Dear Dr. Richard Perry and other SPM experts,
With regard to the Dr. Kyoung-Min Lee's question,
what do you think about the following analysis using EXCLUSIVE masking?
Masking the simple effect of Act2 vs Act1a (0 -1 1 0 0 0) with
the simple effect of Act3 vs Act1b (0 0 0 0 -1 1) in an EXCLUSIVE way,
namely, showing voxels which are significantly activated in the comparison
of Act2 vs Act1a and are NOT significantly activated in the comparison
of Act3 vs Act1b.
or
Masking the simple effect of Act2 vs Act1a (0 -1 1 0 0 0) with
the jointed effect of 'Act2 vs Act1a' and 'Act3 vs Act1b' (0 -1 1 0 -1 1)
in an EXCLUSIVE way, namely, showing voxels which are significantly activated
in the comparison of Act2 vs Act1a and are NOT significantly activated in the
jointed comparison of 'Act2 vs Act1a' and 'Act3 vs Act1b' .
Such kind of masking can be done by slightly modifying a part of the
spm_projections_ui.m,
and as far as I know, spm in MEDx supports this.
I wonder if there are any statistical reasons why SPM supports INCLUSIVE
masking alone.
Any comment would be appreciated.
Regards,
P.S. To Dr. Richard Perry
I send the same e-mail by mistake to you instead of to the spm mailing list.
Hope this e-mail or the previous one will be discarded. Thanks.
_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/
Kota KATANODA
Department of Cognitive Neuroscience,
Faculty of Medicine, University of Tokyo
7-3-1 Hongo Bunkyo-ku Tokyo Japan
ZIP: 113-0033
TEL: +81-3-3812-2111 (Ext. 3574)
FAX: +81-3-5802-3329 (Should be addressed to KATANODA)
E-mail: [log in to unmask]
_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/
At 0:11 AM +0900 99.3.5, Richard Perry wrote:
>Dear Dr Lee,
>>
>>When conditions are Rest1 Act1 Act2 Rest2 Act1 Act3,
>>would it be reasonable to mask contrast (0 -1 1 0 0 0)
>>by (0 -1 1 0 -1 1) in order to get voxels that show higher signal only in
>>comparison between Act2 and Act1, and not in comparison between Act3 and
>>Act1 ? Or, would it be better to mask it by (0 0 0 0 -1 1) ?
>>Any comments will be greatly appreciated.
>>
>Masking one contrast with another shows voxels which satisfy the criteria
>of BOTH contrasts. Thus for example, masking contrast 0 -1 1 0 0 0 with
>contrast 0 0 0 0 -1 1 will yield voxels which give more signal in Act2 than
>in Act1 AND show more signal in Act3 than in Act1, which is clearly not
>what you are after.
>
>As you are clearly aware (from the 'subject' entry of your query), what you
>are looking for is a kind of interaction. You will not be able to obtain
>this directly by masking.
>
>If Act1 on the first presentation is different from Act1 on the second
>presentation (eg. because of a session effect), then perhaps we should call
>them Act1a and Act1b respectively. The contrast 0 -1 1 0 1 -1 will give
>you voxels in which the effect Act2 - Act1a is greater than the effect Act3
>- Act1b.
>
>However, you should be aware of the fact that this interaction may include
>voxels in which there is no significant difference between Act2 and Act1a,
>but in which Act3 produces far less signal than Act1b. If you want to
>exclude these voxels (as your wording of the question seems to imply), then
>you should mask the 'interaction' contrast 0 -1 1 0 1 -1 with the 'simple
>effect' Act2 vs Act1a, for which the contrast is 0 -1 1 0 0 0.
>
>Incidentally, if there is absolutely no difference between Act1a and Act1b,
>as might be the case in a counterbalanced group study in which half of the
>subjects saw Rest Act1 Act2 Rest Act1 Act3 and the other half saw Rest Act1
>Act3 Rest Act1 Act2, then you would expect the interaction to collapse to
>the effect Act2 vs Act3 (contrast 0 0 1 0 0 -1), in which case using 0 -1 1
>0 1 -1 would just introduce unnecessary noise,
>
>Best wishes,
>
>Richard Perry.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|