Thank you for your quick reply!
I am using SPM5 on a Windows XP x64 installation of Matlab R2008b.
I kept the options at defaults except for 'Set origin'. I've attached a
copy of a template script used to generate these images.
They are m0 instead of m because the images are modulated only by the
non-linear compression map. To quote the tooltip: "Modulated images
can be optionally saved by correcting for non-linear warping only
... I recommend this option if your hypothesis is about effects of
relative volumes which are corrected for different brain sizes. This
is a widely used hypothesis and should fit to most data...These
modulated images are indicated by 'm0' instead of 'm'."
Regards,
Neil
John Ashburner wrote:
> My best guess is that the modulated images are stored as some form of integer
> datatype, with a scalefactor. To help me narrow down the cause of the
> problem, I need a few more details before I try to figure out the cause.
>
> Which version of SPM are you using?
> Which options did you use to generate the modulated images?
> Why are they called m0wc1*.nii instead of mwc1*.nii?
>
> Best regards,
> -John
>
> On Tuesday 05 May 2009 21:05, Neil Chatterjee wrote:
>
>> Dear SPMers,
>>
>> I came across several oddities in my m0wc1*.nii (modulated, normalized,
>> gray matter) images yesterday, and I am hoping someone here can shed some
>> light on the situation. Apologies for the length of this correspondence,
>> but I wanted to be precise in explaining the problem observed.
>>
>> Anyways, looking at a typical m0wc1*.nii image, the voxel values have the
>> following strange properties:
>>
>> 1) There are no voxels with a value greater than 1
>> 2) There are ~580,000 voxels with a value of exactly 1. Actually, they all
>> have a value of exactly 1.000000059138983, which in itself is kind of
>> strange. 3) Of the non-zero valued voxels in the m0wc1*.nii image, 42.4% of
>> them are exactly (to double precision) the same as in the wc1*.nii images.
>>
>> In a previous thread, Dr. Ashburner said that
>>
>>
>>> The contents of a modulated image are a voxel compression map multiplied
>>>
>> by tissue belonging
>>
>>
>>> probabilities (which range between zero and one)
>>> ...
>>> The total volume of grey matter in the original image can be
>>> determined by summing the voxels in the modulated, spatially
>>> normalised image and multiplying by the voxel volume (product of voxel
>>> size).
>>>
>> That the total volume of gray matter in the original image can be
>> determined by integration implies conservation of probability of gray
>> matter. It follows that the voxel compression map would have values >1 in
>> areas where there has been positive compression (shrinking) and values <1
>> in areas where there has been negative compression (expansion). With this
>> in mind, the properties described above lead me to the following
>> conclusions:
>>
>> A) There are no voxels with high probability (p~1.0) of being gray matter
>> that were positively compressed (shrunk) in normalizing, else there would
>> exist modulated voxels with value > 1.
>> B) There exist several voxels that either i) had a gray matter probability
>> of exactly 1 and were not compressed even one iota or ii) were compressed
>> in exact (to double precision!) proportion to their uncertainty of being
>> gray matter. Else there would not exist modulated voxels with value = 1
>> exactly C) 42.4% of probable gray matter voxels neither shrunk nor expanded
>> in the process of morphing to standard space.
>>
>> I just can't wrap my head around any of those conclusions. I feel like
>> either I'm totally misunderstanding what happens with modulation or
>> something is very very wrong with my images. I understand that the
>> non-linear only modulation (m0 instead of m) changes things, but
>> substituting "non-linear compression" for "compression" above does not make
>> the observations any less strange. If any guru out there can make sense of
>> all this, it would be much appreciated.
>>
>> Regards,
>> Neil
>>
>> Neil Chatterjee
>> Research Assistant
>> Stanford Systems Neuroscience and Pain Lab
>> 650-724-0522
>> [log in to unmask]
>>
>
>
|