fMRIfMRI Block Design and Data AnalysisBlock Design and Data Analysis

David C. Zhu, Ph.D.Cognitive Imaging Research Center

Departments of Psychology and Radiology

Reading materials

Henderson JM, Larson CL, Zhu DC. Cortical activation to indoor versus outdoor scenes: an fMRI study. Exp Brain Res. 200 1 9 842007;179:75-84.

Subsystem Subsystem Subsystemf(t) Subsystem1

Subsystem 2

Subsystem n y(t)

System with operator TSystem with operator T

y(t) = T{f(t)}

(1) Find T => Event-related design( ) g

(2) A T’ T b d d l(2) Assume T’ = T based on some modelFind expected y’(t) = T’{f(t)}Compare y(t) and y’(t)Compare y(t) and y (t) => Block related design

Linear System

( )f(t) System y(t)

g(t) System z(t)

af(t) + bg(t) System ay(t) + bz(t)

f(t) System y(t)f(t) System y(t)

f(t t ) S t y(t t )f(t-t0) System y(t-t0)

Present for 2 seconds

impulse impulse response

(t) S t h(t)

impulse impulse response

(t) System h(t)

ON ON ON

Traditional (Slow) Event Related Design

Stimuli

ON ON ON

2s 15 s 2s 15 s 2s 15 s

Response

sec

ON ON ON

2sStimuli 5s 5s 2s

Response

sec

ON

Stimuli

ResponseResponse

Time

Block Design

Present 10 pictures With 2 seconds each

Present for 2 seconds Event related design

hhph0

hm

p

p

mmhH

0

Block related designPresent 10 pictures With 2 seconds each

m 0

H

Modeling of fMRI

Mf(t) Subsystem

1Subsystem

2Subsystem

n y(t)Stimuli Measurement

System with inpulse response function (IRF) h(t)

thtfty )()()(

deconvolution

t

dthf

thtfty

0)()(

)()()(

0

h(t)

Continuous: t

dthfthtfty0

)()()()()(

Discrete times:

n

m

ttmtnhtmftny0

)()()(

0

m 0

mn

n

mn hfy In short hand,

n

mnm

mn

hf

fy

0

mm

mn hf

0

Assume hm = 0 for n ≥ p, then

m

p

mnn hfy m0

Lathi BP. Linear Systems and Signals. Berkeley-Cambridge Press. 1992.

Using the matrix notation,

Z 0...1 ffp

1p

p

ZZ

11

0

........11...1

p

p

ffpffp

X

..Z

.

...

.

...

.

.X

1

.

NZ,

11 ...11 pNN ffN ,

0

1p

p

h0

1

0

.

.

.

..

1N

ph.

p

hfZ k nmm

mnn hfZ

0

Error

k

Errormeasurement yn

Including constant baseline + linear trend, the MR signal measured

Z

npnpnn

nnn

fhfhfhnnyZ

11010

10

For n = p, p+1, …, N-1

kZ nnn kynZ 10

For n = 0, 1, …, N-1

Using the matrix notation,

0Z 01 y

1

0

ZZ

1

0

1101

yy

..

Z

X

1

.

NZ ,,

111 pNyN

1

0

..

k1

0

1

.

N

Z = X +

Error

The MR signal intensity at a voxel Baseline signal +

term

from a 7-min runBaseline signal + linear trend + IRF

0ˆ

The design matrix (when th ti l

k ZXXX tt 11

0

)(ˆ

ˆˆ

the stimulus ON and OFF)

k

h(t) Model

)()()( thtfty

Model In AFNI, gamma functions:(1) D f l C i l

)(')()(' thtfty

(1) Default: Cox special (2) Mark Cohen)(')( tkhth if

h)(')( tkyty

then

% i l hH

% signal change

ˆHk

0̂

E lExample

Henderson JM, Larson CL, Zhu DC. Cortical activation to indoor versus outdoor scenes: an fMRI study. Exp Brain Res. 2007;179:75-84.

Engineering ProcessIndoor and Outdoor belongs to different scene sub-categories. Does

Cognitive Question

the bran process them differently?

Cognitive Question

fMRI QuestionDo they have different BOLD activation at scene processing area PPA (parahippocampal place area)?

Experimental Design

Pilot Data

PPA (parahippocampal place area)? How about retrosplenial cortex?

Data AcquisitionFull1. Block design for good detection.2. Need Face as control to find PPA.3 N d l t f i ti li t

Data Analysis

Time & $

3. Need a lot of unique stimuli to find out if Indoor and Outdoor have different BOLD responses.Time & $

160 unique faces

160 unique indoor pictures

160 unique outdoor pictures160 unique outdoor pictures

160 neutral faces, 160 indoor pictures and 160 outdoor pictures.

Detailed Design

, p pNo pictures will be repeated.Block design.4 runs4 runs.Stimuli are randomized.At each run, 6 "+" initial baseline + 12 blocksAt each block, 10 pictures + 6 "+".

10 Face, Indoor or Outdoor pictures in random order

E h15 s

25 s15 s +···+ +···+

6 “+”

Each run

Data Acquisition (194 points with 2.5 s TR)4 points discarded

+ + + +

4 points discarded

8 min 15 sec

Why choose 15 sec of baseline?

25 sec ON15 sec OFF

25 sec ON12 sec OFF

25 sec ON10 OFF10 sec OFF

Other considerations: Why choose 25 sec ON and 15 sec OFF?y

1 Good duty cycle shorten the scan time for same detection power1. Good duty cycle shorten the scan time for same detection power.2. Multiple of TR (2.5 sec).3. We believe 25 sec is still good for maintaining the attention.

fMRI Design

Generate stimulus files (1D files) and the expected BOLD response based on the model.

E-Prime

expected BOLD response based on the model.

Quality Assurance

Integration Test on scanner

Data Collection

Data Analysis

Generate stimulus files (1D files) and the expected BOLD response based on the modelexpected BOLD response based on the model.

Reading Materials:Chapter 8 and Chapter 10

Data Analysis

1. Data Pre-processing:(1) Registration to AFNI( ) g(2) Slice timing adjustment(3) Motion correction(4) Spatial blurring(4) Spatial blurring(5) Mask generation

2. Data Processing:(1) Deconvolution analysis(2) Noise analysis (optional)(3) Overall significance level(4) Group analysis (ANOVA’s)(4) Group analysis (ANOVA s)(5) ROI analysis

Block Design Hand-On

• Scripts to generate a block-design experiment

• Run the script to do the data analysis

E l i h i• Explain the scripts.

Designrandomize.screateWave.s

TS1_analy.1D, TS2_analy .1D, TS3_analy.1D, TS4_analy.1D

indoor_TSall_wave.1Doutdoor_TSall_wave.1D

face TSall wave 1Dface_TSall_wave.1D concat.1D

E-Prime Programming

register s: register many image files to AFNI formatCollect Data register.s: register many image files to AFNI format.3dTshift: slice timing adjustment3dvolreg: motion correction3dTcat: concatenate all fMRI data together3dmerge: spatial blurring

analyze_ts_blur4fwhm_group1.s3dmerge: spatial blurringmask generation: identify brain region with 3dcalc or 3dAutomaskLink design matrix (*_TSall_wave.1D, concat.1D)3dDeconvolve (deconv2.s): Compare fMRI image data with design matrix

decon_Scene_Face_motion (the “bucket”)

PerSigChan.s: calculate % signal change

group_ANOVA2.s: group analysis from all subjects

MonteCarlo moreIterfwhm4 s l (3d l d 3d ) l l iMonteCarlo_moreIterfwhm4.s (AlphaSim)

gen_cluster.s (3dclust and 3dmerge): cluster analysis

Final results