task • keeping task and stimulus fixed and variable « internal state » [N. George et A.L. Paradis]

Data Analysis

Cerebral processing theoriesCerebral processing theories

Functional segregation : spatial cluster of cells having a same functional role.

Functional specialisation : a cortical area is specialized in a (sub-)processing of one (several) function(s).

Functional integration : transitory cooperation of several areas for the realization of one function

Edelman et Tononi, 2001

Data anaysis principles for specialization

Seggregation :Area localizationChronometrySequential processing

MEG/EEG analysis : segregation

Sequential processing : evoked potentials

Dawson, 1951

repetition of stimulations and

conditions

reproducibility of neural events

evoked by the condition

(task+stimulus) and subject state

Interpretation : chronology

Early latencies (<< 200 ms)

Exogeneous waves

• processing in sensory areas or •Depend on the physical properties of the stimulus

Late latencies > ( 200 ms)

Endogeneous waves

• associative and cognitive processing•Depend on the task and subject state

NomenclatureEn EEG: Pxxx ou Nxxx, positif or negative potential peaking at culminant à xxx msEn MEG: Mxxx, magnetif field peaking at xxx ms

N145P100

Source localisation

Reconstruction in time and space of neural sources at the origin of MEG and EEG surface signals

J?

Inverse problem

• ill posed problem• focal or distributed source models

Direct problem

• conductivity values of head tissues• numerical resolution

CapteurMEG

B(t)

V(t)ElectrodeEEGJ(t)

Dipolar models

HypothesisFew areas are activated siumltaneouslyFocal activation modeled by one dipolar current

thumbindex

middlelittle

ExampleHand finger somatotopy at 30 ms

Right hand

index Little finger

Distributed models

ResolutionLinear inverse problemRegularisationOne image for each time sample

HypothesesNo prior on the number of activated areasDistribution of sources normal to the cortical surface Estimation of sources amplitudes

350-400ms

descent

P. Senot et al., in revision

catch

Senot et al

Application : ball catching

Temporal series fMRI voxel time

course

Statistical image(SPM)

amplitude

time

General Linear ModelFittingstatistical image

fMRI data analysis

J.B. Poline

Quantification of voxel activation by a model of the hemodynamic response (function gamma, de Poisson ou gaussienne).

fMRI analysis principle

1 s 15 s

ONOFF

Convolution of the the time serie of the protocol with this function

Test of significativity at each voxel (comparison between bold isgnals and convoluted function) Individual and group statistics

Gaze direction

Head orientation

direct averted

straight

oblique

fMRI : example

4 experimental conditions :

D’après N. George

All conditions versus rest

IRM fonctionnelle :results

Fusiform gyrus area : direct versus averted

Cooperation :Network characterization :Interaction between areasCerebral connectivityParallel processing

Data anaysis principles for cooperation

Oscillations in MEG/EEG

HypothesesTransient neuronal assemblies (Varela et al, 2001)Any cognitive process corresponds to the emergency of a neuronal assemby distributed, specific, transient and synchronous

Local oscillatory activitiesLocal signs of loop activation Time frequency analysis

Long distance synchronyInteraction between areas (from sensors or sources)Coherence, phase synchronisation

12

10

8

6

4

0 400 800-400

b 60

40

20

Pui

ssan

ce é

mis

e (

)

12

10

8

6

4

0 400 800-400

a 60

40

20

Fre

quen

ce (

Hz)

Per

cept

ion

Temps (ms)

FR

EQ

UE

NC

E

PERCEPTION NO PERCEPTION

180 -360 ms 180 -360 ms 360 -450 ms360 -450 mssynchronie De-synchrony Absence of synchrony

Rodriguez et al, Nature, 1999

Gamma band : 30 – 60 Hz

Networks imaging

– Inverse problem + dynamical analysis of synchronies

Cortical sourcesDynamic links

t

… .. .… …Rivalité Binoculaire

Cosmelli et al

Functional integrationAnalyses of inter-regional effects :

fMRI analysis for cooperation

From SPM course

Functional connectivity

= the temporal correlation = the temporal correlation betweenbetween spatially remote spatially remote neurophysiologicalneurophysiological events events

Functional connectivity

= the temporal correlation = the temporal correlation betweenbetween spatially remote spatially remote neurophysiologicalneurophysiological events eventsMODEL-free

Effective connectivity

= the influence that the elements = the influence that the elements of a neuronal system exert over of a neuronal system exert over anotheranother

Effective connectivity

= the influence that the elements = the influence that the elements of a neuronal system exert over of a neuronal system exert over anotheranother

MODEL-dependent

Defined by the correlation of the BOLD signal between regions A and B

Area A

Area B

Ensemble de régions

Functional connectivity

From H. Benali

R-Cereb Cx

L-Cereb Cx

Ant-Cereb

L-Visual A.

R-Visual A.

R-Pre-Cu

L-Pre-Cu

L-Par Cx

R-Par Cx

R-M1

L-M1

SMA

R-PM

L-PMR-DLPFC

L-DLPFC

Cing

L-Put

R-Put L-Thal

R-Thal

L-CaudN

R-CaudN

r

0

1

Example Motor network

From H. Benali

Effective connectivity : Dynamical Causal Modelling

Friston et al., NeuroImage, 2003; David et al., NeuroImage, 2006

• Neuronal variables:– Synaptic time constant

– Synaptic efficacy

– Inhibition/Excitation

– Connectivity (networks)

• Macroscopic data at the brain level:– Local field potentials

– Scalp EEG/MEG

– Functional MRI

Forward problemForward problemGiven the generative model, one can Given the generative model, one can predictpredict the measured data the measured data

Inverse problemInverse problemGiven the measured data, one can Given the measured data, one can estimateestimate the generative model the generative model

From O. David

DCM : forward models

ERP/ERF

EEG/MEG Spatial forward model g

),( xgy

data y

),,( uxfx

Dynamics f

Input u

parameters θ

states x

data y

ssignal

qdHb

s

0

,

vfout

0inf

00

0,

E

EfEf inin

vvolume

0

,

vq

vfout

f

inf

1

s

s

infflow

Ballon model

x(t)

Bold signal

),( xgy

From O. David

SI

SII SII

input

ForwardBackward

Lateral

27

.68

(1

00

%)

2.6

7 (

10

0%

)

3.57 (99%)

0.95 (53%)

DCM : Somatosensory Evoked Potential

From O. David

Limits of Brain Imaging

• Large scale observation (1 million neurons)

• Correlation between behavior and brain images no causality (necessary condition)

•Reveal only images linked to the observed task difficult generalisation, multiple parameters

• Brain : dynamical system : do not forget TIME !!!!!!!!!