Brain Surface Extraction for
Visualization
This uses the Simple Brain Extract
component of the segmentation
tool:
This uses a series of mathematical
morphology operations to extract and label a connected 3D volume from
the image data using the method of [Hohne and Hansen JCAT, Vol 16(2), 1991, 285-194]. This is most useful for doing a quick brain
extraction for brain surface visualisation. It is a semi
interactive process allowing the quick refinement of values to extract
a surface of interest from the data. Note: This is used for approximate surface extraction ONLY...
intended for quick data visualization rather than high accuracy
cortical surface delineation!
The steps to extract a working
brain
surface for rendering are as follows:
Step 1: Get a
starting range of Brain Tissue Intensities
With an MRI loaded and the segmentation tool selected:
a) Make sure that the Active Intensity Range
in the SegTool is reset to maximum (use the reset button)
b) Manually Delineate a sample of intensities in the brain using the
Draw and Fill Mode:
Make sure that you include gray and white matter BUT EXCLUDE CSF!
c) Select 'From Labels' in the
Active
Intensity Range window to capture the range of intensities in
the
selected region.
2. Apply
Intensity Range to Extract Brain
In the Simple
Brain Extract control, select the 'Apply to Label Region'.
On my example data set this created the following segmentation:
This is an 'over-segmentation'
of the brain since it has leaked out and included (in the green label)
regions of
scalp and other non brain head tissues.
3. Refine
Intensity Range to Remove Leaks
The minimum intensity value should be increased and the button 'Apply
to Label Region' re-selected to try another data range, In this case
the
range 72 to 120:
This is better but still includes some scalp (bright regions around the
brain). To remove these try using either a higher
minimum value or a lower maximum value (to exclude the bright fat
tissues in the scalp that are brighter than any tissues in the brain)
In this example case I increased the minimum value again to 85:
This is region is an 'under
segmentation' of the brain, since regions of the cortex,
particularly at the top of the brain and in the cerebellum, are missing
from the green label region. In this example I then tried an
intermediate value (between the previous 85 and 72) for the lower value
of the intensity range: 75 which gave the following acceptable brain
mask:
This looks like a pretty reasonable brain volume. (perhaps some regions
missing from the top of the cortex ?)
Note:
If you cannot get a usable brain label you can simply manually edit out
any unwanted regions or fill in any holes in the volume and then use
option a) in the next section to create a surface file from the current
label dataset.
Step 4:
Create a 'Surface' Volume
This can be done in two ways:
a) Using the segmentation currently
created (in this case the green brain mask) using:
b) Directly by using the simple brain extraction step again but to
create a surface volume dataset
directly rather than a label volume for the segmentation display using:
You can use (b) if the simple brain extraction step provides a usable
brain segmentation without any manual editing, otherwise if you need to
edit the regions manually to get an accurately labeled brain then you
should use 1)
Here we will use the 2nd option and select 'create surface and view'
button:
This uses the same morphology operations
to create a binary brain region label without modifying the
edited labeled regions displayed in rview. It then smooths
this to create a
clean surface volume. The display mode is then changed and updated to
show:
Here the segmentation is shown along with the brain surface rendered:
(Note it is coloured green here because
the rendering is showing the original segmentation labeling we created
below the brain surface)
This is a slightly over segmented brain surface but gives good
delineation of the sucal structures.
Step 5: Saving the Surface Volume Data
If you want to return to a surface volume view later by saving the
rview setup file: you should save the surface volume first and then
save the setup file (to ensure that the setup file contains a reference
to your saved surface volume).
Notes:
The simple morphology based brain extraction is highly dependent on
global
intensity values:
As such, if your data has a significant Bias field artifact, it is
advisable to apply an inhomogeneity
correction algorithm to the data first.
The surface dataset created is simply another image volume stored
within rview. It can be saved using
the 'Save Surface' button in the ISO surface tool. Once this surface
dataset exists it can be used to created
other rendered views by selecting and using the Render Tool. (See Options/Tools Menu)
An example is a larger format displays like this:
Example ISO Surface Rendering of a Brain Surface Extracted Using
the Morphology Segmentation Tool
