Hi John,
Sorry to keep you busy, but I'm still struggling with this issue.
In the VBM/wiki (http://en.wikibooks.org/wiki/SPM/VBM), I found
several words about modulation: "
If you take the limiting case of extremely precise (not necessarily
extremely accurate) registration and segmentation, the pre-processed
concentration images are likely to be all identical. An analysis of
these data would not show you anything."
which is basically the same as what you suggested in this thread of
email. But I can't understand why the pre-processed concentration
images are likely to be all identical if an extremely precise
registration and segmentation was achieved. Should we still get a
subject specific probability map (c* map) for each tissue type, whose
value should not be changed by registration?
Thanks
Ze
On 10/28/2010 11:00 AM, Ze Wang wrote:
> Hi John,
> Except the registration errors, will the unmodulated analysis
> pick up the voxelwise tissue (or other property) difference? Like sub
> A, B have a grey matter probability of 0.8 and 0.7 in Hippocampus,
> respectively. Instead of looking at the volumetric difference, should
> we use the unmodulated analysis to see this voxel wise difference?
> Also, as people are also interested in the correlation between
> tissue intensity (actually the probability after segmentation) and
> some behavior scores. Should we also use the unmodulated normalized
> images?
>
> I appreciate your comments!
>
> Thank you!
> Ze
> On 10/28/2010 7:48 AM, John Ashburner wrote:
>> Unmodulated analysis shows regions where registration error differs
>> significantly among the populations, whereas modulated analysis
>> attempts to show regions of volumetric difference.
>>
>> Everyone accepts that using a different linear model to fit data will
>> give different results. In the same way, different pre-processing is
>> also expected to produce different findings (as this also changes how
>> differences among the data are modelled).
>>
>> Best regards,
>> -John
>>
>> On 28 October 2010 12:07, Cullen, Alexis<[log in to unmask]>
>> wrote:
>>> Hi SPM experts
>>>
>>> I have been running some analyses on a on a set of structural
>>> paediatric MRI
>>> images using the VBM 5 toolbox in SPM 5. I have carried out the
>>> segmentation
>>> and normalisation steps according to the VBM 5 manual with the
>>> exception
>>> that I have segmented against a custom template produced using the
>>> Template-O-Matic toolbox. A 2-sample t-test has been used to compare
>>> my two
>>> groups and these analyses have identified sig differences in areas that
>>> would be expected in the literature. However, I am concerned as the
>>> results
>>> produced from the modulated (non-linear) and unmodulated images differ
>>> substantially. I understand that the analysis of modulated and
>>> unmodulated
>>> images will answer different questions and that the two techniques
>>> will give
>>> varying results. However, I would have expected based on the
>>> literature and
>>> my understanding of the modulation process that analyses of
>>> unmodulated data
>>> would give differences generally in the same areas as the modulated
>>> but with
>>> more extensive group differences (e.g., Fortino et al, 2009;
>>> Schizophrenia
>>> Research). Or at least that group differences would be in the same
>>> direction
>>> using both image types i.e., areas of reduced volume in both
>>> analysis types.
>>>
>>> In contrast I have found group differences in opposite directions
>>> using the
>>> two images, namely that in modulated analyses my group of interest
>>> shows
>>> predominately increased GM and WM relative to the control group, but in
>>> unmodulated analyses the group of interest show predominately
>>> decreased GM
>>> and WM (albeit in different areas). I have one grey matter finding
>>> that is
>>> maintained in both analyses (interest group> control group), otherwise
>>> there seems to be no overlap between the results.
>>>
>>> I have repeated the analysis several times as the sample has
>>> expanded and
>>> the same finding has come up. I also checked my segmentation outputs
>>> and
>>> none of them look strange and there are no real inhomogeneity issues.
>>>
>>> I am just wondering whether anyone has any thoughts on whether this
>>> is some
>>> sort of error or whether the findings are possible? The areas of
>>> decreased
>>> GM and WM are more in line with what I would expect to see in an adult
>>> population but it is possible that in this child/adolescent
>>> population some
>>> specific areas might be increased in volume.
>>> I would be keen to hear if anyone has any thoughts/explanations for
>>> this.
>>> Many thanks, Alexis
>>>
>>> Alexis Cullen
>>> Research Worker/PhD Student
>>> R&R and CHADS projects
>>> Department of Forensic Mental Health Science (PO23)
>>> Institute of Psychiatry, De Crespigny Park
>>> Denmark Hill, SE5 8AF
>>> [log in to unmask]
>>> 020 7848 5678
>>>
>>>
>
>
--
--------------------------------------------------
Ze Wang, Ph. D
Research Assistant Professor of Biomedical Engineering,
Department of Psychiatry,
School of Medicine,
University of Pennsylvania,
3900 Chestnut Street,
Philadelphia, PA 19104, USA
Tel: 215-222-3200 ext 123
Fax: 215-386-6770
|