1. preprocessing of fmri data fmri graduate course october 22, 2003

1. Preprocessing of FMRI Data

fMRI Graduate Course

October 22, 2003

What is preprocessing?

• Correcting for non-task-related variability in experimental data– Usually done without consideration of

experimental design; thus, pre-analysis– Occasionally called post-processing, in

reference to being after acquisition

• Attempts to remove, rather than model, data variability

Signal, noise, and the General Linear Model

MYMeasured Data

Amplitude (solve for)

Design Model

Noise

Cf. Boynton et al., 1996

Signal-Noise-Ratio (SNR)

Task-Related Variability

Non-task-related Variability

Preprocessing Steps

• Slice Timing Correction• Motion Correction• Coregistration• Normalization• Spatial Smoothing• Segmentation• Region of Interest Identification

• Bias field correction

Tools for Preprocessing

• SPM

• Brain Voyager

• VoxBo

• AFNI

• Custom BIAC scripts

Slice Timing Correction

Why do we correct for slice timing?

• Corrects for differences in acquisition time within a TR– Especially important for long TRs (where expected HDR

amplitude may vary significantly)– Accuracy of interpolation also decreases with increasing TR

• When should it be done?– Before motion correction: interpolates data from (potentially)

different voxels• Better for interleaved acquisition

– After motion correction: changes in slice of voxels results in changes in time within TR

• Better for sequential acquisition

Effects of uncorrected slice timing

• Base Hemodynamic Response

• Base HDR + Noise

• Base HDR + Slice Timing Errors

• Base HDR + Noise + Slice Timing Errors

Base HDR: 2s TR

0

0.2

0.4

0.6

0.8

1

1.2

TR:-1 TR:0 TR:1 TR:2 TR:3 TR:4 TR:5

Slice1

Base HDR + Noise

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

TR:-1 TR:0 TR:1 TR:2 TR:3 TR:4 TR:5

Noise1

Noise2

Noise3

r = 0.77

r = 0.80

r = 0.81

0

0.2

0.4

0.6

0.8

1

1.2

TR:-1 TR:0 TR:1 TR:2 TR:3 TR:4 TR:5

Slice1

Slice11

Slice12

Base HDR + Slice Timing Errors

r = 0.85r = 0.92

r = 0.62

HDR + Noise + Slice Timing

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

TR:-1 TR:0 TR:1 TR:2 TR:3 TR:4 TR:5

Slice1

Slice11

Slice12

r = 0.65

r = 0.67

r = 0.19

Interpolation Strategies

• Linear interpolation

• Spline interpolation

• Sinc interpolation

Motion Correction

Head Motion: Good, Bad,…

… and catastrophically bad

Why does head motion introduce problems?

507 89 154

119 171 83

179 117 53

663 507 89

520 119 171

137 179 117

A B C

Simulated Head Motion

Severe Head Motion: Simulation

Two 4s movements of 8mm in -Y direction (during task epochs)

Motion

Severe Head Motion: Real Data

Two 4s movements of 8mm in –Y direction (during task epochs)

Motion

Correcting Head Motion

• Rigid body transformation– 6 parameters: 3 translation, 3 rotation

• Minimization of some cost function– E.g., sum of squared differences

Effects of Head Motion Correction

Limitations of Motion Correction

• Artifact-related limitations– Loss of data at edges of imaging volume– Ghosts in image do not change in same manner as

real data

• Distortions in fMRI images– Distortions may be dependent on position in field, not

position in head

• Intrinsic problems with correction of both slice timing and head motion

Prevention is the best medicine

A B

DC

Coregistration

Should you Coregister?

• Advantages– Aids in normalization– Allows display of activation on anatomical images– Allows comparison across modalities– Necessary if no coplanar anatomical images

• Disadvantages– May severely distort functional data– May reduce correspondence between functional and

anatomical images

Normalization

Standardized Spaces

• Talairach space (proportional grid system)– From atlas of Talairach and Tournoux (1988)– Based on single subject (60y, Female, Cadaver)– Single hemisphere– Related to Brodmann coordinates

• Montreal Neurological Institute (MNI) space– Combination of many MRI scans on normal controls

• All right-handed subjects– Approximated to Talaraich space

• Slightly larger• Taller from AC to top by 5mm; deeper from AC to bottom by 10mm

– Used by SPM, National fMRI Database, International Consortium for Brain Mapping

Normalization to Template

Normalization Template Normalized Data

Anterior and Posterior Commissures

Anterior Commissure

Posterior Commissure

Should you normalize?

• Advantages– Allows generalization of results to larger population– Improves comparison with other studies– Provides coordinate space for reporting results– Enables averaging across subjects

• Disadvantages– Reduces spatial resolution– May reduce activation strength by subject averaging– Time consuming, potentially problematic

• Doing bad normalization is much worse than not normalizing

Slice-Based Normalization

Before Adjustment (15 Subjects)

After Adjustment to Reference Image

Registration courtesy Dr. Martin McKeown (BIAC)

Spatial Smoothing

Techniques for Smoothing

• Application of Gaussian kernel– Usually expressed in

#mm FWHM– “Full Width – Half

Maximum”– Typically ~2 times

voxel size

Effects of Smoothing on Activity

Unsmoothed Data

Smoothed Data (kernel width 5 voxels)

Should you spatially smooth?

• Advantages– Increases Signal to Noise Ratio (SNR)

• Matched Filter Theorem: Maximum increase in SNR by filter with same shape/size as signal

– Reduces number of comparisons• Allows application of Gaussian Field Theory

– May improve comparisons across subjects• Signal may be spread widely across cortex, due to intersubject

variability

• Disadvantages– Reduces spatial resolution – Challenging to smooth accurately if size/shape of signal is not

known

Segmentation

• Classifies voxels within an image into different anatomical divisions– Gray Matter– White Matter– Cerebro-spinal Fluid (CSF)

Image courtesy J. Bizzell & A. Belger

Histogram of Voxel Intensities

0

0.005

0.01

0.015

0.02

0.025

0.03

0.035

0.04

Anatomical

Functional

Region of Interest Drawing

Why use an ROI-based approach?

• Allows direct, unbiased measurement of activity in an anatomical region– Assumes functional divisions tend to follow

anatomical divisions

• Improves ability to identify topographic changes– Motor mapping (central sulcus)– Social perception mapping (superior temporal sulcus)

• Complements voxel-based analyses

Drawing ROIs

• Drawing Tools– BIAC software (e.g., Overlay2)– Analyze– IRIS/SNAP (G. Gerig)

• Reference Works– Print atlases– Online atlases

• Analysis Tools– roi_analysis_script.m

ROI Examples

-2

-1

0

1

2

3

4

-3-1

.5 01

.5 34

.5 67

.5 91

0.5 12

13

.5 15

16

.5 -3-1

.5 01

.5 34

.5 67

.5 91

0.5 12

13

.5 15

16

.5 -3-1

.5 01

.5 34

.5 67

.5 91

0.5 12

13

.5 15

16

.5 -3-1

.5 01

.5 34

.5 67

.5 91

0.5 12

13

.5 15

16

.5 -3-1

.5 01

.5 34

.5 67

.5 91

0.5 12

13

.5 15

16

.5 -3-1

.5 01

.5 34

.5 67

.5 91

0.5 12

13

.5 15

16

.5 -3-1

.5 01

.5 34

.5 67

.5 91

0.5 12

13

.5 15

16

.5 -3-1

.5 01

.5 34

.5 67

.5 91

0.5 12

13

.5 15

16

.5 -3-1

.5 01

.5 34

.5 67

.5 91

0.5 12

13

.5 15

16

.5 -3-1

.5 01

.5 34

.5 67

.5 91

0.5 12

13

.5 15

16

.5 -3-1

.5 01

.5 34

.5 67

.5 91

0.5 12

13

.5 15

16

.5 -3-1

.5 01

.5 34

.5 67

.5 91

0.5 12

13

.5 15

16

.5 -3-1

.5 01

.5 34

.5 67

.5 91

0.5 12

13

.5 15

16

.5

80 75 70 65 60 55 50 45 40 35 30 25 20

Distance Posterior from the Anterior Commissure (in mm)

Left Hemisphere - Gaze Shifts Right Hemisphere - Gaze Shifts

60 55 50 45 40 35 30 25 20 15 10 5 0

BIAC is studying biological motion and social perception – here by determining how context modulates brain activity in elicited when a subject watches a character shift gaze toward or away from a target.

Additional Resources

• SPM website– http://www.fil.ion.ucl.ac.uk/spm/course/notes01.html– SPM Manual

• Brain viewers– http://www.bic.mni.mcgill.ca/cgi/icbm_view/

2. Issues in Experimental Design

fMRI Graduate Course

October 23, 2003

What is Experimental Design?

• Controlling the timing and quality of presented stimuli to influence resulting brain processes

• What can we control?– Experimental comparisons (what is to be measured?)– Stimulus properties (what is presented?)– Stimulus timing (when is it presented?)– Subject instructions (what do subjects do with it?)

Goals of Experimental Design

• To maximize the ability to test hypotheses

• To facilitate generation of new hypotheses

What are hypotheses?

• Statements about the relations between independent and dependent variables.

A B C D

Psychological Hypotheses

Hemodynamic Hypotheses

Neuronal Hypotheses

Independent Variables• Aspects of the experimental design that we want to

manipulate– Often have multiple levels (e.g., experimental and control

conditions)– Critical design choice lies in determining how to choose stimuli

to match independent variableA B C

Dependent Variable: BOLD signal

-5

-4

-3

-2

-1

0

1

2

3

4

5

0 50 100 150 200 250 300

Causal and non-causal relations between variables

A B

Is the BOLD response epiphenomenal?

-8 -6 -4 -2 0 2 4 6 8 10 12 14 16

Detection vs. Estimation

• Detection: What is active?

• Estimation: How does its activity change over time?

Detection

• Detection power defined by SNR

• Depends greatly on hemodynamic response shape

SNR = aM/M = hemodynamic changes (unit)

a = measured amplitude

= noise standard deviation

Estimation

• Ability to determine the shape of fMRI response

• Accurate estimation relies on minimization of variance in estimate of HDR at each time point

• Efficiency of estimation is generally independent of HDR form

Optimal Experimental Design

• Maximizing both Detection and Estimation– Maximal variance in stimulus timing

(increases estimation)– Maximal variance in measured signal

(increases detection)

• Limitations– Refractory effects– Signal saturation

fMRI Design Types

1) Blocked Designs

2) Event-Related Designsa) Periodic Single Trial

b) Jittered Single Trial

c) Staggered Single Trial

3) Mixed Designsa) Combination blocked/event-related

b) Variable stimulus probability

1. Blocked Designs

What are Blocked Designs?

• Blocked designs segregate different cognitive processes into distinct time periods

Task A Task B Task A Task B Task A Task B Task A Task B

Task A Task BREST REST Task A Task BREST REST

PET Designs

• Measurements done following injection of radioactive bolus

• Uses total activity throughout task interval (~30s)

• Blocked designs necessary– Task 1 = Injection 1– Task 2 = Injection 2

Choosing Length of Blocks• Longer block lengths allow for stability of extended responses

– Hemodynamic response saturates following extended stimulation• After about 10s, activation reaches max

– Many tasks require extended intervals• Processing may differ throughout the task period

• Shorter block lengths allow for more transitions– Task-related variability increases (relative to non-task) with increasing

numbers of transitions

• Periodic blocks may result in aliasing of other variance in the data– Example: if the person breathes at a regular rate of 1 breath/5sec, and

the blocks occur every 10s

Effects of Block Interval upon HDR

0

2

4

6

8

10

12

14

0 20 40 60 80 100 120

0

2

4

6

8

10

12

14

0 20 40 60 80 100 120

0

2

4

6

8

10

12

14

0 20 40 60 80 100 120

0

2

4

6

8

10

12

14

0 20 40 60 80 100 120

0

2

4

6

8

10

12

0 20 40 60 80 100 120

0

1

2

3

4

5

6

7

8

9

10

0 20 40 60 80 100 120

0

1

2

3

4

5

6

7

8

0 20 40 60 80 100 120

0

1

2

3

4

5

6

7

0 20 40 60 80 100 120

40s 20s 15s 10s

8s 6s 4s 2s

What baseline should you choose?

• Task A vs. Task B– Example: Squeezing Right Hand vs. Left Hand– Allows you to distinguish differential activation

between conditions– Does not allow identification of activity common to

both tasks• Can control for uninteresting activity

• Task A vs. No-task– Example: Squeezing Right Hand vs. Rest– Shows you activity associated with task– May introduce unwanted results

Interpreting Baseline Activity

From Gusnard & Raichle, 2001

Non-Task Processing

• In many experiments, activation is greater in baseline conditions than in task conditions!– Requires interpretations of significant activation

• Suggests the idea of baseline/resting mental processes– Emotional processes– Gathering/evaluation about the world around you– Awareness (of self)– Online monitoring of sensory information– Daydreaming

From Shulman et al., 1997 (PET data)

From Binder et al., 1999

From Huettel et al., 2001 (Change Detection)

From Huettel et al., 2002 (Baseline > Target Detection)

Power in Blocked Designs

1. Summation of responses results in large variance

Single, unit amplitude HDR, convolved by 1, 2, 4 ,8, 12, or 16 events (1s apart).

HDR Estimation: Blocked Designs

Power in Blocked Designs

2. Transitions between blocks

Simulation of single run with either 2 or 10 blocks.

-1.5

-1

-0.5

0

0.5

1

1.5

2

2.5

1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 73 76 79 82 85 88 91 94 97 100

-1.5

-1

-0.5

0

0.5

1

1.5

2

2.5

1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 73 76 79 82 85 88 91 94 97 100

Power in Blocked Designs

2. Transitions between blocks

-1.5

-1

-0.5

0

0.5

1

1.5

2

2.5

1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 73 76 79 82 85 88 91 94 97 100

-1.5

-1

-0.5

0

0.5

1

1.5

2

2.5

1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 73 76 79 82 85 88 91 94 97 100

Addition of linear drift within run.

Power in Blocked Designs

2. Transitions between blocks

-1.5

-1

-0.5

0

0.5

1

1.5

2

2.5

1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 73 76 79 82 85 88 91 94 97 100

-1.5

-1

-0.5

0

0.5

1

1.5

2

2.5

1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 73 76 79 82 85 88 91 94 97 100

Addition of noise (SNR = 0.67)

Limitations of Blocked Designs

• Very sensitive to signal drift – Sensitive to head motion, especially when only a few

blocks are used.

• Poor choice of baseline may preclude meaningful conclusions

• Many tasks cannot be conducted repeatedly

• Difficult to estimate the HDR

2. Event-Related Designs

What are Event-Related Designs?

• Event-related designs associate brain processes with discrete events, which may occur at any point in the scanning session.

time

Why use event-related designs?

• Some experimental tasks are naturally event-related

• Allows studying of trial effects

• Simple analyses– Selective averaging– No assumptions of linearity required

Event-Related and Blocked Designs give Similar Results

A

B C

2a. Periodic Single Trial Designs

• Stimulus events presented infrequently with long interstimulus intervals

500 ms 500 ms 500 ms 500 ms

18 s 18 s 18 s

Trial Spacing Effects: Periodic Designs

20sec

8sec 4sec

12sec

From Bandettini and Cox, 2000

ISI: Interstimulus Interval

SD: Stimulus Duration

2b. Jittered Single Trial Designs

• Varying the timing of trials within a run

Randomization = Jittering

Dale & Buckner, 1997

Extracting different task components

A B

Effects of Jittering on Stimulus Variance

Effects of ISI on Power

Birn et al, 2002

2c. Staggered Single Trial

• By presenting stimuli at different timings, relative to a TR, you can achieve sub-TR resolution

• Significant cost in number of trials presented – Resulting loss in experimental power

• Very sensitive to scanner drift and other sources of variability

Two HDR epochs sampled at a 3s TR.

Each row is sampled at a different phase.

+0s

+1s

+2s

-0.5

0

0.5

1

1.5

2

2.5

10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50

Two of the phases are normal.

But, one has a change in one trial (e.g., head motion)

+0s

+1s

+2s

Post-Hoc Sorting of Trials

From Konishi, et al., 2000

Data from old/new episodic memory test.

Limitations of Event-Related Designs

• Differential effects of interstimulus interval– Long intervals do not optimally increase

stimulus variance– Short intervals may result in refractory effects

• Detection ability dependent on form of HDR

• Length of “event” may not be known

3. Mixed Designs

3a. Combination Blocked/Event

• Both blocked and event-related design aspects are used (for different purposes)– Blocked design is used to evaluate state-dependent

effects – Event-related design is used to evaluate item-related

effects

• Analyses are conducted largely independently between the two measures– Cognitive processes are assumed to be independent

… …

Mixed Blocked/Event-related Design

Target-related Activity (Phasic)

Blocked-related Activity (Tonic)

Task-Initiation Activity (Tonic)

Task-Offset Activity (Tonic)

Mixed designs

Donaldson et al., 2001

3b. Variable Stimulus Probability

• Stimulus probability is varied in a blocked fashion – Appears similar to the combination design

• Mixed design used to maximize experimental power for single design

• Assumes that processes of interest do not vary as a function of stimulus timing– Cognitive processing– Refractory effects

Random and Semi-Random Designs

From Liu et al., 2001

Summary of Experiment Design• Main Issues to Consider

– What design constraints are induced by my task?– What am I trying to measure?– What sorts of non-task-related variability do I want to avoid?

• Rules of thumb– Blocked Designs:

• Powerful for detecting activation• Useful for examining state changes

– Event-Related Designs: • Powerful for estimating time course of activity• Allows determination of baseline activity• Best for post hoc trial sorting

– Mixed Designs• Best combination of detection and estimation• Much more complicated analyses