Hi Derorah,
answers to your question are in this paper http://journal.frontiersin.org/article/10.3389/fnins.2014.00001/abstract
and the code is here https://github.com/CPernet/spmup/blob/master/spmup_hrf_boost.m
cyril
--
Dr Cyril Pernet,
Centre for Clinical Brain Sciences (CCBS)
Neuroimaging Sciences
The University of Edinburgh
Chancellor's Building, Room GU426D
49 Little France Crescent
Edinburgh EH16 4SB
[log in to unmask]
http://www.sbirc.ed.ac.uk/cyril
http://www.ed.ac.uk/edinburgh-imaging
________________________________________
From: SPM (Statistical Parametric Mapping) <[log in to unmask]> on behalf of Deborah Talmi <[log in to unmask]>
Sent: 11 August 2015 16:29
To: [log in to unmask]
Subject: [SPM] boosting con images with temporal and dispersion derivatives
Hi SPMrs,
I’ve included temporal and dispersion derivatives in my first level design matrices to check whether poor fit to the canonical response explains the limited activations we observe at the second level. To answer the question I ran a full factorial model with my 2 experimental factors (A and B, each with 2 levels) and ‘basis’ as a 3rd factor (3 levels: canonical, temporal, dispersion). There were a large number of FWE<.05 activations corresponding to the F tests for each derivative (across other factors), suggesting that my events were not always modelled optimally with the canonical response.
Questions...
1. In order to improve inference I’ve tried to follow Calhoun’s (2004) ‘boosting’ method to create modified con images. My first question is whether Dr. McLaren’s script (https://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=SPM;fa723f27.0908) is valid for SPM12 (the message mentions version-specific adjustments), what to do with the 2 kinds of beta images the script generated (how to transform those to con images?) and whether it is possible to include dispersion derivatives as well (the script applies only to the temporal derivative).
2. I tried entering 3 con images (the canonical and the 2 derivatives) into the imcalc batch tool for each onset regressor of interest, and modify the first with the function
matlabbatch{1}.spm.util.imcalc.expression = 'sign(i1).*sqrt(i1.^2+i2.^2 +i3.^2)'.
Rik’s reply on one of the threads suggests that this is the correct way to include the dispersion derivative, but how do I adjust this function to reflect the changes in normalization across SPM versions?
3. Finally, our scanning acquisition was a little bit unusual. We used the dual echo sequence (http://onlinelibrary.wiley.com/doi/10.1002/hbm.22463/abstract), where two EPI read-outs are serially collected at a short and longer echo time, and the two images added together. My final question therefore is whether the informed basis set is applicable to this method of acquisition/analysis.
Apologies for the lengthy message – would very much appreciate any insight on these issues.
Cheers, Deborah
--
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.
|