Print

Print
Awesome,
 
Thanks for the advice Stephen!
 
To be clear, the way I interpret what you are saying below would break down into the following steps:
 
1. Run FEAT without ICA denoising to do first regular preprocessing steps, registration, and such
2. Run FEAT on the prior FEAT output from step one, without further preprocessing ?, except ICA denoising
3. Select the IC from the output of step 2 that appear to be noise, and remove these IC from the data by the commands line in the terminal as outlined in the FEAT ICA instructions.
4. Run FEAT again, without further preprocessing steps?,  to output the ICA denoised 4D functional data
5. Run MELODIC on the output from step 4 (In the MELODIC GUI there is an option to run without preprocessing and registration? I see this option in FEAT, but for some reason not able to locate this in MELODIC, maybe looking too hard)
 
Super Thankful,
Varina
 
From: FSL - FMRIB's Software Library [[log in to unmask]] On Behalf Of Stephen Smith [[log in to unmask]]
Sent: Friday, September 16, 2011 3:53 AM
To: [log in to unmask]
Subject: Re: [FSL] Resting State - Better to denoise first with FEAT with ICA then MELODIC?

Hi - the initial ICA is run on the filtered_func_data fully preprocessed 4D data - and it is this file that you should apply the ICA-based denoising to.
Then, when you feed this into the further ICA analysis, you should not apply any further preprocessing again.

Apart from that this looks ok,
Cheers.


On 15 Sep 2011, at 21:39, Varina Wolf wrote:

Good afternoon,

Sorry about the hyper-vigilance, but am working pretty much solo and need to be sure of myself...

I have single 10 minute runs of resting state fmri- one run per subject  

I took the following steps, as below, and my question is if this is the best way of going about it?  (image is 37 slices, 3.4 mm thick)
1.  BET structural image for each individual and the single MNI 2 yo image.  
2. FEAT with:
DATA
- high pass filter at 100.  (I tried lower numbers like 65 and seemed to allow too much noise)
PRE-STATS
- motion correction with MCFLIRT
- B0 unwarping
- Slice time correction for interleaved
- Brain extraction
-Smoothing 5 mm
-highpass temporal filtering
- MELODIC ICA data exploration
STATS
-FILM prewhitening
-motion parameters added to moodel
POST-STATS
-cluster threshold 2.3, P 0.05 (although don't think I really need this step, but was interesting to see what was activated by this analysis)
REGISTRATION
-Main structural with Linear Normal Search and 12 DOF
-Standard space with Linear Normal Search and 12 DOF

Then I viewed the 174 ICA noise sources and picked the ICs that were most obviously noise and reran FEAT renduring the  Denoised ICA functional data, which I then went to MELODIC:
DATA - same as above
PRE-STATS - same as above
Registration - same as above
STATS
- Variance-normalise timecourses
-Automatic diminsionality estimation
-Single session ICA
POST-STATS
-Threshold IC maps 0.5
- No model or contrasts as this is purely resting state data

It looks like I have redundancy in my steps between FEAT and MELODIC, but not sure about the work around with denoising... and motion correction added in... Although do remember Christian Beckmann saying this was Gaussian source of noise so would be accounted for in MELODIC.  

Most Gratefully,
Varina



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