Hi Steve,

Is it not normally recommended to analyze non-resting-state data using FSL MELODIC? 

To clarify, for the finger tapping test, we first used fslsplit and fslmerge to separate and concatenate the functional data into two sets of data: finger_tapping_rest and finger_tapping_task. Then we ran two separate FSL MELODIC with the two files (i.e. one with finger_tapping_rest and one with finger_tapping_task).

Is there a problem with this methodology?

Thanks for your comment!

Liang

---

Date: Fri, 30 May 2014 16:37:20 +0100
From: [log in to unmask]
Subject: Re: [FSL] Running MELODIC on Prospective Data
To: [log in to unmask]

Hi Liang - I'm a little confused: we would not normally recommend doing a functional connectivity analysis on a rest-task dataset....?

Cheers, Steve.

On 28 May 2014, at 19:38, Liang <[log in to unmask]> wrote:

Dear FSL Experts,

I'm attempting to analyze fMRI data collected from a prospective intervention study consisted of an experimental group and control group. Both groups were scanned at baseline and 12-months (post-intervention). During the scanning sessions, two tasks were performed:

1. Block-design finger tapping task.
2. Event-related design cognitive task.

What I'm interested in finding is to determine if there are differences in changes in functional connectivity/network over the 12 months between groups.

Pleas correct me or point me to the right direction if my following rationale is wrong: I came up with two potential ways to perform the analysis for the block-design finger tapping task:

1. I would first input only the baseline functional data into MELODIC - TICA; then I would do the same for the 12-months functional data and end up with two Melodic_IC.nii.gz files that contains the group components (one for baseline and one for 12-months). Subsequently, I can use fslmaths -sub with these two files to calculate the difference between scanning sessions (12-months minus baseline). Then use the output file from fslmaths as the input for Dual Regression to find group differences.

or,

2. I would select both baseline and 12-months as inputs into MELODIC-TICA, then define the contrasts in the GLM that will be later inserted into Dual Regression, for example:

        Experimental_baseline     Experimental_12M      Control_baseline        Control_12M
C1                   -1                                    1                                 0                             0
C2                    0                                     0                                -1                             1
C3                    1                                     -1                                0                             0
C4                     0                                    0                                  1                             -1
C5                    1                                      0                                 -1                             0
C6                     0                                     1                                   0                            -1

Where C1-C4 determine differences within group over time; C5 and C6 determine differences between group within the same scanning session.


Next, for the event-related design task, I will apply either one of the above rationale to perform MELODIC - Multi-session Temporal Concatenation, and in the GLM setup I will load the text files containing stimulus onset information?

I apologize for the lengthy question but I would really appreciate any insight or advice on how to approach this analysis,

Thanks in advance!

Liang

---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director,  Oxford University FMRIB Centre

FMRIB, JR Hospital, Headington, Oxford  OX3 9DU, UK
+44 (0) 1865 222726  (fax 222717)
[log in to unmask]    http://www.fmrib.ox.ac.uk/~steve

---------------------------------------------------------------------------

Stop the cultural destruction of Tibet