You rightly figured that this thread would catch my attention, Donald.
;)
Naroa, I think Donald makes some sound recommendations. I wonder why
you thought it necessary to run separate template creation processes on
each patient in combination with the controls? If you can reasonably
argue that your small set of patients deviate from the control
population only in disease/treatment effects and not on a more
fundamental level (i.e., age, gender skew, etc.), then you should be
able to run each patient against a single control template (i.e.,
Donald's number #5 suggestion below). I've seen some of the ADNI
project papers that have taken this approach in regards to group-wise
genotype x GM comparisons, but I would argue that this sort of approach
is best suited to the small, multiple case series examinations. If you
have enough patients and controls to power a group-wise comparison, I
think you might be on safer ground methodologically to throw all
participants into the template creation process; this conservative
approach (I think) guards against biasing your results towards finding
illusory differences between patients and controls. That being said, I
haven't seen a published comparison of the control template method vs.
control + patient template method. If I were a betting man, I would
posit that the former leads to Type I error increases. It's equally
possible that the latter props up the Type II error. I guess the
decision on which to use should be guided by which error type is more
detrimental for the patient population/sample in question and the
size/representativeness of the control sample template.
J
_______________________________
Jeffrey N. Browndyke, Ph.D.
Duke University Medical Center
2200 West Main St., Suite A-200
Durham, NC. 27705
_______________________________
-----Original Message-----
From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]]
On Behalf Of MCLAREN, Donald
Sent: Wednesday, April 22, 2009 12:21 PM
To: [log in to unmask]
Subject: Re: [SPM] individual analysis with DARTEL
A couple of comments:
(1) With this approach you can't compare the effect in one patient to
another because they are different spaces.
(2) Your results are unexpected, if you have a smaller brain, then you
have less tissue present. Thus lower TBV results in less tissue
everywhere. The interesting results are when atrophy in one area
exceeds what would be expected for global atrophy. This is
accomplished by adding TBV as a covariate.
(3) Be careful how you are defining ICV/TBV, CSF is poorly estimated
from the T1 and probably shouldn't be trusted as a quantitative
measure of CSF. I generally use TBV=grey matter volume+white matter
volume as a covariate.
(4) Did you use modulation?
(5) I just had a similar discussion another day about comparing
patients and controls in terms of template creation. We decided the
optimal solution would be to use a set of controls not in the present
study to create a template and then warp the patients and controls in
our study to that template. Alternatives include: (a) your approach --
multiple case studies due to space differences, (b) use all your
controls and all your patients, (c) use all your patients and an equal
number of controls (this way the template is balanced). I've always
opted for (c), but its still an open question. Jeff, do you have any
input on this question?
On Wed, Apr 22, 2009 at 10:35 AM, NAROA IBARRETXE BILBAO
<[log in to unmask]> wrote:
> Dear SPMrs,
>
> We did individual analyses preprocessed with DARTEL and runned it for
each
> of the patients, therefore we obtained as many templates as patients
we
> have, composed of the patient itself and all the controls. The steps
are the
> following ones (that were repeated for each of the patients)
>
> 1. Segment
> 2. Initial import
> 3. Run dartel (create template)
> 4. Create warped
> 5. Smooth
> 6. 1 patient versus controls
>
> We checked the results and realized that those patients with the
smallest
> intracraneal volume (generally women) were those that presented the
highest
> degree of atrophy (brains on flames :-) ).
>
> Is the procedure correct? or it would have been better just to run one
> dartel for all the sample? If the procedure is ok, should we just
introduce
> the intracraneal volume as a covariate? Is that possible? or it is not
> correct because one of the groups is just composed of one patient?
>
> Thank you for your help,
>
> Naroa.
|