neoSeg
Table of Contents
Introduction
This repository contains the work during the PhD of Carlos Tor-Díez, titled "Automatic segmentation of the cortical surface in neonatal brain MRI". It includes two main contributions:
- patchBasedSegmentation.py, where several methods of label fusion (the final step of the multi-atlas segmentation approaches) are performed, including IMAPA [1].
- topologicalCorrection.py, where a topological correction for segmentation is presented, which consists in a multi-scale, multi-label topological-preserving deformation [2].
Requirements
All scripts were coded in python 2.7, but we are working to be compatible to python 3.7.
Python packages
- argparse
- nibabel
- numpy
- scipy
- time
- itertools
- multiprocessing
- numba
- math
- random
- matplotlib
- scikit-image (skimage)
- scikit-fmm (skfmm)
The file called requirements.txt helps to install all the python libraries.
- Using pip:
pip install -r requirements.txt
- Using anaconda:
conda install --file requirements.txt
Run
patchBasedSegmentation.py
Example of IMAPA application using an atlas set of two pairs of images using two iterations (alpha = 0 and alpha = 0.25) using 4 threads in parallel:
python neoSeg/patchBasedSegmentation.py -i brain.nii.gz -a atlas1_registered_HM.nii.gz atlas2_registered_HM.nii.gz -l label1_propagated.nii.gz label2_propagated.nii.gz -mask mask.nii.gz -m IMAPA -hss 3 -hps 1 -k 15 -alphas 0 0.25 -t 4
i: input anatomical image
a: anatomical atlas images in the input space
l: label atlas images in the input space
mask: binary image for input
m: segmentation method chosen (LP, S_opt, I_opt, IS_opt or IMAPA)
hss: half search window size
hps: half patch size
k: k-Nearest Neighbors (kNN)
alphas: alphas parameter for IS_opt and IMAPA methods
t: Number of threads (0 for the maximum number of cores available)
Note: We recommend to previously register the intensity image from the atlas set to the input image, apply a histogram matching algorithm and propagate the transformations to the label maps.
topologicalCorrection.py
Example of topological correction using two segmentation maps (White Matter and cortical Gray Matter) as segmentation of reference and an atlas prior of the brainstem in order to apply a topological relaxation. Most relevant defaults parameters are to add the background as an extra label, start the correction with omega = 2, apply the relaxation step in omega = 1, stop the correction in omega = 0 and consider a connectivity 6-26 (6:WM and GM, 26:background).
python neoSeg/topologicalCorrection.py -sref neoSeg/examples/31_wm.nii.gz neoSeg/examples/31_ribbon_cortex.nii.gz -rp neoSeg/31_brainstem_drawem.nii.gz -opath neoSeg/results -spath neoSeg/topology/intermediate
sref: input segmentation maps to be corrected (multiple entries are accepted)
rp: binary image providing the spatial prior where the topological relaxation will be applied
opath: Output path
spath: Step path
Note: A set of files for a demonstration is available in the repository 'github.com/rousseau/neoSeg/examples'. In order to check obtained results, they are available in 'github.com/rousseau/neoSeg/results'.
Publications
-
[1] C. Tor-Díez, N. Passat, I. Bloch, S. Faisan, N. Bednarek and F. Rousseau, “An iterative multi-atlas patch-based approach for cortex segmentation from neonatal MRI,” Computerized Medical Imaging and Graphics, 70:73–82, 2018, hal-01761063.
-
[2] C. Tor-Díez, S. Faisan, L. Mazo, N. Bednarek, Hélène Meunier, I. Bloch, N. Passat and F. Rousseau, “Multilabel, multiscale topological transformation for cerebral MRI segmentation post-processing,” In 14th International Symposium on Mathematical Morphology (ISMM 2019), pp. 471–482, 2019, hal-01982972.