algorithms.segmentation.brain_segmentation
Module: algorithms.segmentation.brain_segmentation
Functions
-
nipy.algorithms.segmentation.brain_segmentation.brain_segmentation(img, mask_img=None, hard=False, niters=25, labels=('CSF', 'GM', 'WM'), mixmat=None, noise='gauss', beta=0.2, freeze_prop=True, scheme='mf', synchronous=False)
- Perform tissue classification of a brain MR image into gray
matter, white matter and CSF. The image needs be skull-stripped
beforehand for the method to work. Currently, it is implicitly
assumed that the input image is T1-weighted, but it will be easy
to relax this restriction in the future.
For details regarding the underlying method, see:
Roche et al, 2011. On the convergence of EM-like algorithms for
image segmentation using Markov random fields. Medical Image
Analysis (DOI: 10.1016/j.media.2011.05.002).
Parameters : | img : nipy-like image
- mask_img : nipy-like image
Brain mask. If None, the mask will be defined by thresholding
the input image above zero (strictly).
- beta: float
Markov random field damping parameter.
- noise: string
One of ‘gauss’: Gaussian noise assumption or ‘laplace’: Laplace
noise assumption.
- freeze_prop: boolean
If False, consider relative tissue volume proportions as free
parameters. Otherwise, use equal proportions.
- hard: boolean
If True, use FSL-FAST hard classification scheme rather than the
standard mean-field iteration (not advised).
synchronous: boolean :
Determines whether voxel are updated sequentially or all at
once.
- scheme: string
One of ‘mf’: mean-field or ‘bp’: (cheap) belief propagation.
- labels: sequence of strings
Label names.
|
Returns : | ppm_img: nipy-like image :
A 4D image representing the posterior probability map of each
tissue.
- label_img: nipy-like image
Hard tissue classification image similar to a MAP.
|
-
nipy.algorithms.segmentation.brain_segmentation.initialize_parameters(data, klasses)
Rough parameter initialization by moment matching with a brainweb
image for which accurate parameters are known.
Parameters : | data: array :
klasses: int :
Number of desired classes.
|
Returns : | means: array :
Initial class-specific intensity means
stdevs: array :
Initial class-specific intensity standard deviations
props: array :
Initial class-specific volume proportions
|