Preprocessing

1
Preprocessing
Katherine Aumer-Ryan MRI Methods Class Fall
2006 References of images are noted at end of
slideshow. All other images are from Huettel,
Song and McCarthy (2004). Functional Magnetic
Resonance Imaging
2
Why preprocess?
--Unwanted (non-task) variability from -head
motion -physiological changes -inhomogeneities
in static field -differences in timing
acquisition
Goals of preprocessing
--Increase Functional SNR --Prepare data for
statistical analyses. --All accomplished through
removal and NOT remodelling.
3
Increase Functional SNR
Task-related variability
Non-task-related variability
4
Quality Assurance
Purpose --Identify problems with the scanner
or session. Identify artifacts.
Benefits --Maintain quality of images and
analyses. --Prevent discarding data.
Methods --Observe scans (during
session). --Raw SNR (during session). --Phantom
processing (pre-session).
5
Online Qualitative Assurance Methods
--Interocular Trauma Test
6
Pre-session Quality Assurance Methods
--Phantoms
7
Pre-session Quality Assurance Methods
--Phantoms
Normal
8
Slice Acquisition Errors
Causes --Acquiring slices at different times
within the TR (problem for both interleaved and
ascending/descending acquisition).
Consequences --Inaccurate representation of the
HDR.
9
Slice Acquisition Time Correction
Purpose --Eliminate differences in HDR across
slices due to acquiring those slices at
different times.
Benefits --Provides an accurate representation
of the HDR.
Method --Temporal Interpolation (linear,
spline, and sinc). --Interleaved slices (done
before MC)
10
Slice Acquisition Time Errors
--Interleaved Slice Acquisition --Activation 15
16 17 --3 s TR
11
Interpolation
12
Interpolation
Slice 15 Slice 16 Slice 17
Hemodynamic response (arbitrary units)
TR1 TR2 TR3 TR4 TR5
13
Head Motion Variability
Causes --Participant fatigue. --Startle. --Ta
sk-related (talking).
Consequences --Displacement of
voxels. --Misestimate timing. --Some spins do
not get excited. --Lost data.
14
Head Motion Correction
Purpose --Reduce/remove variability caused by
head motion (translation or rotation).
Benefits --Provides an accurate representation
of signal intensity.
Methods --Prevention. --Coregistration. --Mode
ling (covariate).
15
Head Motion Prevention
16
Head Motion Correction using Coregistration
Coregister --spatially align two images.
Problem image is coregistered with a reference
volume. Usually Rigid Body Transformation. Minim
ize a cost-function -sum of the square
residuals -Mutual information Iterative procedure
Estimate new values --Spatial interpolation
use more than one dimension to derive
appropriate values.
17
Head Motion Coregistration
Motion
Two 4s movements of 8mm in Y direction (during
task epochs)
18
Head Motion Coregistration
Corrected
19
Head Motion Coregistration in action
--Rigid-body (6 DOF - translation and rotation
only) --Each iteration alters the parameters
slightly so that the resulting image looks more
and more like the target --Done for each volume
separately
20
Distortion
Causes --Static field inhomogeneities
(geometric distortion). --Excitation field
inhomogeneities (intensity distortion).
Consequences --Distorted representation of the
anatomy and the function of the brain.
21
Distortion Correction
Purpose --Reduce inaccurate representations.
Benefits --Accurate representation of anatomy
and function.
Methods --Shimming Coils. --Magnetic Field
Mapping (higher field strengths). --Biased
Field Estimation.
22
Magnetic Field Mapping
Field Map --Reduce inaccurate representations.
TE 30ms
TE 50ms
Field Map
Reconstruction Routine
Corrected Image
23
Bias-Field Estimation
Bias-Field Estimation --Using the distorted
image to estimate what the amount of bias or
distortion and then covariate this out to
create the accurate image.
Corrected
Distorted
Total Magnetic Field Estimation
24
Non-Normalized Non-coregistered Data
Causes --Differences in functional and
anatomical resolution. --Intersubject
differences in brain shape/size.
Consequences --Inability to locate and compare
appropriate areas of activation.
25
Coregistration
Purpose --Map low resolution functional
images onto high resolution anatomical images.
Benefits --Accurate representation of anatomy
and function.
Methods --Rigid Body-Transformation.
26
Coregistration
Functional
Anatomical
Coregistered
27
Normalization
Purpose --Fit brain images of various size and
shape into one standardized shape.
Benefits --Accurate representation of anatomy
and function.
Methods --Stretching, squeezing, and pulling to
fit these maps -Talairach or stereotaxic
space -MNI Template
28
Talairach
--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
29
MNI
--Montreal Neurological Institute (MNI)
space --Combination of many MRI scans on normal
controls --All right-handed subjects --Approximate
d 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
30
Other uninteresting variation
Causes --Known variations with estimated
frequency or amplitude. Non- interesting noise.
Consequences --Increases noise, therefore
decreases SNR.
31
Filtering
Purpose --Retains data that occur at the task
frequency, minimizes changes occurring at other
frequencies.
Benefits --Prepares data for better statistical
analyses. --Reduces likelihood of Type I
errors. --Also reduces spatial resolution.
Methods --Filtering or smoothing.
32
Temporal Filtering
Low-Pass Retains low frequencies, attenuates
high frequencies. Allows low- frequencies to
pass. Good for removing physiological
responses. -Heart-rate 1-1.5 HZ (cycles
per second) -Breathing .2-.3 HZ
-Stimulus .04 HZ (lowest)
33
Spatial Filtering

• Gaussian
• Spreads activation to fit normal distribution.
  Thus intensity from one voxel will be spread to
  neighboring voxels.
• -Usually expressed in mm FWHM
• (Full Width Half Maximum)
• -Typically 2 times voxel size

Matched filters Match the frequency of
interest. Works best.
34
Spatial Filtering
Unsmoothed Data
Smoothed Data (kernel width 5 voxels)
35
Summary
Preprocessing --Done BEFORE statistical
analyses. --Reconstructing K-space
data. --Primary purpose increase SNR. --Many
methods, should be done in specific order.
36
References
http//www.brainvoyager.com/BV2000OnlineHelp/Brain
VoyagerWebHelp/TalairachBrain2.gif
http//www.biac.duke.edu/education/courses/fall03/
fmri/handouts/W9_Preprocessing_Design_2003.htm
37
References
http//www.fmrib.ox.ac.uk/fslcourse/lectures/fmri/
s_1278_data_expintro.htm
38
References
www.medrose.com.tw
www.magmedix.com
39
References
http//www.ausp.memphis.edu/ael/