 |
 |
Tom the suggested change works as predicted, by additionally commenting out lines 563 and 565 related to the "if" loop. However, if no voxels survive thresholding there's a rather spectacular core dump relating to passing a 0-voxel k value via line 534 to spm_results at line 260. I can live with this in the meantime until I have a look at that. Many thanks for your help. Regards - Mile > -----Original Message----- > From: Thomas E. Nichols [mailto:[log in to unmask]] > Sent: Thursday, January 20, 2005 2:23 PM > To: Glabus, Michael > Cc: [log in to unmask] > Subject: Re: Tom Nichol's conjunction analysis : spatial extent > > Mike, > > > I have installed these modifications and they now work very well. > > It seems that it's possible to exert complete control of the > > conjunction via the three options. > > Great. Glad to hear it. > > > However, why is it not possible to specify the spatial extent? I > > accept that in terms of conjunction analysis that might > violate some > > primary rule but an explanation would be welcome - I'd like > to tidy up > > some of the results by excluding single voxels/small clusters. > > Good point. There is no reason to forbid it, though there is > no theory to provide the P-values. You should be able to > apply an arbitrary threshold. If you delete " & nc == 1" > from line 540 of the revised spm_getSPM.m you should be asked > for an extent threshold. > > > If you use an Intermediate or the Global Null, you'll get no > P-values, as there is no theory for conjunction spatial > extent (yet... Keith Worsley had a student working on it). > If you use the Conjunction Null you'll get P-values, but > they'll be wrong, fortunately, in the conservative direction. > > (Background: The random field theory for obtaining the > P-values will be assuming that the statistic image is not a > conjunction, rather just a single statistic image. Since a > conjunction is effectively an intersection of images, and, > all things equal, an intersection of images will have smaller > clusters than a single image, the actual clusters will be > smaller than the theory expects, and hence the P-values will > be too large). > > > Please implient this change in spm_getSPM.m and let me know if works. > If it does, I'll post the change to spm2_updates. > > > > Also, is there a preprint of the manuscript explaining the > > intermediate stage, alluded to below? > > It should be appearing soon, but maybe Karl/Will can post it? > > -Tom > > > -- Thomas Nichols -------------------- Department of > Biostatistics > http://www.sph.umich.edu/~nichols University of Michigan > [log in to unmask] 1420 Washington Heights > -------------------------------------- Ann Arbor, MI 48109-2029 >