SPM5 Tutorial Part 2

SPM5 Tutorial Part 2

Tiffany Elliott

May 10, 2007

•To begin setting up your model click on the Specify 1st-level button

Double click fMRI model specification to expand the node.

Click Directory and Specify Files…

Select your stats folder.

Double click the Timing parameters node to expand it.

Enter Timing Parameters

• Units for design scans (the units we will use to report our onsets)

• Interscan interval 2 (this is the TR)

• Microtime resolution 16, default

• Microtime onset 1

Data & Design

Make 3 New “Subject/Session” nodes for each of our 3 VOD runs.

Double click on the first Subject/Session node to expand it and click on the Scans node below it. Now click specify files to select your data files.

Select your data for the first session

Click on the Multiple conditions node.

We will now load previously defined task condition mat files.

These will be for our 4 task conditions: frequents, targets, alcoholic beverage novels and non-alcoholic beverage novels.

Each of the previously defined mat files contains information for the conditions for that session.

There are three defined variables contained within…

names – a cell array of the names of the conditions

Onsets – a cell array of the event onsets (in scan units)

Click High-pass filter and then Specify Text. Enter 128 for your high pass filter cutoff.

A high pass filter removes low frequency confounds from your data. I use this trick to remember the definition… “higher frequencies pass the test of a high pass filter”

We will set ours to a wavelength of 128 seconds.

Repeat scan, condition and high pass filter selection for the other two sessions…

Factorial Design

Do nothing for this node.

Basis Functions

Choose Canonical HRF and Model Time Derivatives

•Basis functions are functions that you want to base your model on

•The model will be a linear combination of your basis functions

•Modeling the time derivatives allows for variations in subject-to-subject and voxel-to-voxel responses

Hemodynamic Response Function

Stimulus Onset Convolved with HRF

Model Interactions (Volterra)

Do not model interactions

Global Normalisation

Always select “None” for this

Here’s why…

•It removes signal that covaries with the global mean•In fMRI large activations bias the global mean calculation• Thus removing the global mean will also remove those activations

Note: this isn’t a problem in PET scanning

Explicit mask

Leave this as <None>

This would be useful if you want to apply a mask to your results. For example one might pick the standard brain mask to mask out things outside of the brain.

Serial Correlations

Select AR(1)

•fMRI produces a lot of low-frequency noise•This noise is highly correlated from one timepoint to the next•This node will perform corrections to account for this

When you have your model set

• Save in the parameters like you did during the preprocessing tutorial by clicking on the save button and saving in a file type of your choice

• Run the analysis by clicking Run

To Review your design matrix click design, select your SPM.mat and select Design Matrix from the dropdown menu.

Your design will be displayed in the graphics window.

Each of the 3 blocks to the right represent the model for that session

Now select Design orthogonality.

•Your design orthogonality is plotted in the table below your design matrix.

•Each little shaded square is representative of the collinearity between two columns of your design matrix.

•The darker the square the more the two conditions correlate with each other and the less orthogonal they are.

•When modeling fMRI you don’t want your separate conditions to be collinear. You want them to be more orthogonal (i.e. linearly independent).

Now we will estimate our model

Click Estimate and select your SPM.mat

Double click the Model estimation node to expand it.

Select the SPM.mat file you created from the previous step. It will be in the folder you chose to run your stats in.

We will leave the Method setting at “Classical”

Then click Run.

Now you will have images in your stats folder associated with your model.

Images you will see in your stats folder after estimation…

• beta_000k.img – estimated regression coefficient where k indexes the kth regression coefficient

• mask.img – binary images that indicates which voxels were used in the analysis

• ResMS.img – variance of the error

• RPV.img – the estimated resels per voxel (a resel is a resolution element)

Now click the Results button to view your results and select your SPM.mat file in your stats folder.

You will see a window pop up that will allow us to specify contrasts to look at.

Contrasts are simply comparisons of your regression coefficients (beta images).

We will specify a new contrast to look at the difference between the alcoholic and non-alcoholic novel conditions.

Enter this in the contrast vector box. And hit OK when you are done.

0 0 0 0 1 0 -1 0 0 0 0 0 1 0 -1 0 0 0 0 0 1 0 -1 0

Each 1 corresponds with the novac condition from each session and each –1 corresponds with the novnc condition from each session.

Here we will define the novac-novnc contrast.

Now we are ready to look at the contrast.

Highlight it and click Done.

Mask with other contrasts? NoTitle for comparison: novac-novnc P Value adjustment to control: NoneThreshold {T or p value}: 0.05

Enter these values for the prompts that follow…

• Mask with other contrasts? No

• ROI Analysis? No

• Title for comparison: novac-novnc

• P value adjustment to control: none

• Threshold {T or p value}: 0.05

• Extent threshold {voxels}: 0

Say hello to your SPM results.

You can move the little red cursor around by drag and drop.

If you right click on the brain you can select from some shortcut options that include jumping to the global maxima.

The voxel you are on is displayed in coordinates to the left (see the red circle in the picture).

Can you find the voxel coordinates for the global maxima in the results?