Fast Marching Segmentation Module

1. Fast Marching segmentation module

This is an active contour 3D segmentation module.
Developed at the Georgia Institute of Technology, Biomedical Imaging Lab, for Slicer2 by Eric Pichon.
MICCAI 2003 paper Coming soon...

1a. Installation

First make sure that the module is not already installed.

If the module is installed, it should appear as an Editor effect.

If you cannot see it, you probably need to compile it first.

the module is installed...

To compile the module under Linux / Solaris, do something like:

[gte444x@sarah slicer2]$ cd Modules/vtkFastMarching/
[gte444x@sarah vtkFastMarching]$ mkdir builds
[gte444x@sarah vtkFastMarching]$ cd builds/
[gte444x@sarah builds]$ mkdir linux
[gte444x@sarah builds]$ cd linux
[gte444x@sarah linux]$ ccmake ../..

(note: use the same name for the build directory and the Slicer Base build directory. For example, if you build Slicer in <slicer>/Base/builds/linux then they build the module in <slicer>/Modules/vtkFastMarching/builds/linux)

Answer CCMAKE's questions, at some point you should be able to "Press [g] to generate and exit".
Then, just do:

[gte444x@sarah linux]$ make

The compilation should succeed without any warnings. When starting Slicer again you should now see the module.

To compile the module under Windows, the instructions are the same except that you should run CMakeSetup instead of ccmake. Inside there you can pick the source and build directories and the compilation environment you will be using (nmake, Visual Studio 6, etc).

1b. Segmenting a volume

In this step-by-step example we will segment a cortex from an MRI data set. This is the same data.xml file found in the tutorial.tar.gz provided with the Slicer.

Open the data (Data / Add Volume / ...)
Enter the Editor (Editor / Start Editing)
Enter the Fast Marching Module. If you cannot find it, go back to the previous section.

At this point the Editor panel should look like that:

the Fast Marching panel

Now we have to define some "seed points" inside the object you want to segment.

This is done by creating Fiducials. Fiducials are created by moving the cursor to the desired location and pressing the 'p' key. To delete Fiducials, move the cursor over the desired location and press the 'd' key.

Fore more information about Fiducials, see the Fiducials section of the User Guide.

Why seeds are important....

The seeds define an intensity and homogeneity range. This is then used to compute a speed function to move a surface. As more points come into the surface they are used to get a better estimate of the statistics. As the volume of the surface increases very rapidly, the initial contribution of the seeds is soon negligible, this means that
there is no really good way to correct a "bad start".

Rule of thumb: put as many seeds as possible, make sure that they are representative of the object you want to segment and that there is at least one in every hard-to-reach region.

Here we have defined seed points in all lobes of the two hemispheres.

Note that in this case it is a good idea to put seeds inside the white and gray matter. Or else the statistics of the sole white
matter for instance will be taken into account. [the algorithm
is robust enough to give OK results even in that case, this is mostly an example].

setting the seeds

We are now ready to start segmenting !

Select a color for the label map by clicking on the "label" button.
Click on the "EXPAND" button on the left panel. The volume which was initially composed by the seed points expands and will slowly fill in the object to be segmented.

If the segmentation is not over, click on "EXPAND" . The entry field right of the "EXPAND" button determines how much more volume (measured in milliliters) is to be added to the segmented volume for each expansion. The total volume which can be placed within an average brain is 1200 mL.

If the segmentation went too far and "leaks out" of the object you were trying to segment, you can use the slider to go back in time to the point right before leaking occurred. If you are pleased with the segmentation, jump to the next section. If some regions of your object have not been reached by the volume yet, try adding some new seed points there and start another expansion. If this process does not converge to an acceptable solution, you should probably start again with more seed points...

Moving the slider completely to the left will only go back right before the last time you hit EXPAND. Therefore you should make sure that the segmentation has not leaked out of the object before you start another expansion.

Also, think about increasing the expansion amount when segmenting a large object (as in this example). You will obtain exactly the same segmentation with a lot less clicks...

Screenshot of the segmentation of the cortex

You can then use other Editing tools on the labelmap. For instance you can perform a morphological dilation to regularize the segmented volume.

You can now create a model to visualize the segmented volume. When creating the model, make sure that you selected the correct volume (here: Working) and the correct labelmap (here: 14).