Resting-state functional connectivity analysis (using Conn)

Let's talk about resting state functional connectivity (rsFC) analysis. FC analysis gives us a better understanding of the widespread activation patterns of the brain than a purely modular view. Using FC anlaysis, researchers have been able to identify a variety of various networks that appear to be fairly robust in their activation in response to various environmental conditions. rsFC has revealed certain networks which seem most active in a "task-negative" state. The analysis steps described below are all conducted through the functional connectivity toolbox (CONN; https://www.nitrc.org/projects/conn). As I learned the hard way while working through various issues in my most recent analysis, I would recommend making sure you have the latest version of SPM12 and Conn in your path before you get started.

To get started, open Conn in the folder containing your data. If you preprocessed your data using FSL, be sure to use the gunzip function in MATLAB or bash to first decompress the images before using them in Conn (Conn seems to be able to decompress them itself, but my preference is to unzip them separately).

1. SETUP TAB

Basic: Input number of sessions, repetition time and acquisition type. For number of sessions and repetition time, if the same for each subject, you can enter one number in each box—Conn will expand this into that value for each subject.

If your dataset contains subjects with multiple sessions, you can enter that here. We generally prefer to include each session as a separate subject, and then apply group differences as a covariate.

Structural: Input a normalized high resolution structural image (MPRAGE) for each subject. Individual preprocessing steps can be applied to the structural image at this time, but preprocessing should be done in advance.

Functional: Select all preprocessed functional images for each subject. Again, preprocessing steps can be applied here but should be done in advance.

ROIs: Your ROIs should include one for Grey Matter, White Matter and CSF as well as your specific regions of interest.

Our lab tends to use ROI images we extracted from a different atlas, but you can use the atlas in Conn to pick your ROIs.

For Grey Matter, White Matter and CSF, if you do not select files, Conn will segment the uploaded MPRAGE files for you and load in those files. Otherwise, if you have previously segmented the MPRAGE, you can load in those files for each subject.

We use the default number of dimensions.

For Grey Matter, “Subject-specific ROI” should be checked.

For White Matter and CSF, “Subject-specific ROI,” and “Regress out covariates” should be checked.

For your specific ROIs, check whichever boxes are appropriate. We use “Mask with Grey Matter” and “Use ROI source data.” Our ROIs are neither subject nor session specific, and we do not put multiple ROIs in one file. When the ROI is not subject specific, Conn will easily allow you to select one file for all subjects.

Conditions: For rsFC analysis, we only need one condition, called rest, with onset 0 and duration inf. It should include all subjects, with one session. Leave the optional fields set to the defaults.

Covariates, 1st level: Create a covariate called realignment and select your file created during realignment (SPM uses the prefix “rp” for those files). If more advanced motion correction is used, use those files instead.

Preprocessing: This is optional and depends on your dataset but we tend to use ART-based outlier detection.

Covariates, 2nd level: The second level should include your group identifications and any other covariates you are interested in controlling for (age, gender etc). For example if you have 3 groups with 10 subjects each, you would create a covariate for group 1, and add a 1 for each subject in the group and a 0 for the other subjects. Then make a separate covariate for each group and do the same). If you have subjects who overlap in certain groups, you can add 1s wherever appropriate.

If entering a continuous variable as a covariate, it is recommended to enter centered values.

Options: Select all analysis types you are interested in (keeping in mind the amount of storage each takes up). We tend to use seed-to-voxel only.

Click DONE to run the setup.

2. DENOISING TAB

Select the effects you want to remove as confounds— these include white matter, CSF, motion and “effect of rest” on top of any other covariate of no interest. Defaults for other fields should be sufficient, unless you have a specific reason to change them.

Click DONE to run the denoising.

3. FIRST-LEVEL ANALYSES

Create a new analysis and name it something useful. If using bivariate correlation, you can include all the seeds you wish to examine in one analysis. If using semipartial, only include the seeds you wish to control for (so generally, seeds for one network at a time). Select which weighting you wish to use—it is likely best to begin with no weighting, unless you have specific reason to use one. We do only seed-to-voxel analysis, but if you wish to include ROI-to-ROI as well and have the space, feel free.

Click DONE to run the first-level analyses.

4. SECOND-LEVEL ANALYSES

Make sure to select the correct first-level analysis. Under subject effects, you will see the subject groups you created—highlight those you wish to compare or examine, then you can either hand-input a contrast or select from one of Conn’s options - we tend to use any difference among the selected groups. We only have one condition, rest, which should be selected. Pick your ROIs of interest - is using multiple, you can select the weighting of ROIs in the options. We tend to use Average for the main effects of the seeds.

Click results explorer (in the lower left). This will create SPM.mat files for the contrasts you have selected, as well as open them in Conn’s display. You can either use the Conn results explorer, or open these SPM.mat files in SPM’s results viewer to perform any post-hoc tests and use SPMs display. Conn also does have a button in the results viewer, “SPM display,” which will open SPM’s contrast manager directly for you.

In the results explorer, you can view parametric or non-parametric results across various thresholds of interest. If you are interested in a different network, simply close the results explorer, select the new seeds and run a new analysis.

**Note: As was likely obvious throughout this post, there are a LOT of options that can be selected for your specific analysis, and I just reported what our lab uses. The Conn documentation should be able to help you decide if any of the other options are appropriate for your analysis.**