biau brain image analysis unit biostatistics & computing institute of psychiatry london, u.k....

54
B Brain I Image A Analysis U Unit Biostatistics & Computing Institute of Vincent P. Vincent P. Giampietro Giampietro [email protected] ntroduction to ntroduction to fMRI Analysis fMRI Analysis

Upload: adrian-miller

Post on 28-Mar-2015

218 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

BBrain IImage AAnalysis UUnitBiostatistics & ComputingInstitute of PsychiatryLondon, U.K.

Vincent P. GiampietroVincent P. [email protected]

Introduction toIntroduction to

fMRI AnalysisfMRI Analysis

Page 2: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

LondonLondon

Page 3: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

LondonLondon

Page 4: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

LondonLondonThe Institute of Psychiatry is in Camberwell…

Page 5: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

LondonLondonThe Institute of Psychiatry is in Camberwell…

Page 6: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

My goalsMy goals

To give you a good idea of what fMRI analysis really does

To fight against the black box way of analysing fMRI datasets

Without showing you any of these:

)())(,,,,

(0,,

)())((1),,(

,,,,

1

,,,,,,22

)3()2()1(

xtUxxDtUxT

Lu

uuuJmlk

j

L

jk x

T

Page 7: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

The fMRI challengeThe fMRI challenge

In fMRI, the signal change due to activation (BOLD effect) is very subtle: it amounts to about less than 4% of the baseline signal at 1.5T (double that at 3T)The challenge is to detect a small signal embedded in background noisefMRI analysis is a digital signal processing problemSome of the analysis methods are directly adapted from other signal processing domains, such as voice recognition (wavelet transforms)

Page 8: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

How large is a 1.5T magnetic field?How large is a 1.5T magnetic field?Roughly the same as

More or less 50000 times

Page 9: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

A simple fMRI experimentA simple fMRI experiment

39s

30s

AU

DIO

VIS

UA

L

5mn

3s (TR)

3D brain volume

16 7.7mm thick slicesmatrix size = 64x64

1 image voxel

Voxel=pixel in 3D

7.7mm

3.75mm

3.75mm

64 voxels

64 v

oxel

s

10 volumes

Multiplex audio-visual (e.g. internal control check for global changes in drug trials)

Page 10: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

The dataThe data

One slice100 images

One image4096 voxels

Voxels

64 voxels

64 v

oxel

s

3.75mm

3.75

mm

1 TR = 3st=1xTR=3s

t=100xTR=300s

Page 11: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

fMRI analysisfMRI analysis

Getting the images from the MR scannerPre-processing the raw dataAnalysing a single subject experimentAnalysing a group of subjectsComparing different groupsUsing more advanced analysis methods

I have scanned a subjectWhat should I do next?

How do I get the red blobs?

Page 12: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Getting the data from MRGetting the data from MRThe grand image format debate

Public server

your Sun/PC

Soon - DICOM formatDDigital IImaging and COCOmmunication in MMedicine

Now - Native MR scanner format

Soon - NIfTI formatNNeuroimaging IInfformatics TTechnology IInitiative

MR scanner

Now - Analyze format

The files are anonymised

http://nifti.nimh.nih.gov/

http://medical.nema.org/http://www.psychology.nottingham.ac.uk/staff/cr1/dicom.html

Page 13: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Pre-processing the raw dataPre-processing the raw dataDo we need it ?

Without With

Really bad dataset with huge head motion

Simon Surguladze

Page 14: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Pre-processing the raw dataPre-processing the raw dataraw data

movement correction

detrending

smoothing

preprocessed data

1

2

3

Page 15: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Pre-processing the raw dataPre-processing the raw dataMovement correction

The two main movement artefacts in fMRI are:

0

0.005

0.01

0.015

0.02

0.025

0.03

0.035

0.04

1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96

Time (TR)

Mot

ion

corr

ecte

d (v

oxel

)

X Rotation

Experimental model

stimulus correlatedbetween scans

Rozmin Halari

Page 16: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Pre-processing the raw dataPre-processing the raw dataMovement correction (co-registration)

Rigid body realignment (3 translations + 3 rotations)Registered images written by tricubic spline/linear interpolation

3D average registration template

Page 17: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Pre-processing the raw dataPre-processing the raw data

Stimulus correlated motion is fitted as “activated” by the modelWithout motion and spin correction, the results are useless

Stimulus correlated motionAnalysis only

Motion correction + analysis

Motion correction + spin excitation history correction + analysis

Page 18: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Pre-processing the raw dataPre-processing the raw dataStimulus uncorrelated motion

Stimulus uncorrelated motion doesn’t “mess up” the results

Motion and spin correction increase the power the fMRI analysis

Motion correction + spin excitation history correction + analysis

Analysis only

Motion correction + analysis

Page 19: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Movement-related autocorrelationIn the magnet, the positions of the nuclei at time t are spatially and temporally related to the positions of the same nuclei at time t-1 (and actually up to t-3)Can be corrected by using autoregressive pre-whitening

Pre-processing the raw dataPre-processing the raw dataDetrending – Spin excitation history correction

Without With

Page 20: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Problem in space

Magneticfield

Problem in time

Pre-processing the raw dataPre-processing the raw dataDetrending – Spin excitation history correction

Page 21: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Pre-processing the raw dataPre-processing the raw dataDetrending – Scanner drift

Bef

ore

Aft

er

Linear trend

Non-uniformity in the magnetic field

Electronic interferences due to temperature fluctuations in the imaging hardware

time

Page 22: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Pre-processing the raw dataPre-processing the raw data

Ray Norbury

Non-linear but periodic trends

Easily “filtered out” using high/low/band-pass filters

Detrending – Other trends

Page 23: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Pre-processing the raw dataPre-processing the raw dataSmoothing – Spatial filtering

Digital filtering

Filter kernel Original Image

X Y f(X)f(Y)

Filtered Image

= convolution (~the image is multiplied by the filter kernel)

f( )

Page 24: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Pre-processing the raw dataPre-processing the raw dataSmoothing – Spatial filtering

Mean filter

ΣNew value = 1/9 x Pixeli,j x Filteri,j

111

111

111

3x3 mean filter

1125 1014 850

1310 1243 1138

1315 1338 1282

1138

(convolution)1180

126

Page 25: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Gaussian filter

2D Gaussian distribution (mean (0,0) and σ=1)

Discrete approximation to Gaussian function with σ =1.0

Full Width at Half MaximumFWHM = 2.355 x σFWHM = 2.355 voxelsVoxel size = 3.75x3.75mmFWHM = 8.83 mm

standard deviation

Pre-processing the raw dataPre-processing the raw dataSmoothing – Spatial filtering

Page 26: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

The problem…– To maximise the effect, the size of the filter should

match the size of the activated regions in the image– But brain structures come in many different sizes and

shapes so smoothing the images may do more harm than good…

To smooth or not to smooth?– Adaptive (steerable) filtering (CCA - CCanonical CCorrelation

AAnalysis)– No smoothing at all… – But it is worth remembering that some analysis packages

require smoothing for their statistical analysis to work…

Pre-processing the raw dataPre-processing the raw dataSmoothing – Spatial filtering

Page 27: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Time

Pre-processing the raw dataPre-processing the raw dataSmoothing – Temporal filtering

Moving average filter

8-point low pass

Page 28: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Analysing a single subject experimentAnalysing a single subject experimentThe model file – Experiment description

39s

30s

AU

DIO

VIS

UA

L

5mn

0000000000

1111111111

0000000000

1111111111

0000000000000

1111111111111

0000000000000

1111111111111

…0 00 00 00 00 00 00 00 00 00 01 01 01 01 11 1…

…0.501 14.56052.517 25.50998.517 39.509110.517 62.517120.517 72.517133.509 84.517172.501 146.501247.525 158.501259.525 186.501269.525 197.509280.517 211.509…

Model file for block design

Model file for event related design

t=0.501s t=14.560s

1 TR (3s)

Page 29: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Analysing a single subject experimentAnalysing a single subject experiment

4s 8s

Gamma Variate Kernels

The model file – Experiment description

Experiment

Real BOLD response

or- Physiological models (Balloon

model)- Adaptive models (GLM extensions)

- Model free analysis (ICA)

- 1 Gamma function & its 1st derivative

Page 30: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Analysing a single subject experimentAnalysing a single subject experimentThe model file – Experiment description

(convolution)4

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

Page 31: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Analysing a single subject experimentAnalysing a single subject experimentModel fitting

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

Model for the visual stimulation

600

620

640

660

680

700

720

740

760

1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97

Real time series from the visual cortex

790

800

810

820

830

840

850

860

870

880

890

1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97

Real time series from the auditory cortex

The model is usually fitted using least square fitting

Page 32: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Analysing a single subject experimentAnalysing a single subject experimentModel fitting – Good fit (1 gamma function)

Real time series Fitted model

Page 33: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Analysing a single subject experimentAnalysing a single subject experimentModel fitting – Bad fit (1 gamma function)

Page 34: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Calculate a goodness of fit statistic– For each pixel– For each condition

This generates statistical maps of the brain (one per condition and per interaction)Null hypothesis– There is no experimental effect– There is no relationship between the voxel time series and

the experimental model

How do you decide if your statistics are significant or not ?– Parametric statistics (lots of assumptions…among other

things the data need to have a Normal distribution)– Non parametric statistics (distribution-free)

Analysing a single subject experimentAnalysing a single subject experimentModel fitting

Page 35: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Non parametric statistics (the way we do it)– Statistic used: Sum of Square Quotient (SSQ)

– SSQ = ratio of model to residual sum of squares

Analysing a single subject experimentAnalysing a single subject experimentModel fitting

iii

ii

fittedTSrealTS

fittedTSSSQ 2

2

– Use randomisation testing to determinate the p value of the statistic ( this is the non-parametric bit)

– If p< α then the null hypothesis is rejected, there is a statistically significant relationship between the experimental model and the studied voxel time series

– The voxel is activated and gets coloured

Page 36: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Analysing a single subject experimentAnalysing a single subject experiment

Randomisation tests ???

Statistical tests in which the data are repeatedly mixed

A test statistic is computed for each data shuffle

The proportion of data divisions with as large a test statistic value as the value for the original results determinates the significance of the results

Computer intensive and memory hungry…

(E.S. Edgington 1995)

Page 37: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

A simple example– 2 treatments A and B– Hypothesis: A measurements > B measurements

– 4 patients a, b, c and d

Analysing a single subject experimentAnalysing a single subject experiment

A B Patient Measurement Patient Measurement a 6 b 3 c 7 d 4 A=6.5 B =3.5

t=4.24

Randomisation tests ???

Page 38: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

A Simple example– 6 permutations possible of the patients to form 2 groups

– We calculate the t statistic for every permutation

Analysing a single subject experimentAnalysing a single subject experimentRandomisation tests ???

– None of the permutations have a statistical value higher or equal to 4.24 (the statistical value for the real situation).

– The one-tailed significance (p value) associated with the obtained results is therefore 1/6=0.167

A B A B A B A B A B A B 3 6 3 4 3 4 4 3 4 3 6 3 4 7 6 7 7 6 6 7 7 6 7 4 X 3.5 6.5 4.5 5.5 5 5 5 5 5.5 4.5 6.5 3.5 t -4.24 -0.47 0 0 0.47 4.24

real (observed)situation

Page 39: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Analysing a single subject experimentAnalysing a single subject experimentCluster analysis

Clustering– Connects activated voxels from the same brain structure– Can reinforce sub-threshold activations by “pushing them

to the surface” and eliminates single activated voxels

Levels of clustering– Per slice (2D)

– Per volume (3D)

– In time (4D)

You get the idea !

Page 40: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Analysing a single subject experimentAnalysing a single subject experimentThe resultsA

UD

IOV

ISU

AL

Page 41: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Interlude…

Page 42: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Interlude…

Page 43: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Spatial NormalisationSpatial Normalisation

What is it?– Process of transforming an image for an individual

subject to match a standard brain or brain template

What do we want to do with it ?– Check the activations on the standard atlas (functional

localisation)– Compare groups of subjects

How do we do it ?– Mostly by using automatic warping methods

Page 44: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Talairach atlas (Talairach and Tournoux)– “Co-Planar Stereotaxic Atlas of the Human Brain” (1988)– Detailed atlas of brain sections with a coordinate system and Brodmann

regions– “Proportional grid” of brain imaging– But…made from the post-mortem brain of a 60-year old alcoholic french

woman

MNI/ICBM templates– MMontreal NNeurological IInstitute / IInternational CConsortium of BBrain

MMapping– Average of hundreds of brains– 241 brains were manually scaled to the Talairach brain to produce an

temporary template– MNI305 is made of 305 brains mapped to this template– ICBM152 is made of 152 brains registered to MNI305 (current template)– No Brodmann regions…

Spatial NormalisationSpatial NormalisationBrain templates

Page 45: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Spatial NormalisationSpatial NormalisationTalairach mapping

fMRI

1st registration

Structural space

2nd registration

Talairach space

High-resolutionstructural image

Talairach template

Page 46: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

More smoothing if needed…

More statistical analysis…

More cluster analysis…

More pretty pictures…

Spatial NormalisationSpatial NormalisationTalairach mapping

but…

Page 47: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

The resultsSpatial NormalisationSpatial Normalisation

Strongest activated cluster– 64 voxels.– Talairach coordinates: x=-55, y=-14, z=9 (activation focus).– Left side, slice 10.– Brodmann Area 42, Auditory Association Cortex

Page 48: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Spatial NormalisationSpatial NormalisationThe results

Page 49: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Spatial NormalisationSpatial Normalisation

2D

The results

Page 50: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

The results in virtual reality

Spatial NormalisationSpatial Normalisation

Page 51: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Analysing groups of subjectsAnalysing groups of subjects

Individual analysis

Group mapping

ANCOVA

differences

similarities

Individual analysis

Group mapping

Page 52: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

Using more advanced analysis methodsUsing more advanced analysis methods

Improved registration/warping (e.g. non-linear)Constrained BOLD fit / Wavelet denoisingNew randomisation methods (e.g. cyclic wavelet permutation of the residuals)Trend analysisPath analysisConnectivity analysisTime series extractionCorrelation/Partial Correlation analysisReal time fMRI analysisCombined EEG/fMRI…

Page 53: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

ConclusionConclusion

Most of the analysis packages give you robust semi-automated methods of fMRI analysis “from MR to ANCOVA and much more…”The analysis process can look like a black box but we make our best to try to explain what we are doingIn all the labs, there is constant work on improving and validatingvalidating the existing methods and on writing the next ones…Please, be patient and understanding !!! (especially if you are a beta tester…)

Page 54: BIAU Brain Image Analysis Unit Biostatistics & Computing Institute of Psychiatry London, U.K. Vincent P. Giampietro V.Giampietro@iop.kcl.ac.uk Introduction

BBrain IImage AAnalysis UUnitBiostatistics & ComputingInstitute of PsychiatryLondon, U.K.

Vincent P. GiampietroVincent P. [email protected]

Introduction toIntroduction to

fMRI AnalysisfMRI Analysis