Questions about processing pipeline using SPM and the toolbox

[Frances Jin]

Dear CosMoMVPA experts,

I have been using SPM and my own matlab-based MVPA scripts so far, but I would like to start using this toolbox. If I can take a shortcut by asking you, is the following pipeline sounds right in terms of combining SPM, FreeSurfer, and CosMoMVPA? I also have specific questions in post-searchlight processing, please see below:

Start with raw data: T1 + functional epi

Step 1, preprocessing in SPM: 1) realign epi, 2) coregister epi to T1, 3) run GLM on realigned epi to get beta maps, 

Step 2, get surface from FreeSurfer: 1) recon-all T1 --> .pial and .white.

Step 3, resample surfaces to standard topology using  pymvpa2-prep-afni-surf. I guess I have to either use AFNI or PyMVPA just for this, right? Is there a way possible to avoid using python?

Step 4, conducting subject-level MVPA using CoSMoMVPA: 1) define nodes using resampled surfaces, 2) map nodes to voxels which is from the beta maps obtained from Step 1, 3) conducting the actual analyses (e.g., RSA and classification), 4) store result from voxel to the nodes.

Step 4, Group-level analyses, this is where I am most confused. When I conducted conventional volume-based searchlight, I did the analysis at subject's native space, and normalize searchlight result to, say MNI space. Do I need to do any transformation here at all? Is this standard surface thing already taken care of from AFNI's MapIcosahedron or by pymvpa2-prep-afni-surf?

Step 5,  shall I smooth the node-based searchlight results?

Thanks in advance for the detailed answers!

Best,

Frances

[Nick Oosterhof]

Greetings,

> On 28 Jun 2016, at 15:50, Frances Jin <jinjin...@gmail.com> wrote:
>
> I have been using SPM and my own matlab-based MVPA scripts so far, but I would like to start using this toolbox. If I can take a shortcut by asking you, is the following pipeline sounds right in terms of combining SPM, FreeSurfer, and CosMoMVPA? I also have specific questions in post-searchlight processing, please see below:
>
> Start with raw data: T1 + functional epi
> Step 1, preprocessing in SPM: 1) realign epi, 2) coregister epi to T1, 3) run GLM on realigned epi to get beta maps,
> Step 2, get surface from FreeSurfer: 1) recon-all T1 --> .pial and .white.
> Step 3, resample surfaces to standard topology using pymvpa2-prep-afni-surf. I guess I have to either use AFNI or PyMVPA just for this, right?

pymvpa2-prep-afni-surf is a python script that requires AFNI and FreeSurfer. Note that it also takes care of coregistration between volume and surface, and merges left and right hemishere surfaces into a combined bilateral surface.

> Is there a way possible to avoid using python?

I don't see how you could easily avoid using python.

> Step 4, conducting subject-level MVPA using CoSMoMVPA: 1) define nodes using resampled surfaces, 2) map nodes to voxels which is from the beta maps obtained from Step 1, 3) conducting the actual analyses (e.g., RSA and classification), 4) store result from voxel to the nodes.
> Step 4, Group-level analyses, this is where I am most confused. When I conducted conventional volume-based searchlight, I did the analysis at subject's native space, and normalize searchlight result to, say MNI space. Do I need to do any transformation here at all? Is this standard surface thing already taken care of from AFNI's MapIcosahedron or by pymvpa2-prep-afni-surf?

When using surfaces resampled to standard topology (e.g. using MapIcosahedron), then there is no need to transform the surfaces.

> Step 5, shall I smooth the node-based searchlight results?

Searchlight maps are usually already smooth because neighboring searchlights share many voxels. Generally I don't smooth searchlight maps.

Hope that helps.
best,
Nick

[Frances Jin]

Thank you all very much ! Super helpful !