Deletions are marked like this. | Additions are marked like this. |
Line 13: | Line 13: |
= Synopsis = mri_watershed [<options>] invol outvol |
|
Line 21: | Line 18: |
== Required Flagged Arguments == None |
|
Line 57: | Line 51: |
= Description = | |
Line 85: | Line 78: |
If you used -atlas option, then { The template was used to correct the surface. } |
|
Line 92: | Line 83: |
First, the code is constructing a discretized distribution of the whole intensity image (300 points) -> this is what PDcurve is doing. In doing so, I remove noisy voxels by removing the one thousandth brightest voxels. Then, in analyze_curve, I try to determine a meaningfull threshold (that would be SKULL_PD) of this distribution of voxels. To do so, the distribution should be peaking somewhere in the 300 points and then decreasing. In this decreasing region (that I find between 3max/4 and max/5), I do a linear least square approximation to fit the best line to this decreasing region: aX+b and then simply set SKULL_PD = -b/a, i.e. the region where the fit becomes 0. Needless to say that there is plenty of code regions that is not really robust and might fail. |
|
Line 93: | Line 89: |
= References = [[References/Segonne2004]] |
|
Line 97: | Line 90: |
Report bugs to < freesurfer@nmr.mgh.harvard.edu > = Author/s = YasunariTosa |
Report bugs to < freesurfer@nmr.mgh.harvard.edu > YasunariTosa |
Index
Contents
Name
mri_watershed - strip skull and other outer non-brain tissue
Arguments
Positional Arguments
invol |
input volume |
outvol |
output volume |
Optional Flagged Arguments
-atlas |
use the atlas information to correct the segmentation. |
When the segmented brain is not correct, this option might help you. |
-surf [surfname] |
save the BEM surfaces. |
In order to get the surfaces consistent with tkmedit, you have to use the option -useSRAS. |
-useSRAS |
use the surface RAS coordinates (not the scanner RAS) for surfaces. |
|
-noT1 |
don't do T1 analysis. (Useful when running out of memory) |
|
-less |
shrink the surface |
|
-more |
expand the surface |
|
-wat |
use only the watershed algorithm |
|
-T1 |
specify T1 input volume (T1 grey value = 110) |
|
-wat+temp |
watershed algo and first template smoothing |
|
-first_temp |
use only the first template smoothing + local matching |
|
-surf_debug |
visualize the surfaces onto the output volume |
|
-brainsurf [surfname] |
save the brain surface |
|
-shk_br_surf [int_h surfname] |
to save the brain surface shrank inward of int_h mm |
|
-s [int_i int_j int_k] |
add a seed point |
|
-c [int_i int_j int_k] |
specify the center of the brain (in voxel unit) |
|
-r int_r |
specify the radius of the brain (in voxel unit) |
|
-t int_threshold |
change the threshold in the watershed analyze process |
|
-h int_hpf |
precize the preflooding height (in percent) |
|
-n |
not use the watershed analyze process |
|
-LABEL |
labelize the output volume into scalp, skull, csf, gray and white |
|
-man [int_csf int_trn int_gray] |
to change the different parameters csf_max, transition_intensity and GM_intensity |
|
-mask |
mask a volume with the brain mask |
|
--help |
show usage message |
|
--version |
show the current version |
|
Outputs
brainvol |
skull stripped brain volume |
BEMsurfaces |
when you specify the option -brainsurf surfname |
Produce the brain volume from T1 volume or the scanned volume.
Examples
Example 1
mri-watershed -atlas T1 brain
where T1 is the T1 volume and brain is the output brain volume. When the cerebellum is cut-off from the brain or getting the left/right asymmetric brain, you should first try this -atlas option.
Example 2
mri-watershed T1 brain
The same as the first example, but no correction is applied to the intermediate result.
Bugs
None
See Also
Links
Methods Description
The "watershed" segmentation algorithm was used to dertermine the intensity values for white matter, grey matter, and CSF. A force field was then used to fit a spherical surface to the brain. The shape of the surface fit was then evaluated against a previously derived template. The finely grained sphere was fit to the brain. (Segonne 2004)
First, the code is constructing a discretized distribution of the whole intensity image (300 points) -> this is what PDcurve is doing. In doing so, I remove noisy voxels by removing the one thousandth brightest voxels. Then, in analyze_curve, I try to determine a meaningfull threshold (that would be SKULL_PD) of this distribution of voxels. To do so, the distribution should be peaking somewhere in the 300 points and then decreasing. In this decreasing region (that I find between 3max/4 and max/5), I do a linear least square approximation to fit the best line to this decreasing region: aX+b and then simply set SKULL_PD = -b/a, i.e. the region where the fit becomes 0. Needless to say that there is plenty of code regions that is not really robust and might fail.
Reporting Bugs
Report bugs to < freesurfer@nmr.mgh.harvard.edu > YasunariTosa