Rich, John,
I would not venture to suggest that I am "a bit more knowledgeable" than
you on this but would like to agree that using different strategies for
subjects within one analyses is problematic.
The way I see it, you have at least two other option: one, devise a
lesion/cost function masking approach that is applied to all subjects
(for example, use CSF as a mask to avoid CSF spaces and chronic lesions
to influence processing), or use an externally generated DARTEL
template. For example, there is one coming with vbm8 which you could
use, potentially offering the advantage of being (almost) equally far
away from both populations. Which way is "better" will still depend on
your data, lesion size, characteristics etc.
I would generally be careful in using extensive normalization approaches
in subjects with lesions as you want to compare the two groups but at
the same time preserve what essentially characterizes your patients (the
lesion). It may be less fancy ;) but if you follow the approach
suggested by Crinion et al in NeuroImage 2007, does that work?
Cheers,
Marko
John Ashburner wrote:
> Personally, I prefer to see all data treated the same way, so that
> pre-processing is not informed by the linear combinations of design
> matrix columns you intend to use to examine the differences. If you
> leave out a bunch of subjects, but only from one group, then you are
> effectively doing the pre-processing slightly differently for each of
> the two groups. This may introduce a bias.
>
> I am no statistician, so I would welcome the views of others on the list
> who may be a bit more knowledgeable than myself on this.
>
> Best regards,
> -John
>
> On 14 March 2013 11:08, Richard Bethlehem <[log in to unmask]
> <mailto:[log in to unmask]>> wrote:
>
> Dear all,
>
> I'm currently working on a patient-control study with relatively
> small sample of patients (5 patients versus 13 controls), these
> patients all share a similar small lesion and I would like to
> optimize my pre-processing pipeline for the functional scans I've
> acquired. Normally I would use a 'standard' DARTEL procedure that
> will allow me to get some optimal normalisation as follows:
>
> 1. reallign my functional scans
> 2. coregister my functional scans to my anatomical scans
> 3. segment the anatomical scans (using New Segment)
> 4. create DARTEL template and flow-fields from the anatomical scans
> 5. use the flow-fields from step 4 to normalise (DARTEL utility) my
> functional scans to MNI.
> 6. Smooth the functional scans
>
> Now given the balance of my groups (more than twice as many
> controls) I'm a bit concerned that the template created in step 4
> might be a bit biased towards the control group (and hence the
> extend of the lesion in the patient group might be diminished).
> Therefore I was wondering whether the following (somewhat circular
> approach) could be used to prevent that issue:
>
> 1. reallign my functional scans
> 2. coregister my functional scans to my anatomical scans
> 2.1 use a balanced subset of the anatomical scans (5 patients 5
> controls) to segment and subsequently create a DARTEL template
> 2.2 normalise the DARTEL (study-specific) template to MNI (for all
> tissue classes)
> 3. segment the anatomical scans, but instead of using the standard
> TPM's from the segment toolbox, use the DARTEL template created in 2.2
> 4-6 stays the same
>
> The other issue pertains to the smoothing, I've heard there might be
> an issue with using small smoothing kernels in the DARTEL procedure
> when the sample size is small. First of all, perhaps someone is keen
> to elaborate on that issue as it is not immediately clear to me why
> that might be the case. Second, given that the patients all share a
> relatively small lesion I would like to use a medium-sized smoothing
> kernel of 6mm (the original anatomical are acquired at 3mm
> isotropic). Would that be reasonable?
>
> Best,
>
> Richard
>
>
--
____________________________________________________
PD Dr. med. Marko Wilke
Facharzt für Kinder- und Jugendmedizin
Leiter, Experimentelle Pädiatrische Neurobildgebung
Universitäts-Kinderklinik
Abt. III (Neuropädiatrie)
Marko Wilke, MD, PhD
Pediatrician
Head, Experimental Pediatric Neuroimaging
University Children's Hospital
Dept. III (Pediatric Neurology)
Hoppe-Seyler-Str. 1
D - 72076 Tübingen, Germany
Tel. +49 7071 29-83416
Fax +49 7071 29-5473
[log in to unmask]
http://www.medizin.uni-tuebingen.de/kinder/epn/
____________________________________________________
|