Automated Fetal Brain Segmentation from 2D MRI Slices for Motion Correction

This GitHub repository contains a complete example to segment the fetal brain in MRI and perform motion correction in a fully automatic way using the code from [1] and [3], which is publicly available at:

https://github.com/BioMedIA/IRTK

A step-by-step installation guide with a frozen version of IRTK is available at:

http://kevin-keraudren.blogspot.co.uk/2014/10/automated-segmentation-of-fetal-brain.html

Prerequisites

In order to install a working Python environment with all the necessary packages (numpy, scipy,cython, scikit-learn and OpenCV), it is simpler to use a prepackaged installation such as the one provided by Continuum Analytics in Anaconda.

http://continuum.io/downloads

Then you need to build IRTK, enabling the Python wrappers:

git clone https://github.com/BioMedIA/IRTK.git
cd IRTK
mkdir build
cd build
cmake -D BUILD_WITH_PNG=ON -D WRAP_CYTHON=ON ..
make -j 3

Lastly, you now need to update your PYTHONPATH:
get the full path of "irtk/build/lib" and add to your ~/.bashrc :

export PYTHONPATH=full_path/irtk/build/lib:$PYTHONPATH

Note:
This code was last tested with Anaconda 2.1.0, which comes with the following versions of the different Python modules: https://docs.continuum.io/anaconda/old-pkg-lists/2.1.0/py27

In particular, scikit-learn 0.15.2 was used when training and saving the provided models. Using a different version of scikit-learn may lead to the following error message: AttributeError: 'LinearSVC' object has no attribute 'classes_'.

Training a model

Trained models are already provided in the "model/" folder. If you wish to train models on your own data, you need to install SimpleITK, which can be done with the following command:

easy_install SimpleITK

1. Learn a vocabulary of 2D SIFT features (extracted using OpenCV) with MiniBatchKmean from scikit-learn and taking the cluster centers:
   create_bow.py

2. Train an SVM classifier on histogram of SIFT features extractede within MSER regions (which are first filtered by size using the gestational age):
   learn_mser.py

Running the example

In order to run the example, you first need to edit the script "runs.sh" so that the variables SCRIPT_DIR and BIN_DIR point respectively to The example can be run with the following command:

./run.sh -a

It performs the following 3 steps on the dataset provided in the folder "data":

1. Brain detection

2. Brain segmentation

3. Motion correction

Usage: ./run.sh -a

Valid options are: -1, -2, -3, -a, -d
    -1    to run only the detection step
    -2    to run only the segmentation step
    -3    to only the motion correction step
    -a    to run all steps
    -d    for debugging

Dataset

The folder "data/" contains 8 stacks of a subject at 29.7 gestational weeks.

2 coronal stacks (1-2):

2 sagittal stacks (3-4):

and 4 transverse stacks (5-8):

Output

Segmented brains:

Motion corrected volume:

References

[1] Keraudren, K., Kuklisova-Murgasova, M., Kyriakopoulou, V., Malamateniou, C., Rutherford, M.A., Kainz, B., Hajnal, J.V., Rueckert, D., 2014. Automated Fetal Brain Segmentation from 2D MRI Slices for Motion Correction. NeuroImage.
http://www.sciencedirect.com/science/article/pii/S1053811914005953

[2] Keraudren, K., Kyriakopoulou, V., Rutherford, M., Hajnal, J., Rueckert, D., 2013. Localisation of the Brain in Fetal MRI Using Bundled SIFT Features, in: MICCAI, Springer. http://www.doc.ic.ac.uk/~kpk09/publications/MICCAI-2013.pdf

[3] Kuklisova-Murgasova, M., Quaghebeur, G., Rutherford, M.A., Hajnal, J.V., Schnabel, J.A., 2012. Reconstruction of Fetal Brain MRI with Intensity Matching and Complete Outlier Removal. Medical Image Analysis.
http://www.medicalimageanalysisjournal.com/article/S1361-8415(12)00096-5/fulltext