Hello,
(sorry, slightly misunderstood your experimental design in my previous
response- Narender is quite correct that it's better to implcitly model the
'baseline' task, i.e. not give it a separate column in the design matrix).
I am interested however in your group by task interaction. A random effects
analysis is relevant to generalisation to the population, this is clearly
impossible (and not actually the question of interest) in a single case
study. I still don't really follow your design but am I to assume that your
desire to analyse single patients (in comparison to a control group)
reflects the fact that you have a heterogeneous set of patients who cannot
be meaningfully grouped together? If this is so maybe it would be simpler
to use your control group data (expressed at the second level of analysis)
to generate a series of regions of interest which might then be applied to
your patients (separately). This is side stepping the issue a bit (as I
don't know how you would carry out a comparison of the single con.img
produced by your second level On vs Off analysis in the patients with the
same contrast in a single patient (any answers?)). However, it would have
the advantage that it would bring down your number of comparisons. After
all, to compare the control groups with a series of patients individually,
at each voxel, would involve an immense number of contrasts, so limiting
your number of regions would be a boon.
I shall be very interested to hear thoughts on the best way to approach this
Good luck
Paul Fletcher
At 17:58 01/12/00 +0000, you wrote:
>Thanks for your advice on the model setup.
>
>Can you clarify one point concerning contrasts.
>
>My goal is to average my normal data and use that average to compare to each
>patient's brain. (ie average of 10 vs 1 patient) This approach was used in a
>talairach study.
>My initial attempt at this was to use contrasts once the model set up was
>completed..
>
>ie contrast -1/10 <---nine more of these.. v.s. 1
>
>>From your e-mail, I gather that the process won't be that simple
>ie I need to preform a ramdom effects analysis.
>
>Could you comment on the above. ie confirm that the analysis won't be that
>simple, or yes analysis can be performed this way.
>Thanks for your advice on the model setup.
>
>
>[log in to unmask]
>
>>From: [log in to unmask]
>>To: Gregory Krolczyk <[log in to unmask]>
>>CC: [log in to unmask], [log in to unmask]
>>Subject: Re:model & contrast
>>Date: Fri, 1 Dec 2000 12:54:40 GMT
>>
>>Dear Gregory,
>>
>>As I understand it, in your experimental design you have two groups (Normal
>>Controls and Patients).
>>You want construct a design matrix in which your stimulus is represented as
>>ON for 4 blocks (40 scans duration), and the first block starts at scan 14.
>>All the others scans occur during the OFF condition.
>>
>>I assume that you are primarily interested in localising areas in which
>>there is an interaction between Stimulus state (ON vs. OFF) and Group
>>(Normal vs. Patient).
>>
>>You should therefore construct a design matrix that includes all your
>>subjects from both groups (e.g. subjects 1-10, nomal; subjects 11-20,
>>patient).
>>
>>For a given subject, the simplest way of analysing this data would be to
>>represent ON and OFF in the same column.
>>
>>To acheive this, you should specify the number of trial types as 1 (instead
>>of 2). Then,
>>
>>Stochastic Design, NO.
>>SOA, variable.
>>vector of onsets, [enter onset times of ON blocks] In your case, this is 14
>>94 174 254
>>variable durations, NO.
>>parametric modulation, NONE.
>>Are these trials: EPOCHS.
>>Select type of response: e.g. Box-car.
>>Convolve with hrf: e.g. YES....
>>Epoch length, 40
>>Interactions among trials, NO
>>
>>User specified regressors [1]...
>>Enter user speficied regressor [if you have any]...
>>
>>The second part of your enquiry relates to contrasts.
>>
>>On vs. OFF for single subject (first level analysis):
>>For a given subject, if you wanted to find areas in which activity related
>>to ON was greater than activity related to OFF, you would use a t-contrast
>>of +1. To find areas where the inverse was true, you would simply use the
>>contrast -1.
>>
>>Group x Condition Interaction (second level analysis):
>>In order ro arrive at a meaningful interaction result, you would really
>>need to perform a random effects analysis on the Con images derived from
>>the first level analyses of each subject. to derive a meaningful result for
>>the interaction. There are several previous emails in the archives relating
>>to how this is done practically.
>>
>>I hope this is helpful. If you wish, I could send you an email of the
>>SPM_fMRIDexMtx.mat file for a single subject.
>>
>>Best wishes,
>>
>>Narender Ramnani
>>
>>___________________________________
>>Narender Ramnani, PhD.
>>
>>University Laboratory of Physiology,
>>University of Oxford.
>>
>>FMRIB Centre
>>John Radcliffe Hospital
>>Oxford.
>>___________________________________
>>
>>
>>
>>In message <[log in to unmask]> Gregory Krolczyk
>><[log in to unmask]> writes:
>> > SPMers help!
>> >
>> > I am a new spm user, and I trying to come up with a fMri experimental
>>model
>> > and contrast. Please see if my logic is at fault/imcomplete. I have
>>tried
>> > your on line tutorials and mailing list archives.
>> >
>> > My experiment consists of on (stimulus) and off times. In terms of
>>frames,
>> > the breakdown is
>> >
>> > Condition Frames length
>> > OFF 1-13 13
>> > ON 14-53 40
>> > OFF 54-93 40
>> > ON 94-133 40
>> > OFF 134-173 40
>> > ON 174-213 40
>> > OFF 214-253 40
>> > ON 254-293 40
>> > OFF 294-319 25
>> >
>> > My first problem is with the experimental model. I'm looking for a
>>checkered
>> > pattern ie
>> >
>> > M
>> > M
>> > M
>> > M
>> >
>> > but have not achieved one when I fill SPM in as follows...
>> >
>> > Interscan interval 2.24
>> > scans 319
>> > trails 2
>> > SOA variable
>> > vector on 14 94 174 254
>> > vector off 1 54 134 214 294
>> > NO variable durations
>> > epoch lengths on 40
>> > epoch lengths off 13 40 40 40 25
>> > regressors 0
>> >
>> > PROBLEM, when I look at the fmri design for off, epoch lenth for off is
>> > 13... not 13 40 40 40 25... Why is that....?
>> >
>> > Remedy... have three variables with the first being just 1-13 (off)
>> > The second being the on's
>> > The third being the rest of the off's
>> >
>> >
>> > Problem two (clarification)...
>> >
>> > I now want to take my model and form contrasts. I have 10 normals and I
>> > would like to compare to each and every diseased...
>> >
>> > would the contrasts be
>> >
>> > -1/10 0 <---nine more of these.... 1 0
>> >
>> > note -1/10 represents the on period
>> >
>> >
>>__________________________________________________________________________
___________
>> > Get more from the Web. FREE MSN Explorer download :
>>http://explorer.msn.com
>
>___________________________________________________________________________
__________
>Get more from the Web. FREE MSN Explorer download : http://explorer.msn.com
>
-----------------------------------------------------------------
Paul Fletcher,
Research Department of Psychiatry,
University of Cambridge,
Addenbrooke's Hospital,
Hills Road,
Cambridge,
UK
CB2 2QQ
Tel 01223 336 988
Fax 01223 336 581
|