AFNI Soup to Nuts: How to Analyze Data with AFNI from Start to Finish

At the moment, this information is input via a command-line interface, or with an optional question/answer session (afni_proc.py -ask_me). Eventually, a GUI will become available (but not yet).

Slide 3

Illustration of Stimulus Conditions:

Analysis Steps:

Back to afni_proc.py :

Line-by-Line Explanation of s1.afni_proc.block command:

Line-by-Line Explanation of s1.afni_proc.block command (cont…)

PART I  Process Data for each Individual Subject:

Pre-processing is done by the proc.sb23.blk script within the directory, AFNI_data4/sb23.blk.results/.

STEP 0 (tcat): Apply 3dTcat to copy datasets into the results directory, while removing the first 3 TRs from each run.

STEP 1 (tshift): Check for possible “outliers” in each of the 9 time series datasets using 3dToutcount . Then perform temporal alignment using 3dTshift.

Subject sb23’s outlier files:

in afni, view run 01, time points 38, 39 and 40

Next, perform temporal alignment using 3dTshift.

Subject sb23’s newly created time shifted datasets:

STEP 2: Register the volumes in each 3D+time dataset using AFNI program 3dvolreg.

Subject sb23’s 9 newly created volume registered datasets:

view the registration parameters in the text file, dfile.rall.1D

STEP 3: Apply a Gaussian filter to spatially blur the volumes using program 3dmerge.

results from 3dmerge:

STEP 3.5 : creating a union mask

next, take the union of the run masks

so the result is dataset, full_mask.sb23.blk+orig

STEP 4: Scaling the Data - as percent of the mean

Another example:

Scale the Data:

Compare EPI graphs from before and after scaling

compare EPI images from before and after scaling

STEP 5: Perform a regression analysis on Subject sb23’s data with 3dDeconvolve.

3dDeconvolve command - Part 1

3dDeconvolve command - Part 2

3dDeconvolve command - Part 3

3dDeconvolve command - Part 4

There are a many reasons why 3dDeconvolve might fail:

After running 3dDeconvolve, an 'all_runs' dataset is created by concatenating the 9 scaled EPI time series datasets, using program 3dTcat.

Create ideal response files, in case the user wishes to plot them in the graph window along with the time-series data.

Compute estimates for the amount of blur in the data, as requested via afni_proc.py option, -regress_est_blur_errts.

Generate a script that the user can run to review the unprocessed EPI data.

Let’s view some data:

Now plot the all_runs dataset along with the fitts dataset

For each subject, we need only the 9 beta weights of our stimulus conditions to perform the group analysis. For our class example, the beta-weights are located in the following sub-bricks:

Result: One dataset for each of the 16 subjects, containing only the 9 sub-bricks of regressors of interest. These regressors will be used for the ANOVA.

STEP 7: Warp the beta datasets for each subject to Talairach space, by applying the transformation in the anatomical datasets with adwarp.

PART 2  Run Group Analysis (ANOVA3):

3dANOVA3 Command - Part 1

3dANOVA3 Command - Part 2

3dANOVA3 Command - Part 3

-adiff: Performs contrasts between levels of factor ‘a’ (or -bdiff for factor ‘b’, -cdiff for factor ‘c’, etc), with no collapsing across levels of factor ‘a’.

-aBcontr: 2nd order contrast. Performs comparison between 2 levels of factor ‘a’ at a Fixed level of factor ‘B’

In class -- Let’s run the ANOVA together:

-fa: Produces a main effect for factor ‘a’

Brain areas corresponding to “Telephone” (reds) vs. “Face-to-Face” (blues)

Brain areas corresponding to “Positive Telephone” (reds) vs. “Positive Email” (blues)

Many thanks to NIMH LBC for donating the data used in this lecture