Dear Fenna,
Concerning your first point: anyone wanting access to any specific
revision of SPM can email me their request and I'll provide them with a
link to download that version.
The DCM.mat file stores the SVN revision number of various files, as a
first step to store processing provenance:
* SPM.version: SPMxx (help text of Contents.m)
* SPM.revision: rxxxx (help text of Contents.m). Internally, this
actually contains 'beta' (that's the only difference with the public
version of SPM) as you've noticed in Adeel's file.
* DCM.version: DCMxx (stored in spm_dcm_ui.m)
* DCM.revision: rxxxx (spm_dcm_fmri_csd.m or spm_dcm_estimate.m)
The dates of the revisions you've mentioned are:
r5919 | adeel | 2014-03-14 13:19:21 +0000 (Fri, 14 Mar 2014)
r6080 | guillaume | 2014-07-01 17:00:22 +0100 (Tue, 01 Jul 2014)
r6142 | adeel | 2014-08-27 17:10:29 +0100 (Wed, 27 Aug 2014)
Best regards,
Guillaume.
On 15/09/14 16:52, Zeidman, Peter wrote:
> Hi Fenna,
>
> Replicating the results from the Neuroimage is a very useful project.
>
>
>
> Just a specific comment on your second point. If you run the attention
> dataset batch script with SPM12, you will indeed find the ROIs are
> shifted compared to those described in the manual, and those you’d get
> with SPM8. I’m not sure which coordinates Adeel and colleagues used in
> their paper, but if you want ROIs which exactly match previous versions,
> you can use the attached script. A script like this will probably appear
> on the SPM website in due course. (The difference is that the
> co-ordinates are hard coded in this script to match those found using
> SPM8, rather than allowing SPM to automatically choose the peak
> coordinates.)
>
>
>
> Adeel is best placed to confirm whether these coordinates match those
> used in his analysis.
>
>
>
> Best,
>
> Peter.
>
>
>
> *From:*SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]]
> *On Behalf Of *Fenna Krienen
> *Sent:* 15 September 2014 15:15
> *To:* [log in to unmask]
> *Subject:* [SPM] Issues replicating cross-spectra analysis in Friston et
> al 2014, cont.
>
>
>
> Hi List,
>
> We have been having issues replicating the cross-spectra DCM based on
> the attention dataset as specified in the Friston et al 2014 Neuroimage
> paper, 'A DCM for resting state fMRI'. We have been corresponding
> off-list with Adeel, who has been very helpful in assisting us to debug.
> I'm now copying the list in case others are interested or have any insights.
>
> To summarize the current situation, we are trying to replicate the
> directions and magnitudes in Figure 7 of Friston et al. 2014. When we
> try to run the DCM using the attention dataset available on the SPM
> website, we do not arrive at the same effective connectivity values as
> shown in the paper. One of the authors (Adeel) kindly sent us the
> DCM.mat file used in that paper. In our hands, the values in this .mat
> file replicate the connectivity directions (positive and negative
> effective connectivity at the correct positions in the visual
> hierarchy). However, the magnitudes don't quite match the published
> results.
>
>
>
> Inline image 2
>
>
>
> At this point, we can think of a few possibilities:
>
> *1) Different SPM and DCM versions.*
>
> We are using SPM12b, r.6080, DCM v.5919. The DCM.mat file that Adeel
> sent us used SPM12b r. 'beta', DCM v. 6142. Therefore version difference
> might be an issue. To figure this out, can SPM folks either (A) let us
> know how we can obtain the versions used in the paper (SPM12b 'beta',
> DCM revision 6142)? Alternatively, can anyone involved in that paper try
> to repeat your DCM analysis using the same version as is currently
> publicly available (i.e., SPM12b revision 6080, DCM revision 5194)?
>
> *2) Different data.*
>
> Friston et al. 2014 state that the attention dataset used in that paper
> can be downloaded from http://www.fil.ion.ucl.ac.uk/spm
> <http://l.ion.ucl.ac.uk/spm>. We downloaded the attention dataset from
> here: http://www.fil.ion.ucl.ac.uk/spm/data/attention/ (attention.zip).
> When we tried to replicate the VOIs from Adeel's DCM.mat (or Table 2 of
> the paper), SPM shifts several of our VOIs by a little, resulting in
> different coordinates and different numbers of voxels included compared
> to the paper. We did a scatterplot of Adeel's VOI timecourses and ours,
> and the two sets of timeseries are correlated but not exactly the same.
> To really figure out if it's a data issue, can anyone confirm that
> attention.zip on the website is what was used in the paper?
> Alternatively, can folks involved in that paper repeat your DCM
> analysis, using the attention.zip data on the website?
>
>
>
> Inline image 1
>
> *3) We are running the DCM incorrectly* despite the instructions Adeel
> was kind to describe for us in an earlier posting. To figure out if this
> is the issue, can anyone post both the SPM.mat and DCM.mat using
> SPM12b revision 6080, DCM revision 5919 (i.e., same versions as what we
> have), so we can check if any variables are different?
>
> We think that a combination of the above possibilities might explain why
> we can't replicate the results. If we are failing because of (1) or (2),
> then we are a bit worried that we do not fully understand how sensitive
> the basic result is to different software versions and datasets?
>
>
>
> Many thanks,
>
> Fenna
>
>
>
> [Our previous posting was: Fri, 29 Aug 2014 11:48:09 -0400 with subject
> "Fwd: [SPM] Issues replicating cross spectra DCM in Friston et al
> 2014"]
>
--
Guillaume Flandin, PhD
Wellcome Trust Centre for Neuroimaging
University College London
12 Queen Square
London WC1N 3BG
|