PLS User's Guide did not say that your data need to be divided into blocks of equal volume number. In order to run Seed PLS, extraction of the seed data need to be done with the "Multiple Voxel Extraction" option on the PLS results. Then, using the extracted seed data in Regular Behavior PLS.

So the step is:

1. Running Mean-Centering PLS, and find seed locations from activation map (Brain LV).
2. Using above seed locations to extract voxel values with the "Multiple Voxel Extraction" option on the PLS results.
3. Running Regular Behavior PLS. The behavior data will be the voxel values that you extracted above.

I don't know how to assess the temporal aspect of the FC.

Hi Randy,

Many thanks for your kind answer.
Here is a small description of our data and what we want to do. So, we have two sets of data including two sessions (day1 and day2). Each session includes two functional runs (rest1 & rest2) separated by a memory task. We would like to explore the effect of the memory task on the functional connectivity (FC) of the default network (DN). An example of contrast would be "day1_rest2 - day_rest1". So, we oriented our choice to the seed-voxel analysis based on its possibility to select only one network by using a seed belonging to the desired network (eg. PCC for the DN). As I understood, Seed-PLS could provide a statistical map depicting the difference between correlation maps (correlation maps: seed with rest1 & seed with rest2).
The idea of dividing the time series into blocks mentioned in my previous message is borrowed from the paper of  "Grigg O. and Grady Cheryl L., PloS one, Vol 5, issue 10, 2010". The time series resampling was done in order to asses the temporal aspect of FC of DN as each block represents a time point on the time series data. The used method was the Seed-PLS but I am not sure that they process all blocks in one step.

Hope that this could help us to select the suitable method.

Regarding the two options (methods) that you proposed in your answer, I am not sure to be able to say which one I should use as I am beginner in this field.

Many Thanks,

Mohamed

Hi Mohamed..

in the Grigg paper, the full resting state run was divided into 10-sec blocks (5TRs).. All blocks were entered into the analysis.. Look more carefully at Fig 1. The brain pattern identifies the regions that show mostly positive correlations with the seed across those 10-s blocks for both RS runs. Fig2 shows the second pattern, where the correlations with the seed differed between RS1 and RS2.

So, yes, you can do this kind of analysis using PLS. You would run a Block fMRI Seed PLS, specifying block-lengths smaller than the length of the RS run. If you only want to look at run-to-run differences, you would could derive a metric outside of PLS, as Randy mentioned, or use "properly" preprocessed data and enter those into PLS, using a block length equal to the number of TRs in your study - it's hard to be more helpful than this without really knowing what the question is (Run1-Run2 is not specific enough).

Note: PLS, depending on the structure of the dataset entered, is unlikely to give you "one network" - a region can and most likely will, correlate with activity in multiple sets of regions. - see Fig4 in Grigg- only a subset of regions in LV1 and LV2 showed overlap - i.e., more than one functional network was found. cheers, nancy

Hi Mohamed

Those are part of the results.. the "singular image" is derived from the "brainlv" (u) and the correlations for each block are derived from the "designlv" (v).

[u s v]=svd(data');

This is explained more fully in the 2004 review paper - you may want to re-read that.
n

This is definitely something can be done with PLS.  Cheryl Grady has answered similar questions to the ones you have:

 

http://www.ncbi.nlm.nih.gov/pubmed/19789183

http://www.ncbi.nlm.nih.gov/pubmed/15989937

http://www.ncbi.nlm.nih.gov/pubmed/12921348

Dear PLS experts,

When using univariate method for connectivity analysis (SPM), white matter and CSF signals are usually regressed out from fMRI data prior analysis. fMRI data also are bandpass filtered to keep only the neuronal component. I am wondering if these preprocessing steps also are needed when using PLS?

Any suggestion/explanation will be very welcome,

Many thanks

Dear PLS experts,

I am giving a talk in my lab in two days about the use of PLS as a tool for neuroimaging data analysis and I am wondering if anyone have a simple way (schems, example, or ...) to explain the SVD and the resulted U and V matrix and their relation to the data, to a non-specialist audiance?

Many thanks in advance,

mohamed

Dear PLS experts,

Are there any documentation describing the preprocessing steps (and the methodological reasons) done on the Y (behavioral/design) and X (brain activity) matrices for each PLS variations (Task PLS, behavioral PLS, Mean-Centering PLS, Multi-bloc PLS)?

Many thanks,

Mohamed

I want to speak about the PLS transformations (mean centering, ....) done on the data matrices (Y and X) beforme combining them on a one "correlation" matrix and not the spatial preprocessing of the data.