Goals

  • To understand important factors in optimizing block designs including
  • Saturation
  • Confounding Paradigm Frequency
  • To explore the use of Event Related Averages for fMRI data analysis

Accompanying Data

Tutorial 5 Data (Download the data from this link)

Background

In this tutorial we will explore optimization of block designs from multiple perspectives. Block duration, ordering, and the effects of using baselines are all considered. Examples use Variant 1 – Orders 1 to 7, and Variant 2 – Orders 1 to 4 from the Course Data Description.

For part of the tutorial, we will use a new kind of analysis: Event Related Averaging . Recall that Event Related Averaging can be used for both block and event related designs, and that it shows the average observed response corresponding to the onset of a condition. The following figure shows the average BOLD response with standard error bars corresponding to the Face (pink) and Hand (purple) conditions in a block designed experiment (16 seconds, no baselines). The average response is shown for 2 seconds preceding each block onset (pre) and for 20 seconds post-onset (post). Using 20 seconds, rather than 16 corresponding to the block duration, we can visualise average BOLD response for the 4 seconds following the condition.

This event related average plot could be used as evidence to support the hypothesis that the selected voxel or ROI shows significant activation for faces, and significant deactivation for hands.

General Instructions

Six of the orders (Variant 1 – Orders 1 and 2, Variant 2 – Orders 1 to 4) use 16 second blocks . Using strategies from the previous tutorial (voxelwise GLM analysis, ROI GLM) as well as Event Related Averaging, analyse data recorded using each of these variants. Compare the quality of data recorded using each order, reflecting on the pros and cons of each design.

Four of the orders (Variant 1 – Orders 3 to 6) use variable duration blocks between 4 and 16 seconds , and one order (Variant 1 – Order 7) uses 64 second blocks. Using voxelwise GLM analysis, explore the data and model statistics using voxelwise GLM analysis. Discuss the ability to detect significant activation at reasonable statistical thresholds.

Files

./Variant1/
1 VMR file – Anatomical scan
7 VTC files – Volume time courses, one per order
7 PRT files – Protocol files, one per order

./Variant2/ 1 VMR file – Anatomical scan
4 VTC files – Volume time courses, one per order
4 PRT files – Protocol files, one per order

Specific Instructions

Use the following instructions for simultaneous Event-Related Averaging and ROI GLM analysis. For more details on setting up a GLM, manipulating thresholds, and ROI GLM analysis, refer to the GLM or Spatiotemporal Smoothing tutorials. Use Event-Related Averaging only where needed (i.e. for Question 1 only). Try the specific instruction on Order1 of Variant1 and then use what you learned to answer the Tutorial questions and repeat for the different orders and variants.

1.Open the VMR file in the corresponding Variant folder.

2.Attach a VTC file.

3.From the Analysis menu, select Event Related Averaging...

3.1. Ensure that the VTC file is selected.
3.2. Select both conditions.
3.3. For scans with 16 second blocks, specify a Post event onset duration of 20 volumes.
3.4. Click Create AVG and save the file using a name such as ‘Order1ERA’.

 

4.From the Analysis menu, select General Linear Model: Single Study.

5.Click Define Preds, then click GO.

6.From the Analysis menu, then Overlay General Linear Model.
6.1. Toggle the Predictor Nr. boxes to define a contrast – use Face > Hand for this tutorial.
6.2. Click Balance +/-.
6.3. Click OK.

7.Adjust the statistical threshold using the Increase/Decrease Threshold buttons until you are satisfied with the resulting map. Alternatively, click the Analysis menu and select Overlay Volume Maps for more thresholding options.

8.Locate a region you might identify as the FFA.

  1. Right-click on this region and select Show ROI Time Course.

10.Continue your analysis from the ROI Signal Time Course window.

10.1. Click the little green box to reveal the full window.
10.2. Under Event-related averaging, click Browse, then select and open the AVG file you created earlier. Make sure you select the AVG file for the correct order.
10.3. If it is not checked, click the Enable checkbox.

 

After the Event-Related Averaging Plot appears you can continue your analysis without closing it.

11.Click ROI GLM... to analyse the GLM statistics within the selected ROI. Refer to the Spatiotemporal Smoothing tutorial for more details.

11.1. Click Browse, then select the design matrix file (SDM) corresponding to the correct order.
11.2. Navigate to the Contrasts tab to specify a contrast. For this tutorial, use Face > Hand.
11.3. Select your desired Correction / Output Options.
11.4. Click Fit GLM.

 

Tutorial Questions

Where applicable, use a combination of analysis techniques seen in this and prior tutorials to justify your answers. For example, you could set a contrast for Face > Hand, choose a suitable statistical threshold, then perform an ROI GLM analysis on a region that appears to be highly selective for faces. Statistics including the beta weights and R scores can be used to quantify the quality of the data and corresponding block design.

16 Second Blocks

Question 1:

a) Using GLM statistics and Event Related Averaging, compare model statistics for data recorded using 16 second blocks without baselines (Variant 1 – Orders 1 and 2) to those with baselines (Variant 2 – Orders 1 and 2). Is this a fair comparison?

b) Using Event Related Averaging plots as evidence, discuss the pros and cons of the block designs with baselines (Variant 2 – Orders 1 to 4). Which, if any, of these 4 orders do you think is a superior design?

c) Focusing on Variant 2 – Orders 3 and 4, use Event Related Averaging plots as evidence either in support of or against this block design. Can you think of an alternative design that might be better?

Variable Duration Blocks

Question 2:

a) Using GLM statistics, compare model statistics for data recorded using Variant 1 – Orders 3 to 6. Do any of these block designs seem to be superior? Why?

b) Perform a similar analysis for Variant 1 – Order 7. Are you able to find evidence of Face selective activation at a reasonable statistical threshold? What do you think the problem with this block design is?

Relationship Between Filtering and Block Designs

Belwo find the interactive figure for temporal smoothing from the previous tutorial. Using Voxel 1, adjust the Minimum Freq. slider incrementally while clicking Optimize GLM to see the new beta weights and residuals.

Question 3:

a) At what frequency does high-pass filtering (cutting off frequencies below the minimum) seem to become detrimental? In other words, at what point do you suspect it is filtering out signal instead of strictly removing noise?

b) Knowing that the experimental design for this voxel uses 16 second blocks with 16 second baselines, could this minimum frequency be predicted? How?