fatcat command help p

AFNI Message Board
1

Hi,

So I had a few questions in regards using the FATCAT toolbox. I

So a little background… Basically, I’m trying to combine resting state fmri with dti (copying the example given in the FATCAT paper but with my own data).

So first off, in the example given in the paper. Spatial ICA is used in native functional space with FSL-melodic.

First question I have from this is how to run spatial ICA? I read many documents to make sense of this. There are some papers that claim you can run ICA either in sICA (spatial ICA) or tICA (temporal ICA). But there are no settings in MELODIC that distinctly distinguishes the two? I have come across a paper however that discusses when working with rs-fmri data it is useful to apply sICA because of the resting-state networks are not consistent temporally. So it was advised to concatenate all subjects data temporally first (by running a group PCA reduction on each subject, then concatenate it, and lastly perform a PCA reduction on the concatenated work). Once PCA is performed, one can run ICA, and after a backward reconstruction? Also, how would I do this using a code?
Another method I read online was to download a toolbox: GIFT (Group ICA of fMRI Toolbox) which is a toolbox that works on MATLAB that apparently does all the group independent component analysis with PCA and all, automatically.

Lastly, I’ve also come across a document that used MELODIC for group ICA (multi-session temporal concatenation).

p.s. I’m aware that there is ICA done by averaging all the subjects and then there’s the one where you study each individual subject on its own (the ICA) and can perform in within group and between group study using a t-test?

So my question is: which method is more accurate? Maybe it’s a bit of information overload but now it seems that there are many options and no clear path of what to do.

My next question is when it claims that has to do with the mention of using sICA in native functional space. Does that mean that it wasn’t registered onto standard space? So in MELODIC, under the registration tab, was ONLY the structural image and functional image entered? And not the standard space? And if so, why would that be the case? Also, in the example given in the FATCAT paper, rs-fmri and DTI were combined. However, wasn’t the fMRI image supposed to be transformed to DWI space? And if so, how can I do that? From what I understand, is that fMRI images are supposed to be translated to standard space, then translated into DWI data? I’m a little confused about this, and the concept of having to translate into different spaces?

Last question I have is how do I obtain the z-score? I noticed MELODIC only outputs the z-threshold? Many papers report the z-score, so I’m wondering how to obtain that value with the given output from MELODIC.

Any help would be greatly appreciate it. I apologize for the lengthy question. Hope to hear from you soon .

2

Hi, Sondos-

So first off, in the example given in the paper. Spatial ICA is used in native functional space with FSL-melodic.
So, yes, that’s true. One could also use seedbased correlation for an analogous analysis. The main point of that example is to use a subject’s own data (in that case, resting state FMRI) to generate a network of GM targets among which it would be meaningful to look for associated WM via DTI-based tractography.

First question I have from this is how to run spatial ICA? I read many documents to make sense of this. There are some papers that claim you can run ICA either in sICA (spatial ICA) or tICA (temporal ICA). But there are no settings in MELODIC that distinctly distinguishes the two? I have come across a paper however that discusses when working with rs-fmri data it is useful to apply sICA because of the resting-state networks are not consistent temporally. So it was advised to concatenate all subjects data temporally first (by running a group PCA reduction on each subject, then concatenate it, and lastly perform a PCA reduction on the concatenated work). Once PCA is performed, one can run ICA, and after a backward reconstruction? Also, how would I do this using a code?
I’m afraid that this really isn’t a forum for operating MELODIC. I’ll just note that spatial ICA was run in the above example, because the matrix dimensions wouldn’t permit tICA (one would need a loooot of time points or to have averaged voxels together to make a small number of ROIs). I believe that PCA is used as an initial step on the way to ICA results, amongst other preprocessing. However, those questions about detailed running will have to be addressed to those software makers+maintainers.

Another method I read online was to download a toolbox: GIFT (Group ICA of fMRI Toolbox) which is a toolbox that works on MATLAB that apparently does all the group independent component analysis with PCA and all, automatically.
Again, those questions about the technical differences of those implementations would have to be addressed to those experts and/or to those who use the software a lot.

Lastly, I’ve also come across a document that used MELODIC for group ICA (multi-session temporal concatenation).
p.s. I’m aware that there is ICA done by averaging all the subjects and then there’s the one where you study each individual subject on its own (the ICA) and can perform in within group and between group study using a t-test?
Sure, you could map a lot of subjects’ FMRI data to standard space, and use group ICA to get group-wide network maps. Those could then be mapped to individual’s DTI spaces and tracked. The main aim is still the same as above: use FMRI data to generate meaningful networks of GM to track, whether based on an individual or a group. The group-based one would have the benefit of higher SNR from pooling many subjects, but the detraction of less direct specificity of individual differences (from the same reason: pooling many subjects). In reality, practical considerations might make pooling subjects the more appealing option-- depends on the situation, data and questions being asked.

So my question is: which method is more accurate? Maybe it’s a bit of information overload but now it seems that there are many options and no clear path of what to do.
That is the world of MRI in general, I think! I’m sorry that I can’t pinpoint a best way through. One could also use non-ICA approaches, such as AFNI’s Group InstaCorr as a way to combine seedbased correlation with group level analysis, or other methods that determine meaningful GM networks.

My next question is when it claims that has to do with the mention of using sICA in native functional space. Does that mean that it wasn’t registered onto standard space? So in MELODIC, under the registration tab, was ONLY the structural image and functional image entered? And not the standard space? And if so, why would that be the case? Also, in the example given in the FATCAT paper, rs-fmri and DTI were combined. However, wasn’t the fMRI image supposed to be transformed to DWI space? And if so, how can I do that? From what I understand, is that fMRI images are supposed to be translated to standard space, then translated into DWI data? I’m a little confused about this, and the concept of having to translate into different spaces?
Depending on how you proceed, you can map your FMRI data to group space and then take results to individual DWI spaces. For example, if pooling subjects as discussed above, that might be necessary. If you wanted to use something like FreeSurfer parcellations of anatomical dsets, you would not have to do that (but it’s a different type of data being used to make meaningful GM networks).

Last question I have is how do I obtain the z-score? I noticed MELODIC only outputs the z-threshold? Many papers report the z-score, so I’m wondering how to obtain that value with the given output from MELODIC.
This is also a question for the MELODIC-makers. (Though, I thought it was directly output?)

Any help would be greatly appreciate it. I apologize for the lengthy question. Hope to hear from you soon .
A lot of these actually sound like technical questions for other softwares, if you want to use ICA, that is. It will also tie in with your study design and questions(s) of interest-- which also would be something I can have limited input on, except possibly some of the technical considerations when walking down various paths.

–pt