 |
 |
Hi Varun,your pipeline looks OK to me, but I suggest running group ICA *after* cleaning the data.
In fact, running multi-session temporal concatenated ICA in this case is not as a "dimensionality reduction” step, but it’s used to obtain a template of resting state networks to then extract the single-subject versions and compare them across subjects. If you derive the template from noisy data, dual regression will basically try to derive noisy components (which will fit the template) from your cleaned data.
Note that the filtered_func_data.nii.gz files you obtain are from the preprocessing pipeline inside Melodic, not from the ICA decomposition itself.
Summarising, here is what I suggest:- Preprocessing (with Melodic gui for example) without setting group ICA decomposition. Output: filtered_func_data.nii.gz for each subject- Denoising pipeline as you described, up to applywarp (note that we usually do it differently, with single-subject ICA and FIX, but in principle your approach is ok)- Group ICA with temporal concatenation (available with melodic command line only) on the cleaned data to derive melodic_IC- Dual regression on the cleaned data with the clean template
Cheers,Ludovica
—
Ludovica Griffanti, PhD
Postdoctoral Researcher
Wellcome Centre for Integrative Neuroimaging (WIN)
Oxford Centre for Functional MRI of the Brain (FMRIB)
Nuffield Department of Clinical Neurosciences, University of Oxford
John Radcliffe Hospital
Oxford, OX3 9DU, UK
email: [log in to unmask]
On 28 Nov 2019, at 11:40 am, Varun Chandran <[log in to unmask]> wrote:
Dear Ludovica,
Thank you very much for your prompt reply.
Just to clarify if I have done my analysis correctly, I am bringing up my pre-processing and dual regression analysis pipeline.
In the first step, I pre-processed our rs-fMRI dataset and applied dimensionality reduction (20 components) using multi-session temporal concatenated ICA. From this step, we have the output melodic_IC.nii.gz and also the subject-specific filtered_func_data.nii.gz. Then, I used the following commands (below) to clean up the subject-specific filtered_func_data.nii.gz (295 volumes) separately.
Denoising:
Fast : to segment the T1-w data.
Fslmeants : to extract the time series of WM and CSF from the filtered_func_data.
Fsl_motion_outliers : to extract the timepoints explaining the effects of large motion correction (9 parameters).
Text2Vest : to convert the design.txt into design.mat
fsl_glm -i filtered_func_data.nii.gz -d design -o filtered_func_data_cleaned.nii.gz
applywarp : to register the fMRI dataset to the standard space
Then, I applied the dual regression using the following command:
dual_regression p201/resting+.gica/groupmelodic.ica/melodic_IC.nii.gz 1 AT_fMRI_56subjs.mat AT_fMRI_56subjs.con 0 dual_reg08_6 `cat filelist`
An important note that I have used the path for all the subject-specific filtered_func_data_cleaned_std.nii.gz in the `cat filelist`
randomise -i dr_stage2_ic0001.nii.gz -o dr_IC0001 -d AT_fMRI_56subjs.mat -t AT_fMRI_56subjs.con -m mask.nii.gz -T
Could you please help me to ensure that I am not doing anything wrong in this analysis. I would be grateful for your help, thanks !
best,
Varun Arunachalam Chandran
PhD candidate in Neurosciences,
School of Psychology and Language Sciences,
University of Reading,
Berkshire, UK.
Dear Varun,noise removal and dual regression happen at two different stages.
Here is a common pipeline:- preprocessing at the single subject level- noise removal at the single subject level (either regressing out motion, WM and CSF signal or, single subject ICA + manual classification or FIX or AROMA)- registration in standard space- group ICA with temporal concatenation (on the cleaned and registered data)- dual regression to test group differences
You can find more info on our FSL course material: https://fsl.fmrib.ox.ac.uk/fslcourse/in particular the ICA and dual regression practical covers the above steps:
Hope it helps,Ludovica
—
Ludovica Griffanti, PhD
Postdoctoral Researcher
Wellcome Centre for Integrative Neuroimaging (WIN)
Oxford Centre for Functional MRI of the Brain (FMRIB)
Nuffield Department of Clinical Neurosciences, University of Oxford
John Radcliffe Hospital
Oxford, OX3 9DU, UK
email: [log in to unmask]
On 27 Nov 2019, at 8:28 pm, Varun Arunachalam Chandran <[log in to unmask]> wrote:
Dear experts,
This may be a very basic question to post about the dual regression using multi-session temporal concatenated ICA.
I am trying to understand how does the time-series of the white matter and cerebrospinal fluid nuisance signals are being regressed out, when we run the dual regression using multi-session temporal concatenated ICA.
Otherwise, should we just add the time-series of WM and CSF as confounding variables in the GLM, when we run the dual regression after the multi-session temporal concatenated ICA.
Could someone please help me to clarify this doubt? Any help would be very much appreciated, thanks !
best,
Varun
PhD in Neurosciences,
School of Psychology and Language Sciences,
University of Reading,
Berkshire, UK.
########################################################################
To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1
To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1
To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1
To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1
To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1