Dear Yoshitaka & Mirko,
I briefly have access to MatLab4 and SnPM99b, and have had a look into your
problems with Suprathreshold Cluster size tests in SnPM96b
Unfortunately I am unable to replicate your problem, however, I think I
know what the problem is.
Basically, MatLab4 uses two forms of indexing: You can index using a "mask"
vector of 1's and 0's, or you can index the locations directly. When a mask
vector has length 1 MatLab4 thinks it's an index, which can lead to trouble
if the "mask" vector was zero: "Index exceeds matrix dimensions".
We were using a "mask" vector for filtering out interesting voxels: It
looks like your plane 36 only has a single voxel in it, for which the
statistic is not above the (internal) threshold for identifying
suprathreshold clusters, so the index "mask" is zero, causing the problem!
You can pick up a fixed spm_snpm.m (version 1.17) from:
ftp://ftp.fil.ion.ucl.ac.uk/spm/snpm96b_updates/spm_snpm.m
Let me know if this doesn't fix the problem!
Hope this helps, and sorry for the delay...
-andrew
At 23:38 06/10/99 -0400, Yoshitaka KUMAKURA wrote:
| Thank you very much for your kind reply, and Really sorry to be late
| in expressing appreciation. The message we got from you helped us a
| lot, and we can understand your situation. We realized our
| explanation was a little short. Please forgive us to ask you some
| details again.
|
| > At 16:57 09/09/99 -0400, Yoshitaka KUMAKURA wrote:
| >| I work at the MNI with Prof. M. Diksic and we would like to use non-
| >| parametric version of SPM (SnPM) in our comparisons of two groups.
| >| Whenever we try it, it seems to us that there are some problems
| >| which we can not find solution to. We have also posted two
| >| questions on the Web site, however we have not received any
| >| answer so far.
| >
| > I'm afraid the two of us who developed the non-parametric
| > SnPM software have moved on, and unfortunately don't have much
| > time for SPM/SnPM software development!
| >
| > My ability to help with SnPM is particularly difficult now,
| > as I've just lost my access to Matlab 4.2c!
| >
| >| We are trying to use SnPM suprathreshold cluster size statistics
| >| for the comparison of two groups. The data are generated by
| >| PET and alpha-methyl-L-Tryptophan which enables to evaluate
| >| serotonergic synthesis in the brain. The two groups are
| >| constituted by 11 psychiatric patients and 11 normal controls.
| >| If we can get a comment about following problem from you, it
| >| will be great help for us.
| >|
| >| The compute process (spm_snpm.m) to collect suprathreshold
| >| cluster size resulted in an incomplete termination with the
| >| following message.
| >|
| >| Plane 35: 0'10" ( 0% Sm)
| >| Plane 36: ??? Index exceeds matrix dimensions.
| >|
| >| Error in ==> __________________/matlab/snpm/spm_snpm.m
| >| On line 475 ==> spm_append('SnPM_ST',[...
| >
| > This usually indicates that the information on image
| > dimensions in the header doesn't match the data in the image file.
| >
| > Does the SnPM analysis work if you don't collect supre-threshold
| > cluster statistics? (Can you analyse this data in SPM96?)
|
| Yes. We can analyse this data in SPM96, and SnPM without collecting
| supre-threshold cluster statistics works well .
|
| We would also like to ask about another example using 6 males and 5
| females. (The number of samples are lesser than above example.) When
| we tried comparing 6 males and 5 females in the same manner, the
| compute process was not interrupted. However, the SnPM_ST.mat (or
| SnPM_STC.mat?) file was not produced in working directory. We
| suppose these files for cluster size statistics might have been
| deleted during the calculating process becase of excess of limited
| file size for our UNIX system as mentioned in SnPM primer.
| Meanwhile, SnPM calculation processes without suprathreshold cluster
| size statistics work well in all kinds of examples we tried.
| However, the results without using suprathreshold cluster size
| statistics showed unfortunately no significant area. This is why we
| would like to try suprathreshold cluster size statistics and pick up
| some pixels somehow. Our target area has large number of pixels and
| not so high peak in pixel values. So, I would lile to know what we
| should do to fix our SnPM problem, how extent we can expect for
| effectiveness of suprathreshold cluster size statistics in analysis
| of our data, and if there is another solution to meet our purpose.
|
| >| The main parameters set up in SnPM_cfg.mat are as follows.
| >|
| >| sDesFile = spm_snpm_MG2x (version 1.5)
| >| sDesign = Two Group, 1 scan per subj: 11(GrpA),11(GrpB)
| >| sPiCond = 200 permutations of conditions, bhPerms=0 (approx. test)
| >| vFWHM = 0 0 0
| >| No volumetrically.
| >
| > Sounds OK.
| >
| > Make sure you have downloaded all the bug fixes from:
| > ftp://ftp.fil.ion.ucl.ac.uk/spm/snpm96b_updates
| > (although non specifically relate to the problem you are having).
|
| We confirmed it again. It is sure we are using the latest version.
+- Dr Andrew Holmes [log in to unmask]
| ___ __ __ Wellcome Department of Cognitive Neurology |
| ( _)( )( ) Functional Imaging Laboratory, Stats & |
| ) _) )( )(__ 12 Queen Square, Systems |
| (_) (__)(____) London. WC1N 3BG. England, UK |
+------------------------------------- http://www.fil.ion.ucl.ac.uk/ -+
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
