Dear Frederique,
You're right, there is a problem entering preT and postT data in the same
model by virtue of their differences in TR.
I would suggest that you run the analysis for each subject / time period
separately in a first analysis, then use ImCalc to create a "postT-preT"
difference image from the con images of each subject. Scanner differences
should affect control subjects in the same way as the patients, so the
"postT-preT" difference images can be entered into a fixed effects model to
look for treatment effects on patients vs. control subjects. I believe
this will address the treatment / scanner confounds that Marko pointed out.
The tricky part is that you must use a p threshold of 1.0 on the first
stage analysis so that all postT-preT differences will be saved in the con
file, not just those that exceed some threshold. Otherwise you could end
up with artificial differences that result from some voxels barely
exceeding threshold postT but not preT (or vice versa).
Hope this helps.
Doug
At 08:51 AM 7/19/2006 +0100, Frederique Liegeois wrote:
>Dear Doug,
>
>Many thanks for your reply. I must have misunderstood: how can I have all
>these in the same model (am assuming you mean a Fixed effects where I
>enter all my scans as if they were one big experiment) if my interscan
>interval-ISI- is not the same for preT and postT? The question is asked at
>the beginning of "fMRI /design" and to my knowledge I don't have the
>option of entering say two different ISIs-or can I? There may be a way
>that I don't know about, so apologies for this.
>
>As for inference, these are patients with a unique genetic disorder so
>there was no way to recruit more-and my conclusion will only apply to
>their disorder.
>
>Kind regards,
>Frederique
>
>
> > Dear Frederique,
> >
> > I'm not sure how much you can infer from a sample size of only 4 patients
> > /
> > 4 controls, but in the same model you can try:
> > (postT patients - preT patients) vs. (postT controls - preT controls)
> >
> > This should control for change in scanner as well as change in time.
> >
> > Doug Burman
> >
> > At 06:50 PM 7/18/2006 +0100, Frederique Liegeois wrote:
> >>Dear SPM users,
> >>
> >>I have an unusual question and hope that one of you can help me. I have
> >>conducted a preliminary study investigating the brain changes associated
> >>with a course of treatment in 4 patients. The patients were therefore
> >>scanned before therapy (preT) and after therapy (postT) in a simple Task
> >>vs Rest fMRI block design. To control for the effect of "time elapsed
> >>between preT and postT", I have also scanned matched controls twice at
> >>about the same time, but the controls did not undergo any treatment. In
> >>theory this enables me to explore the brain regions that are under- and
> >>overactive in the patients relative to the controls before and after
> >>therapy, as well as the difference.
> >>
> >>HOWEVER, my problem is that, due to unforeseen circumstances, the
> >>treatemnt did not start at the time originally planned, and our scanner
> >>was REPLACED in the middle of the study.
> >>This means that my preT data were acquired on Scanner#1 and my postT data
> >>on Scanner#2. The TRs and TEs are different.
> >>
> >>Had this not happened, I would have put all the data in a fixed effect
> >>model and tested the interaction:
> >>(preT Patients vs. PreT controls) vs (postT Patients vs. postT controls)
> >>1 -1 -1 1 and -1 1 1 -1
> >>Unfortunately I cannot do that because the scanning parameters are
> >> different.
> >>
> >>I am wondering whether there is any way I can still compare my preT and
> >>post T data within SPM2, using a method that would be acceptable to a
> >>reviewer.
> >>
> >>For instance, would it be acceptable/legal to bring my Patients vs
> >>Controls contrasts (con1_preT and con2_postT) to a second level analysis
> >>to explore the interaction I am intested in?
> >>If so, is that considered a one-sample t-test, a two-sample t-test, or a
> >>paired t-test? These are the same subjects (so repeated measures) but as
> >>they were not scanned on the same machine I am not sure about which one
> >> to
> >>use to be as conservative as possible.
> >>
> >>Should I instead/additionally compare (Patient preT vs Patient postT) and
> >>(controls preT vs control postT) in a second-level analysis (two-sample t
> >>test), and bring those two contrasts to a third-level?
> >>
> >>Is there another way I can process my data that would enable me control
> >>for the effect of scanner change?
> >>
> >>Any help would be greatly appreciated as these treatment studies are very
> >>time-consuming and I really hope I can still use my data.
> >>
> >>Many thanks in advance for your help,
> >>
> >>Frederique
> >>
> >>--
> >>Frederique Liegeois, PhD
> >>UCL Institute of Child Health
> >
> > Dept. of Communication Sciences & Disorders
> > Northwestern University
> > 2240 Campus Drive
> > Frances Searle Building, Room 2-356
> > Evanston, IL 60208
> > phone 847-467-1549
> > fax 847-491-4975
> > email: [log in to unmask]
> >
> >
>
>
>--
>Frederique Liegeois, PhD
>Lecturer in Cognitive Neuroscience
>Developmental Cognitive Neuroscience Unit
>Institute of Child Health, UCL
>30 Guilford Street
>LONDON WC1N 1EH
>U.K.
>
>------
>phone: 44 (0)20 7905 2728
>fax: 44 (0)20 7905 2616
>E-mail: [log in to unmask]
Dept. of Communication Sciences & Disorders
Northwestern University
2240 Campus Drive
Frances Searle Building, Room 2-356
Evanston, IL 60208
phone 847-467-1549
fax 847-491-4975
email: [log in to unmask]
|