introduction to neuroimaging

Sunghyon [email protected]

Institute of Behavioural Science in Medicine, Yonsei University College of Medicine

Introduction to Neuroimaging -PET, fMRI, VBM, and DTI-

mailto:[email protected]

Sunghyon Kyeong (Yonsei Univ) Introduction to Neuroimaging: Methods and Preprocessing steps p 2

with Ctrl-key, select multiple regions

with Ctrl-key, select multiple regions

Outline• Positron Emission Topography (PET) Imaging • Principles of BOLD signal generation • Review on fMRI preprocessing steps • Functional Network Construction • Morphometric Brain Network • Network from Diffusion Tensor Imaging

Positron Emission Tomography:Two photo detector


Positron Emission Tomography

gamma raydetectors Unstable parent

nucleus

Proton decays to neutron in positron and neutrino emitted

Positron combines with electron and annihilates

Two anti-parallel 511 keVphotons produced

p� n + �+ + ⇥ebeta decay process :

NaI(Tl), bismuth germanate oxide (BGO), gadolinium oxyorthosilicate (GSO), lutetium oxyorthosilicate (LSO) are used for the crystal.


Coincidence Detection


Types of Coincidence Events

• A scattered coincidence is one in which at least one of the detected photons had undergone at least one Compton scattering event prior to detection

• Random coincidence occur when two photons not arising from the same annihilation event are incident on the detectors with the coincident time window of the system


• Unstable positron-emitting isotopes are synthesised in a cyclotron by bombarding elements such as oxygen, carbon, or fluorine with protons.

• Isotopes : 15O(half-life 2min), 18F(110 min), 11C(20min) • When the radio-labeled compounds are injected into the blood

stream, they distribute according to the physiological state of the brain, accumulating preferentially in more metabolically active areas.

• The structure of F-18-FDG is similar to the glucose, so it can used to diagnosis the abnormality of glucose metabolism.

Isotope in PET imaging

Blood Oxygen Level Dependent Signal for

functional MRI


2D iFFT

Raw Data

k-Space Image

Complex Data in Image Domain

M = |R + iI|

P = tan�1(I/R)

fMRI Data Acquisition


Detection of MRI Signal• Applying RF pulse to tip down bulk magnetisation (Mz) to

the transverse plane. • Mz tends to align the external magnetic field as time goes

on (T1 recovery). • Mz decays in the transverse plane as time goes on (T2

decay).

Good Contrast Good Contrast

B0

MR scan

ner

magnetic field due to solenoid


Tissue T1 (ms) T2 (ms) Gray matter (GM) 950 100

White matter (WM) 600 80

Muscle 900 50

Cerebrospinal fluid (CSF) 4500 2200

Fat 250 60

Blood 1200 100~300

Tissue Specific T1 and T2

B0 = 1.5 T

T = 37�C

obtained at

• T1 recovery and T2 decay time ranges from tens to thousands of milliseconds for protons in human tissue over the main field. Typical values for various tissues are shown in following table.

• Applying the pulse sequences, we can discriminate brain tissues; The different sequences should be applied to obtain the specific image, for example, anatomic, functional, angio images.


• The abbreviation BOLD fMRI stands for Blood Oxygen Level Dependent functional MRI.

• The BOLD contrast mechanism alters the T2* parameter mainly through neural activity–dependent changes in the relative concentration of oxygenated and deoxygenated blood.

• Deoxyhemoglobin is paramagnetic and influences the MR signal unlike oxygenated hemoglobin.

Detecting BOLD fMRI Signal


Contrast Agents for fMRI ?• Definition : Substances that alter magnetic susceptibility of tissue of

blood, leading to changes in MR signal - Affects local magnetic homogeneity: decrease in T2*

• Two types- Exogenous : Externally applied, non-biological compounds.- Endogenous : Internally generated biological compound (e.g., dHb)

• BOLD functional magnetic imaging method doesn’t need the external contrast agents.


O2 Ratios in Blood

High ratio deoxy :→ deoxygenated blood → fast decrease in MRI signal

Low ratio deoxy :→ oxygenated blood → slow decrease in MRI signal

Normal blood flow High blood flow

BOLD signal =HB

dHB

dHbHb

deoxyhemoglobin (paramagnetic) oxyhemoglobin (non-magnetic)

• BOLD contrast measures inhomogeneities in magnetic field due to changes in the level of O2 in the blood.


Mechanism of BOLD fMRI

Time

BO

LD s

igna

l T2* task

T2* control

TEoptimal

ΔS

↑ Neural Activity ↑ Blood Flow ↑ Oxyhemoglobin

↑ T2*

↑ MR Signal

Sunghyon Kyeong (Yonsei Univ) Introduction to Neuroimaging: Methods and Preprocessing steps p

Hemodynamic Response

16

BOLD

Sig

nal C

hang

e

Time (second)

0 5 10 15 20

• BOLD signal은 자극 이 제시되고 5~6초 후에 최대 반응을 보임

• Fast event related + jittered ISI is the optimal design

Reference for FMRI Experimental Design, http://afni.nimh.nih.gov/pub/dist/HOWTO/howto/ht03_stim/html/stim_background.html

http://afni.nimh.nih.gov/pub/dist/HOWTO/howto/ht03_stim/html/stim_background.html


Block Designed fMRI

MR

I

Language Area Motor Area


Resting State fMRI

• Resting state fMRI measures “low-frequency (0.01~0.08 Hz)” slow oscillation. • Resting state means “Keep eyes closed resting state but not sleep for

several minutes”. • Resting state functional connectivity considered as “intrinsic connectivity”. • Modular structure in RSFC were found in many studies. • Default mode network (DMN) alteration in Psychiatric patients (e.g.

schizophrenia).

18

steps in the spatial preprocessing

fMRI preprocessing


Summary of PreprocessInput Output

EPI1.niiEPI2.nii

…

aEPI1.niiaEPI2.nii

…

aEPI1.niiaEPI2.nii

…

meanaEPI.nii aEPI1.nii (realigned) aEPI2.nii (realigned)

rp_EPI.txt …

meanaEPI.nii anat.nii

meanaEPI.nii anat.nii (coregistered)

anat.nii aEPI1.niiaEPI2.nii

…

wanat.nii waEPI1.nii waEPI1.nii

…

waEPI1.niiwaEPI2.nii

…

Slice Timing

Realignment

CoregistrationT1 → meanEPI

Normalisation

Smoothing

Event related fMRI analysis

Resting state fMRI analysis

Preprocessing• Specify 1st-level in SPM

Individual GLM with Stimulus onset and rp_EPI.txt as regressors

• Specify 2nd-level in SPMGroup-wise GLM analysisone sample, two sample, factorial design, flexible design

• Linear detrending of EPI time series at each voxel.

• bandpass filtering (0.009~0.08Hz) to capture Low-frequency fluctuation

• regression nuisance parameters such as head motion, white matter, ventricle, and global signal

• Functional connectivity analysis and Complex network analysis

swaEPI1.niiswaEPI1.nii

…


Realignment

21

...

motion parameters mean-fMRI

sagittal

coronal

axial

100 dynamic images


Coregistration

22

Bef

ore

Cor

egA

fter C

oreg

• High Resolution T1 data is registered to mean-fMRI

• Rigid-body transformation only (translation & rotation)

T1 mean-‐fMRI

T1 mean-‐fMRI


coregistered T1 T1 template

normalized T1 (wT1)

fMRI images

...

...

normalised fMRI (wfMRI) images

...

smoothed fMRI (swfMRI) images

Nonlinear normalisation (T1→Template)

w

w

spatial gaussian ?ilter (FWHM=6 or 8mm)

S

Normalisation and Smoothing

Resting State Functional Connectivity

Michael D. Fox (2005) PNAS

Seed-ROI based connectivity analysis Graph theoretical analysis

fMRI preprocessingsteps in the temporal preprocessing


0 100 200 300 400 500 600 700 800720

730

740

750

760

770

780

790

time course at voxel i (before linear detrending)

increasing trend due to heat

0 100 200 300 400 500 600 700 800−25

−20

−15

−10

−5

0

5

10

15

20

25

after detrending (i.e. removing long term increasing trend) time course with linear function

Linear Detrending


Nuisance parameter regression

27

0 200 400 600 800

YGS

YCSF

YWM 0 200 400 600 800

−0.8

−0.6

−0.4

−0.2

0

0.2

0.4

0.6

0.8

x translation y translation z translation

0 200 400 600 800−0.02

−0.015

−0.01

−0.005

0

0.005

0.01

0.015

0.02

pitch roll yaw

GM WM CSF

Tx

Ty

Tz Rx

Ry

Rz

0 50 100 150 200 250 300 350 40065

70

75

80

85

90

0 50 100 150 200 250 300 350 400−10

−5

0

5

10

Volume (inter-‐volume interval = 2 sec)

Y = β1Tx + β2Ty + β3Tz + β4Rx + β5Ry + β6Rz + β7YGS + β8YCSF + β9YWM + ε

Head motions were regressed out to remove spin-history artefact.

Before After


0 0.05 0.1 0.15 0.2 0.250

100

200

300

400

500

600

Bandpass Filtering Region (0.01 -‐ 0.08 Hz)

Bandpass Filtering

28

0 50 100 150 200 250 300 350 400−4

−3

−2

−1

0

1

2

3

4

Bandpass Ailtering (0.01-‐0.08 Hz) : removing vary slow wave, cardiac & respiratory noise

• very low frequency regions are related to drift (<0.01 Hz)

• high frequency regions are related to respiratory & cardiac noise

Frequency (Hz)

Volume (inter-‐volume interval = 2 sec)


Functional Connectivity

29

0 50 100 150 200 250 300 350 400−30

−20

−10

0

10

20average time course within a node

computing the pair-wire correlation coefficients for functional connectivity

AAL atlas

weighted undirected

Adjacency Matrix (Aij)

Thresholding

Graph

Graph Theory and Matrix


Types of Graph

Binary Undirected

BinaryDirected

Weighted Directed

1

3

6

5

2

4

0 1 1 0 0 01 0 1 0 1 01 1 0 0 0 00 0 0 0 1 00 0 0 1 0 10 0 0 0 1 0

Aij =Matrix

k2 = 3k3 = 2k4 = 1

Degree


Graph Visualisation

degree strengthclustering coefficientnode betweenness centralitynode efficiency

edge strengthedge betweenness centrality

modular architecture

Network Properties

Node Properties

Edge Properties

Modular Structure

Network Visualisation

계산된 네트워크의 노드 속성값을가시화 과정에서 노드 크기로 표현함.

계산된 네트워크의 엣지 속성값을가시화 과정에서 엣지의 두께로 표편함.

계산된 네트워크의 모듈구조를가시화 과정에서 노드의 색깔로 표현함.

1

2

Morphometric Brain Network

Hippocampus

Posterior Hipp

time as taxi driver (month)

adju

sted

VBM

resp

onse

spo

ster

ior h

ippo

cam

pus

ante

rior h

ippo

cam

pal c

ross

-se

ctio

nal a

rea

(mm

2 )

Posterior Hipp

Anterior Hipp

Taxi drivers' brains 'grow' on the job

Maguire, E.A. (2000) PNAS


1. Tissue segmentation2. Create Template & Normalisation3. Modulation4. Smoothing5. Network Construction

The data are pre-processed to sensitise the statistical tests to *regional* tissue volumes

Analysis Steps

Voxel-based Morphometry (VBM)


Segmentation

Probability maps

Mixture model

CSF GM WM

• Individual T1 weighted images are partitioned into- grey matter / white matter / cerebrospinal fluid

• Segmentation is achieved by combining with- probability maps / Bayesian Priors (based on general knowledge about

normal tissue distribution)- mixture model cluster analysis (which identifies voxel intensity

distributions of particular tissue types in the original image)

GM WM CSF

T1 weighted image


Modulation

* Jacobian determinants of the deformation field

• Is optional processing step but tends to be applied

• Corrects for changes in brain VOLUME caused by non-linear spatial normalisation

• Multiplication of the spatially normalised GM (or other tissue class) by its relative volume before and after warping*, i.e. IB = IA×(VA/VB).


Example

IB = ?IA = 1 VA = 1 VB = 2

IA = 1 VA = 4

IB = ?VB = 2

Template

Signal intensity ensures that total amount of GM in a subject’s temporal lobe is the same before and after spatial normalisation and can be distinguished between subjects

TemplateIB = 1 × [1 / 2] = 0.5

IB = 1 × [4 / 2] = 2

Modulation

ModulationNormalisation

Normalisation

IB = IA × [VA / VB]

Larger Brain

Smaller Brain


What is GM density• The exact interpretation of GM concentration or density is

complicated.

• It is not interpretable as (i) neuronal packing density or (ii) other cytoarchitectonic tissue properties, though changes in these microscopic properties may lead to macro- or mesoscopic VBM-detectable differences.

• Modulated data are more “concrete”.


Age, VTIV

ROI index (i)

Subject index (j)

After Regression

MijMij is a GMV for a Subject i and ROI j

−1 −0.5 0 0.5 10

200

400

600

800

1000

1200

What’s the meaning of positive and negative associations in the morphometric network?

ROI Based MorphometryRegressors

Adjacency Matrix (Aij) Distribution of Correlation Values

Morphometric network is a part of structural network, and representing group level network.

Diffusion Tensor Imaging


6 directional encoding b=0

Tensor

FA

Construct Structural Network Fiber

Tracking:

DT-MRI

+

+


DTI & Tractography• Diffusion Tensor: at least 6 directional DWIs + non-DWI are required.

• Diagonalization using Singular Value Decomposition

D =

0

@D

xx

Dxy

Dzy

Dyx

Dyy

Dyz

Dzx

Dzy

Dzz

1

A

D = (e1 e2 e3)T

0

@�1 0 00 �2 00 0 �3

1

A (e1 e2 e3) =3X

k=1

�kekeTk


Useful Quantities• Mean ADC (apparent diffusion coefficient)

• FA (fractional anisotropy)

• PDD (principal diffusion direction) what direction is greatest diffusion along? the orientation of finer tract

Trace(D) = hDi = �1 + �2 + �3

3

FA =

p(�1 � �2)2 + (�2 � �3)2 + (�3 � �1)2p

2p

�21 + �2

2 + �23

introduction to neuroimaging

Documents

preprocessing steps

neuroimaging pet

unstable positron

coincidence detection

p n ebeta decay process

detection of mri signal

neutrino emitted positron

detection random coincidence