NCSI 525 March 23, 2006

Data Analysis Lab J. Rosenberg / A. Wall / J. Maisog Room 150

In this laboratory exercise, you will be visiting three workstations, each with a different demonstration:

 Station 1: Motion Correction and Spatial Normalization  Station 2: Spatial and Temporal Filtering  Station 3: Statistical Analysis (t-test) and Overlay

One of us (Rosenberg, Wall, Maisog) will be at each workstation to guide you through the demonstration and field questions. You may not be visiting the stations in the above order. Whichever workstation you visit first, you should make sure that you understand the Section 1 “Orthogonal Views” below; you don’t need to repeat this section when you visit your 2nd or 3rd workstation.

Here is a nice review article on fMRI data analysis:

Smith SM, Overview of fMRI analysis, Br J Radiol. 2004;77 Spec No 2:S167-75.

- 1 - NCSI 525 March 23, 2006

Processing Pipeline for Single-Subject fMRI Data

Image Data or Transform Motion Correction “Raw” fMRI Scans (a.k.a. alignment or registration) A Processing Step

Temporal Filtering (usually high-pass)

Spatial Filtering (usually low-pass, smoothing)

Pre-processed fMRI Scans

Simple Averaging Statistical Analysis Simple Averaging and Operation (e.g., t-test) Differencing

Average Scan Mean Difference Maps Statistical “Map” (Mean) (“Contrast Maps”)

Template Spatial Normalization: Random Effects Group (e.g., “MNI 305 Average”) Compute Transform Statistics

Spatial Transform Spatial Normalization: Spatially Normalized (“Warp”) Apply Transform Statistical “Map”

Structural Scan Image Fusion in Template Space

Within-Subject Image Overlays

- 2 - NCSI 525 March 23, 2006

1. Sample fMRI Data for this session. Implicit reading task [1] as well as a simple finger-tapping paradigm. All fMRI data are transverse (axial), with the left of the subject on the right side of the image (radiologic convention). 2. Processing Pipeline for Single-Subject fMRI Data – On the next page is a block diagram illustrating the way single-subject fMRI data is processed at the Center for the Study of Learning. There is no single “correct” order in which to process data, but for our purposes we have found that this pipeline works well for us. This pipeline corresponds to the section entitled “First-level analysis” in the paper by Smith. Note that the two inputs are a set of “raw” functional MRI scans (usually on the order of 60 to 100 scans), and a “template” image defining a standard brain size and shape, and that the two outputs are a spatially normalized statistical map and a spatially normalized average scan. 3. Station 1: Motion Correction and Spatial Normalization – Rigid-body resampling of images to minimize within-subject artifacts due to head-motion; and non-rigid body resampling of images to “warp” them into a standard anatomic shape such as Talairach or Montreal Neurological Institute (MNI) space. a. The Problem of Head Motion – reasons for doing motion correction (research reasons for motion correction, insert them here). b. Head Motion before Motion Correction, Demonstration (AFNI) – demonstrate AFNI cine loop on Subject # 10043, Visit 1, IR1 before motion correction. c. Correcting Head Motion (AIR) – perform motion correction on sample data. d. (Apparent) Head Motion after Motion Correction, Demonstration – demonstrate AFNI cine loop on Subject # 10043, Visit 1, IR1 after motion correction. e. Stimulus-correlated motion – a special case of motion artifact is stimulus-correlated head motion [2-4]. What is the effect on activation maps? How might one address this problem? f. Spatial Normalization Demonstration – using an MPRAGE from Subject # 10043, Visit 1, IR1. Three methods: Nonlinear AIR “Warp” [5], SPM [6], piecewise linear (MEDx, [7]). Here, we will demonstrate using a nonlinear warp in AIR.

4. Station 2: Temporal and Spatial Filtering – Filtering in both time and 3D space. [8]. What does “filtering” mean? a. Low Frequency Artifacts – why do high-pass temporal filtering [9, 10]? b. High-Pass Temporal Filtering Demonstration #1 – Example on synthetic time- series data (from MEDx tutorial). c. High-Pass Temporal Filtering Demonstration #2 – Example on real fMRI data. d. Three reasons for doing low-pass spatial filtering (smoothing). e. Spatial Smoothing Demonstrations.

- 3 - NCSI 525 March 23, 2006

5. Station 3: Statistics and Overlays –. The data for this Station is from a simple 64-scan finger-tapping fMRI experiment. Most fMRI analyses use some method derived from the general linear model [11, 12]; in this demonstration, we’ll use simple t-tests (both single- as well as two-sample t-tests). Subjects alternated between a “rest” condition and a finger tapping condition in a classic block-design experiment.

a. Within-Subject Statistical Analysis: t-test – First, we demonstrate the analysis of fMRI data within-subject. This will generate, among other things, a statistical Z map for this subject, as well as a contrast map for this subject. The contrast maps from 46 subjects will be used for the random effects group analysis in (b) below. i. In MEDx, load the MEDx folder “/export/w/Methods_Core/Methods- Tutorials/NCSI-525/Station3- StatisticsAndOverlays/SingleSubjectStatistics.f”. This is data that has already been pre-processed using the techniques demonstrated in Stations 1 and 2. ii. This folder has two groups of scans, one for each condition. Select Page  Page Manager, and go to the Group named “Fix”. These are the 32 scans that were acquired during the fixation crosshair (“rest”) condition of the finger- tapping experiment. On the other hand, the 32 scans in the Group named “Tap” were acquired during the finger-tapping task.

- 4 - NCSI 525 March 23, 2006

iii. Select Toolbox  Functional  Group Statistics  Between Groups. Select the Tap group as the “Test Group,” and the Fix group as the “Control Group.” Set the other options to Parametric and Unpaired t-test (the defaults). Then click on the OK button. This will perform the two-sample t-test, contrasting the Tap scans against the Fix scans. iv. Inspect the image named “Tap vs Fix – Mean Diff”. This is the difference between the average of the Tap scans and the average of the Fix scans. This “contrast map” or “mean difference image” would be used as input in a random effects group analysis (analysis across subjects). v. Inspect the image named “Tap vs Fix – Unpaired T Test ZScore”. This is the statistical map showing areas of the brain associated by finger tapping. Note the strong area of activation on the left side of the brain, which is most probably motor cortex. There is also activation in the posterior occipital pole, due to the flashing circle that is present during the finger tapping blocks, but this is less strong than the activation in motor cortex. b. Group Statistical Analysis: Random Effects Analysis – Currently, the “gold standard” analysis for a group analysis (i.e., a statistical analysis across subjects) is a two-level random effects analysis [13]. In the first level, the fMRI scans are processed within-subject, generating a contrast map for each subject. For example, this would be the mean difference map we looked at in (5,a,iv) above. In the second level, the contrast maps are passed on to a statistical test such as a t-test or ANOVA. i. In MEDx, load the folder “/export/w/Methods_Core/Methods-Tutorials/NCSI- 525/Station3-StatisticsAndOverlays/GroupStatistics.f”. These are mean difference images contrasting the Tap condition against the Fix condition, as was demonstrated on one subject in (5,a) above. ii. Select Toolbox  Functional  Group Statistics. This will pop up a window named “Group Statistics.” Select the tab labeled “Within Group”. iii. For the Group, select the MEDx page named “Contrast Maps”. This is a group of mean difference images for the contrast Tap minus Fix, from 46 subjects. iv. For Operation, select “Single Group t-test”. Leave “Compare Mean To” set to the default of 0. Click on the OK button. This performs the single-group t- test, and generates a Results group. v. As with the within-subject t-test we performed in (a), this generates, among other things, a t-test map as well as a Z map. The t-test map is the first one in the Results group. Note that it indicates that the t-test has 45 degrees of freedom (“DOF=45”). vi. Inspect the Z map image; this is the image that has “ZScore” in its name. Note that it shows activation in left motor cortex as well as in the occipital lobe. A superior midline activation is likely SMA. There is a hint of right- sided cerebellar activation. vii. Save this Z map to disk as Analyze (AVW) format in the directory /export/w/Methods_Core/Methods-Tutorials/NCSI-525/Station3-

- 5 - NCSI 525 March 23, 2006

StatisticsAndOverlays, as a file named “Z.hdr”. This will create both Z.hdr and Z.img files. c. Overlays (Image Fusion) – We’ll show voxels that were greater than 3.821 (“activations”) in yellow and red, and voxels that were less than -3.821 (“deactivations”) in blue and green (the +/- 3.821 threshold was computed by selecting a False Discovery Rate [14] of 0.05). i. cd into the directory /export/w/Methods_Core/Methods-Tutorials/NCSI- 525/Station3-StatisticsAndOverlays. ii. Type the command

ThresholdOverlay.csh mask.img Z.img -3.821 3.821

then hit the Enter or Return key (carriage return). This will overlay the group t-test image that we created onto a stripped structural MRI scan that is in MNI space, generating an Analyze format file named Overlay.Z._- 3.821_3.821.hdr, along with an associated .img file. (The file “mask.img” is merely an image defining brain versus non-brain, pre-generated for this tutorial.) Now we’ll load this into VolView for a 3D volume rendering. iii. We will use the program VolView (Kitware, Albany, NY) for 3D rendering. This program is installed on the machine named app1. So, on the machine app1, type the command “volview”, then hit the Enter or Return key (carriage return). Load the .hdr file we generated in the previous step. iv. The data is displayed with a default color and opacity look-up table (LUT). For a somewhat nicer display, load the following alternative LUT, which was custom-built for our data: /export/w/apps/tcl/DeactAndActOverlays.vvt.

Use the left-mouse button to rotate the display, the middle-mouse button to pan, and the right-mouse button to zoom/dezoom. Note that there is a hint of deactivation in the right motor cortex, opposite the activation in the left motor cortex. Try some of the other options, especially cropping.

- 6 - NCSI 525 March 23, 2006

References.

1. Price CJ, Wise RJ, Frackowiak RS. Demonstrating the implicit processing of visually presented words and pseudowords. Cereb Cortex 1996; 6: 62-70.

2. Bullmore ET, Brammer MJ, Rabe-Hesketh S, Curtis VA, Morris RG, Williams SC, Sharma T, McGuire PK. Methods for diagnosis and treatment of stimulus-correlated motion in generic brain activation studies using fMRI. Hum Brain Mapp 1999; 7: 38-48.

3. Field AS, Yen YF, Burdette JH, Elster AD. False cerebral activation on BOLD functional MR images: study of low-amplitude motion weakly correlated to stimulus. AJNR Am J Neuroradiol 2000; 21: 1388-1396.

4. Gavrilescu M, Stuart GW, Waites A, Jackson G, Svalbe ID, Egan GF. Changes in effective connectivity models in the presence of task-correlated motion: an fMRI study. Hum Brain Mapp 2004; 21: 49-63.

5. Woods RP, Grafton ST, Watson JD, Sicotte NL, Mazziotta JC. Automated image registration: II. Intersubject validation of linear and nonlinear models. J Comput Assist Tomogr 1998; 22: 153-165.

6. Ashburner J, Friston KJ. Nonlinear spatial normalization using basis functions. Hum Brain Mapp 1999; 7: 254-266.

7. Talairach J, Tournoux P. Co-planar stereotaxic atlas of the human brain : an approach to medical cerebral imaging. Stuttgart and New York: Thieme Medical Publishers; 1988.

8. Hansen CL. Digital image processing for clinicians, part II: filtering. J Nucl Cardiol 2002; 9: 429-437.

9. Woolrich MW, Ripley BD, Brady M, Smith SM. Temporal autocorrelation in univariate linear modeling of FMRI data. Neuroimage 2001; 14: 1370-1386.

10. Tanabe J, Miller D, Tregellas J, Freedman R, Meyer FG. Comparison of detrending methods for optimal fMRI preprocessing. Neuroimage 2002; 15: 902-907.

11. Friston K, Holmes A, Worsley K, Poline J-P, Frith C, Frackowiak R. Statistical parametric maps in functional imaging: A general linear approach. Human Brain Mapping 1995; 2: 189-210.

12. Bandettini PA, Jesmanowicz A, Wong EC, Hyde JS. Processing strategies for time- course data sets in functional MRI of the human brain. Magn Reson Med 1993; 30: 161- 173.

13. Friston KJ, Holmes AP, Price CJ, Buchel C, Worsley KJ. Multisubject fMRI studies and conjunction analyses. Neuroimage 1999; 10: 385-396.

14. Genovese CR, Lazar NA, Nichols T. Thresholding of statistical maps in functional neuroimaging using the false discovery rate. Neuroimage 2002; 15: 870-878.

- 7 -