1. MNTP Summer Workshop fMRI BOLD Response to Median Nerve Stimulation: A Comparison of Block and Event-Related Design
Mark Wheeler
Destiny Miller
Carly Demopoulos
Kyle Dunovan
Martin Krönke
Todd Monroe
Dil Singhabahu
Elisa Torres
Christopher Walker
Funded by:
NIH R90DA02342

2. MNTP Workshop: Learning Objectives
In-depth understanding of preprocessing of fMRI data
Filtering
Motion correction
Slice Time Correction
Smoothing
Registration
Conduct first-level analyses
Conduct group-level analyses
Investigate two experimental designs

3. The Task: Median Nerve Stimulation
Electrical stimulation of the median nerve by applying pulses to the wrist of the non-dominant hand
Voltage: motor threshold

4. Blocked Design 10 repetitions Pros. Cons.
Stim ON 10s
Stim OFF 16s
Stim ON 10s
Stim ON 10s
15Hz
15Hz
15Hz
Stim OFF 16s
Stim OFF 16s
10 repetitions
Pros.
Cons.
Excellent detection power (knowing which voxels are active)
Useful for examining state changes
Poor estimation power (knowing the time course of an active voxel)
Relatively insensitive to the shape of the hemodynamic response.

5. Event-Related Design Pros. Cons.
Good at estimating shape of hemodynamic response
Provides good estimation power (knowing the time course of an active voxel)
Can have reduced detection power (knowing which voxels are active)
Sensitive to errors in predicted hemodynamic response

6. Event Related Task Design
Three different frequencies: 15Hz, 40Hz, 80Hz (Kampe, Jones & Auer, 2000)
Event length: 4s
Inter-stimulus jitter – 2, 4, 6 seconds
Exponential distribution (Dale, 1999)
15Hz +
40Hz
80Hz
4s (2TR) 4s
Time
2s Jitter
6s Jitter
4s Jitter

7. Data Acquisition Scanner: Allegra 3T N=5 Structural Scan
T1 weighted MPRAGE
176 slices
Voxel Size 1mm
Functional Scans: Median Nerve Stimulation
Volumes
140 for block
233 for event-related
Voxel Size 3.5mm
Slices 34
Interleaved
TR 2s
T2* contrast

8. Single Subject Demonstration
Processing stream
Data-conversion
Dicom2Nifti
Statistical analysis
GLM
Statistical Parametric Mapping
Preprocessing
Block Design
Single Subject Demonstration
Slice-timing
Motion correction
Temporal Filtering
Smoothing
Registration / Normalization

9. Preprocessing: Slice Time Correction (STC)
Huettel, Song, McCarthy 2009
Stronger influence of STC for event-related vs. block-designs
sensitivity to timing / shape of HRF
Slice acquisition order
interleaved slice acquisition (34 slices in 2s)
avoids cross-slice excitation
Debate on STC before / after motion correction?
before head motion (interleaved)
Temporal non-linear sinc interpolation

10. Motion correction
Due to subject movements inside the scanner, a voxel might represent different parts of the brain across time points, introducing artifacts
Huettel, Song, McCarthy, 2004

11. Motion correction Estimation Rigid-body transformation 6 DOF 0.2 mm
-0.1
time (TRs)
radians
0.003
-0.004
time (TRs)
Interpolation
trilinear
Nearest neighbour
(tri-)linear
Non-linear (sinc, B-spline)

12. No Motion correction Motion corrected Z-Value: 3.9
% signal change
Z-Value: 3.9
Crosshair location: Postcentral gyrus
Time (TRs)
Motion corrected
% signal change
Time (TRs)
Z-Value: 3.8

13. Temporal Filtering A highpass filter can remove these unwanted effects
Artifacts like “slow scanner drift” and changes in basal metabolism can reduce SNR
A highpass filter can remove these unwanted effects
Do not want to remove task-related signal
Block Design Task: 10s on, 16s rest
Woolrich et al. (2001) recommends filter of at least 2 epochs duration
52s temporal filter  .019 Hz
Also compared effects of 0 Hz, .038 Hz, .01 Hz
Little difference between
.019 Hz
.038 Hz
.01 Hz
Discrete cosine transform
Low frequencies
0 – 0.1 Hz
0.1 – 0.5 Hz respiratory
Hz cardiac

14. 0Hz / No Temporal Filtering
% Signal Change
Time (TRs)
52s / .019Hz Temporal Filter
% Signal Change
Time (TRs)

15. Smoothing Spatially filters data using Gaussian Kernel to remove noise
Reduces spatial resolution
Improves signal to noise ratio
Consider ROI and voxel size in determining the size of the kernel
Gaussian Weight

16. 0mm smooth
4mm smooth
8mm smooth
20mm smooth

17. Registration / Normalization
Why?
Group analysis
Compare results in common coordinate system (MNI)
Karsten Müller
How?
Estimate transformation
Combining affine-linear (12 DOF) subject  standard space (FSL FLIRT)
nonlinear methods (> 12 DOF) subject  subject (FSL FNIRT)
least squares cost function
2. Resample / Transform / Interpolate
Nearest neighbour
Linear interpolations
Bi-, trilinear
Non-linear interpolations
B-Spline, sinc (Hanning)

18. Preprocessing Summary
Data-conversion
Dicom2Nifti
Block Design
Filtering
Highpass (52s / .019Hz)
Discrete cosine transform
Motion correction
Rigid-body, 6DOF
Trilinear interpolation
Statistical analysis
GLM
1st-level
Group-analyses
Slice-timing
Interleaved
Sinc interpolation
Smoothing
FWHM, 8mm
Registration / Normalization
Affine-linear + Non-linear
Statistical Parametric Mapping

19. Design matrix comparison: Block vs. Event-related
Block design
15Hz
Time
Event-related
40Hz
80Hz
15Hz

20. Block vs. Event-Related Design
Block Design
15Hz activation map
Modeled with gamma function
Event-Related Design
15Hz activation map
Modeled with double-gamma function

21. Functionally vs.structurally defined ROIs
ROI (structure)
ROI (functional 9 mm)
ROI (functional 6 mm)
ROI (functional 3 mm)

22. Effect of Region of Interest on Task Related Median Percent Signal Change
-0.10
0.00
0.10
0.20
0.30
0.40
0.50
15Hz
40Hz
80Hz
80Hz > All*
Functionally Defined
Structurally Defined
Median Percent Signal Change
Median percent signal change of the BOLD response to 15, 40, and 80 Hz median nerve stimulation from baseline. The functionally defined ROI was defined as above; the structurally determined ROI was created using the Harvard-Oxford Structural Atlas defined boundaries for the postcentral gyrus, and masked to include only right hemisphere activity.
*Note: “80 Hz > All” denotes the contrast of the response to 80 Hz stimulation against both the 15 and 40 Hz responses combined. It has been included to demonstrate the potential to model differences between conditions with the GLM approach.
ROI – F (1, 4) = 6.431, p = .064
Frequency – F (2, 4) = , p = .007
Frequency * ROI – F (2, 8) = 5.101, p = .037

23. Future Directions: Condition and Subject Timeseries
Modeled 15 Hz response for 1 subject
Arbitrary Units

24. Event-Related Activation Comparison
15 Hz above baseline
40 Hz above baseline
80 Hz above baseline

25. Future Directions: Overlapping Activation
Investigate condition specific differences in activation patterns

26. References
Dale, A. M. (1999). Optimal experimental design for event-related fMRI. Human Brain Mapping, 8: 109–114.doi:  /(SICI) (1999)8:2/3<109::AID- HBM7>3.0.CO;2-W
Huettel, S. A., Song, A. W. and McCarthy, G. (2004). Functional magnetic resonance imaging. Sunderland, MA: Sinauer Associates
Kampe, K. K., Jones, R. A. and Auer, D. P. (2000). Frequency dependence of the functional MRI response after electrical median nerve stimulation. Human Brain Mapping, 9: 106–114. doi:  /(SICI) (200002)9:2<106::AID- HBM5>3.0.CO;2-Y
Woolrich, M. W., Ripley, B. D., Brady, M., Smith, S. M. (2001). Temporal autocorrelation in univariate linear modeling of FMRI data. NeuroImage, 14,

27. Thank you Mark Wheeler Destiny Miller Seong-Gi Kim Bill Eddy
Tomika Cohen
Rebecca Clark
Fellow MNTPers!
Funded by: NIH R90DA02342 27