AuSoMS - Automatic Segmentation of the Mouse Skull in MR images

Prerequesities

You will need a folder <dicomFolder> containing a DICOM sequence. AuSoMS was developed for slices in the transversal plane.

Installation

Clone the project and open the project directory in MATLAB.

Launch

Add the whole project folder ./segmentation in the MATLAB path. The main code resides in segmentation/functions.
To call the default segmentation, execute

[skull_segmentation, brain_segmentation, ~] = segmentVolume(<path-to-dicomFolder>, struct());

AuSoMS will return a skull segmentation as well as a brain segmentation. They are represented by a 3D matrix each.

Options

In order to run AuSoMS with custom options, create an empty struct

options = struct();

and define attributes for it:

• options.resizeFactor Resizing factor of volume. Default: 1
• options.SEBg Structuring element used for background removal (Caution: Structuring element will be used as is, ignoring the resize factor)
• options.SEBgClose Structuring element used for closing of foreground (Caution: Structuring element will be used as is, ignoring the resize factor)
• options.sizeSEBrain Structuring element size for 3D PCNN brain segmentation in pixels (Caution: Size will be used as is, ignoring the resize factor)
• options.cutoff Number of slices ignored at the start as well as the end of the sequence. Default: 0
• options.nSegmentsForeground Number of segments for the foreground. Default: 4
• options.brainSegmentation Logical volume of existing brain segmentation (same size as input volume!). Alternative to using built in 3D-PCNN segmentation of the brain.
• options.maxSkullThickness Define thickness of skull (in mm). Default: 0.7
• options.evalFile .mat file which contains ground truth of skull skulltruth, scale factor factor in relation to scale of input slices and, optionally, ground truth of the brain braintruth
• options.fScoreBeta Indicates how 'many times' recall is more important than precision. Default: 2
• options.pcnnRepetitions Repetitions for averaging of brain segmentation. Default: 1
• options.outFile Specify file for saving the results. Default: No file specified

You don't have to define every attribute. Defining an attribute solely overwrites parameters in the main function later on. Then call

[skull_segmentation, brain_segmentation, ~] = segmentVolume(<path-to-dicomFolder>, options);

For example, if you want to work with the assumption that the maximum skull thickness is 0.6 mm instead of 0.7 mm (default), the code should look like this

options = struct();
options.maxSkullThickness = 0.6;
[skull_segmentation, brain_segmentation, ~] = segmentVolume(<path-to-dicomFolder>, options);

You can also save the results in a .mat file by defining

options.outFile = <path-to-saveFile>;

That's it! For a more detailed explanation on these parameters, read the thesis in the thesis project directory.

Evaluation

In case you want to evaluate the results, make sure you have a .mat file <evalFile> containing the variables skulltruth as well as the rescale factor factor (use value 1 if image was not rescaled for manual segmentation) and - optionally - braintruth. Now execute

options = struct(); options.evalFile = <path-to-evalFile>;
[skull, brain, eval] = segmentationVolume(<path-to-dicomFolder>, options);

The variable eval contains the Dice Score and the Matthews Correlation Coefficient for the skull segmentation (whole as well as upper half) and the brain segmentation if a ground truth was given. The scores will also be saved in <outFile> if you defined options.outFile before executing the main function.