hide details 08: 02 (5 hours ago)
From: John Ashburner <[log in to unmask]>
Reply-To: [log in to unmask]
To: [log in to unmask]
Cc: [log in to unmask]
Content-Type: text/plain; charset="UTF-8"
Date: Wed, 04 Aug 2010 14:11:44 +0100
Message-ID: <1280927504.7972.104313.camel@zappa>
Mime-Version: 1.0
X-Mailer: Evolution 2.28.3
Content-Transfer-Encoding: 7bit

        When I used the DARTEL tool to get the Gray matter and White
        matter, I have got some questions.

        1. Using modulation once is what the manual tells us. Then when
        we use twice modulation, what result will we get?

I'm not really sure what you mean by this.  Are you referring to
spatially normalising grey matter images that have already been
spatially normalised and Jacobian scaled (instead of using the
"imported" data)?  If so, then the DARTEL algorithm would not deal with
such data very well.  The reason for this is that registration is based
on aligning tissue class images.  If you supply (say) grey matter and
white matter, then the algorithm will make another tissue class that
represents not grey or white, and base the registration on
simultaneously aligning the three tissue types.  It computes the
additional class by subtracting the grey and white from 1.  If these
data were Jacobian scaled, then this background class would contain some
negative values (where Jacobian scaled grey+white were > 1).  Such
negative values confuse the Dartel registration algorithm.

        2. To the result map, such as swr*.img, what does the voxel
        intensity mean? Dose it mean the relative volume per voxel?

It is an image that contains the average of the signal that is under the
Gaussian smoothing kernel at each point.  This image is not really
useful for VBM.

        3. Can the DARTEL compute the density? If it can, how dose it
        do?

It does not compute the density of neuronal cell bodies or anytging like
that.  Segmentation simply partitions the image into different tissue
classes, which are used by Dartel.  When the tissue classes are warped,
the volume of tissue in each brain region is preserved - either by
warping and scaling by the Jacobian determinants (which encode relative
volumes before and after warping), or by actually adding all the
original voxels of the tissue class image into the appropriate places in
the spatially normalised image.

        4. When got the DARTEL result imgfiles, how to compute the
        volume and the relative volume of ROIs?

Each native space tissue class image (c1*.nii or c1*.img) contains
voxels that encode the probability of each voxel containing grey matter
(according to the segmentation model).  Summing up these values will
give the probable number of grey matter voxels.  Multiplying this number
by the volume of a voxel will give the probable volume of grey matter.

For "modulated" and spatially normalised grey matter (mwc1*.nii), a
similar procedure may be applied.  Sum up the values and multiply by the
volume of a voxel.  This works because the total amount of signal is
preserved by the "modulation".

        5. If we use DARTEL to deal with different types of data, e.g.
        some are axial dicom files and some are sigttal ones, can the
        result be trustworthy? For another condition, some are with
        voxel of 0.5*0.5*1(mm3), some are with voxel of 1*1*1, how
        about of the result of this condition?

You can not easily interpret the results of a comparison between one
population of brains scanned one way, with another population of brains
scanned another way.  Any differences you find between the data may be
because of differences between the populations of subjects or
differences between the scanning settings.

If you are not comparing two populations this way, then you could
include additional columns in your design matrix that account for the
scanner settings.  This would allow your analysis to show differences
that could not be explained by the acquisition settings.

Best regards,
-John


--
John Ashburner <[log in to unmask]>