> Many thanks for Matthieu's reply.
> I still don't understand what‘voxvol = det(V(1).mat)/100^3;’means. Why the
> volume is equal to ' tot*voxvol ', not 'tot/100^3'? What does V.mat
> represent?
> Any suggestion will be appreciated.
V.mat represents the voxel-to-world mapping of the image. It basically maps
from voxels (1..Nx, 1..Ny, 1..Nz) to some coordinate system in mm. A simple
voxel to world mapping (no rotations) would have the voxel sizes along the
diagonal, and the determinant of this matrix would be the product of these
lengths - i.e. the volume of a voxel. More generally, det(V.mat), or
det(V.mat(1:3,1:3)) is the volume of a voxel. With flipping, the volume
would be negative, so it is better to use abs(det(V.mat(1:3,1:3))).
>
> I've read the script of spm_global. It only says to calculate the mean
> value but not the volume. How does it calculate the volume then?
spm_global is extremely crude (almost embarrasingly so). It computes the mean
intensity within the image, and assumes that any voxel with an intensity of
over 1/8 of this mean belongs to brain. The "global" is then the average
intensity of these brain voxels.
>
> Also the question in my previous email. If I’d like to compute the global
> intracranial volume, shall I do all the same but only replace the modulated
> gray matter image to the modulated brain image?
I would definately not suggest using the modulated brain image. Basically,
you want a count of the voxels, not a sum of all the voxel values. If an
image consists of ones and zeros, and you sum the values, then you get a
count of the number of ones. If you segment an image in native space, then
the voxel values represent a probability of each voxel belonging to a
particular tissue class. These values go from zero (totally improbable) to
one (totally probable). If you sum these values, then you get a measure of
the probable number of voxels in the class.
If an image has been spatially normalised, then some voxels will shrink,
whereas others will grow. "Modulation" simply multiplies the values by the
ratio of original and warped volumes, such that if you sum up the voxels of a
modulated spatially normalised image, you should get approximately the same
answer as if you sum up the values in the original image (after accounting
for voxel volumes). For example, if a region shrinks by a factor of two,
then the intensities would be doubled in order to preserve the original
signal.
Best regards,
-John
|