Print

Print


Dear Johan,

I found the same problem when doing an analysis with more than 10000 EPI volumes and 2000 regressors (including motion nulling regressors).
As you said, my workaround to continue the analysis with SPM was also to modify my SPM files so that it saves the SPM structure as Matlab v7.3.
The problem I next faced was that loading the SPM matrix each time I wanted to check the results for a contrast took more than 20 minutes. That was inviable.
Therefore, my solution was to specify all contrasts in advance with the batch interface and check the results with AFNI where I could explore the results easier (e.g. changing the threshold). Nevertheless, when overlaying the statistical maps in AFNI one needs to take into account that the statistical maps mean nothing without degrees of freedom. But you can also get the degrees of freedom of the contrast, either because you know them in advance, or getting SPM plot the results also in batch mode.  Within the generated .eps file, one can find out the degrees of freedom.

Hope this helps,
Cesar


On Mar 31, 2011, at 4:17 PM, Johan Carlin wrote:

Dear SPM experts,

I am currently analysing an extremely large dataset (13200 EPI volumes in a single first-level model), and am running into trouble with SPM. 

SPM forces the SPM.mat file to be saved as a Matlab v6 file (e.g. lines 873-883 in spm_spm), which limits its size to 2 GB. This means that the SPM.mat is never saved at the end of spm_spm if the model is too large.

I circumvented this behaviour in my local SPM install, and although saving as a Matlab v7.3 seems to work for a while (ie, I can also get through spm_contrasts after forcing this one to save v7.3 too), the SPM.mat ends up corrupted quite quickly (Loading it in Matlab fails). I guess this is why SPM doesn't use this format by default?

Anyway, my question is: Short of breaking my analysis into separate models, is there a way to analyse large datasets in SPM8 without bumping into this size constraint?

Any help would be greatly appreciated,

Johan Carlin

--
Johan Carlin
Graduate Student
MRC Cognition & Brain Sciences Unit
15 Chaucer Road
Cambridge CB2 7EF
UK

[log in to unmask]
+44 (0)1223 355294 ext 593

Cesar Caballero Gaudes
Hôpitaux Universitaires de Genève
CIBM / Radiologie
Rue Gabrielle-Perret-Gentil 4
(ex-rue Micheli-du-Crest 24)
CH-1211 Genève 14 
Switzerland
 
[log in to unmask]
Phone: (+41) 0223725217