segment/normalize (using SPM segmentation - no links between parameters)

Segment Segment, bias correct and spatially normalise - all in the same model. This function can be used for bias correcting, spatially normalising or segmenting your data. Note that this module needs the images to be roughly aligned with the tissue probability maps before you begin. If strange results are obtained, then this is usually because the images were poorly aligned beforehand. The Display option can be used to manually reposition the images so that the AC is close to coordinate 0,0,0 (within a couple of cm) and the orientation is within a few degrees of the tissue probability map data.

Description

Many investigators use tools within older versions of SPM for a technique that has become known as "optimised" voxel-based morphometry (VBM). VBM performs region-wise volumetric comparisons among populations of subjects. It requires the images to be spatially normalised, segmented into different tissue classes, and smoothed, prior to performing statistical tests. The "optimised" pre-processing strategy involved spatially normalising subjects brain images to a standard space, by matching grey matter in these images, to a grey matter reference. The historical motivation behind this approach was to reduce the confounding effects of non-brain (e.g. scalp) structural variability on the registration. Tissue classification in older versions of SPM required the images to be registered with tissue probability maps. After registration, these maps represented the prior probability of different tissue classes being found at each location in an image. Bayes rule can then be used to combine these priors with tissue type probabilities derived from voxel intensities, to provide the posterior probability. This procedure was inherently circular, because the registration required an initial tissue classification, and the tissue classification requires an initial registration. This circularity is resolved here by combining both components into a single generative model. This model also includes parameters that account for image intensity non-uniformity. Estimating the model parameters (for a maximum a posteriori solution) involves alternating among classification, bias correction and registration steps. This approach provides better results than simple serial applications of each component. Note that multi-spectral segmentation (e.g. from a registered T1 and T2 image) is not yet implemented, but is planned for a future SPM version. This branch contains 3 items: - Data - Output Files - Custom

Parameters

MRI_Nat: 4D Volume ( input )

MRI_Mni_tpmSeg: 4D Volume ( input )

spmJobName: String ( input )

GM: Choice ( input )

grey_Nat: 4D Volume ( optional, output )

grey_Mni: 4D Volume ( output )

WM: Choice ( input )

white_Nat: 4D Volume ( output )

CSF: Choice ( input )

csf_Nat: 4D Volume ( output )

biasreg: Choice ( input )

Bias regularisation MR images are usually corrupted by a smooth, spatially varying artifact that modulates the intensity of the image (bias). These artifacts, although not usually a problem for visual inspection, can impede automated processing of the images. An important issue relates to the distinction between intensity variations that arise because of bias artifact due to the physics of MR scanning, and those that arise due to different tissue properties. The objective is to model the latter by different tissue classes, while modelling the former with a bias field. We know a priori that intensity variations due to MR physics tend to be spatially smooth, whereas those due to different tissue types tend to contain more high frequency information. A more accurate estimate of a bias field can be obtained by including prior knowledge about the distribution of the fields likely to be encountered by the correction algorithm. For example, if it is known that there is little or no intensity non-uniformity, then it would be wise to penalise large values for the intensity non-uniformity parameters. This regularisation can be placed within a Bayesian context, whereby the penalty incurred is the negative logarithm of a prior probability for any particular pattern of non-uniformity.

biascor: Choice ( input )

Bias Corrected This is the option to produce a bias corrected version of your image. MR images are usually corrupted by a smooth, spatially varying artifact that modulates the intensity of the image (bias). These artifacts, although not usually a problem for visual inspection, can impede automated processing of the images. The bias corrected version should have more uniform intensities within the different types of tissues.

biasCorrected: 4D Volume ( optional, output )

Bias Corrected This is the option to produce a bias corrected version of your image. MR images are usually corrupted by a smooth, spatially varying artifact that modulates the intensity of the image (bias). These artifacts, although not usually a problem for visual inspection, can impede automated processing of the images. The bias corrected version should have more uniform intensities within the different types of tissues.

cleanup: Choice ( input )

Clean up any partitions This uses a crude routine for extracting the brain from segmentedimages. It begins by taking the white matter, and eroding it acouple of times to get rid of any odd voxels. The algorithmcontinues on to do conditional dilations for several iterations,where the condition is based upon gray or white matter being present.This identified region is then used to clean up the grey and whitematter partitions, and has a slight in uences on the CSF partition. If you found pieces of brain being chopped out in your data, then you may wish to disable or tone down the cleanup procedure.

ngaus: String ( input )

Gaussians per class The number of Gaussians used to represent the intensity distribution for each tissue class can be greater than one. In other words, a tissue probability map may be shared by several clusters. The assumption of a single Gaussian distribution for each class does not hold for a number of reasons. In particular, a voxel may not be purely of one tissue type, and instead contain signal from a number of different tissues (partial volume effects). Some partial volume voxels could fall at the interface between different classes, or they may fall in the middle of structures such as the thalamus, which may be considered as being either grey or white matter. Various other image segmentation approaches use additional clusters to model such partial volume effects. These generally assume that a pure tissue class has a Gaussian intensity distribution, whereas intensity distributions for partial volume voxels are broader, falling between the intensities of the pure classes. Unlike these partial volume segmentation approaches, the model adopted here simply assumes that the intensity distribution of each class may not be Gaussian, and assigns belonging probabilities according to these non-Gaussian distributions. Typical numbers of Gaussians could be two for grey matter, two for white matter, two for CSF, and four for everything else.

regtype: Choice ( input )

Affine Regularisation The procedure is a local optimisation, so it needs reasonable initial starting estimates. Images should be placed in approximate alignment using the Display function of SPM before beginning. A Mutual Information affine registration with the tissue probability maps (D'Agostino et al, 2004) is used to achieve approximate alignment. Note that this step does not include any model for intensity non-uniformity. This means that if the procedure is to be initialised with the affine registration, then the data should not be too corrupted with this artifact.If there is a lot of intensity non-uniformity, then manually position your image in order to achieve closer starting estimates, and turn off the affine registration. Affine registration into a standard space can be made more robust by regularisation (penalising excessive stretching or shrinking). The best solutions can be obtained by knowing the approximate amount of stretching that is needed (e.g. ICBM templates are slightly bigger than typical brains, so greater zooms are likely to be needed). For example, if registering to an image in ICBM/MNI space, then choose this option. If registering to a template that is close in size, then select the appropriate option for this.

warpreg: String ( input )

Warping Regularisation The objective function for registering the tissue probability maps to the image to process, in-volves minimising the sum of two terms. One term gives a function of how probable the data is given the warping parameters. The other is a function of how probable the parameters are, and provides a penalty for unlikely deformations. Smoother deformations are deemed to be more probable. The amount of regularisation determines the tradeof between the terms. Pick a value around one. However, if your normalised images appear distorted, then it may be an idea to increase the amount of regularisation (by an order of magnitude). More regularisation gives smoother deformations, where the smoothness measure is determined by the bending energy of the deformations.

warpco: String ( input )

Warp Frequency Cutoff Cutoff of DCT bases. Only DCT bases of periods longer than the cutoff are used to describe the warps. The number actually used will depend on the cutoff and the field of view of your image. A smaller cutoff frequency will allow more detailed deformations to be modelled, but unfortunately comes at a cost of greatly increasing the amount of memory needed, and the time taken.

biasfwhm: Choice ( input )

Bias FWHM FWHM of Gaussian smoothness of bias. If your intensity non-uniformity is very smooth, then choose a large FWHM. This will prevent the algorithm from trying to model out intensity variation due to different tissue types. The model for intensity non-uniformity is one of i.i.d. Gaussian

samp: String ( input )

Sampling distance The approximate distance between sampled points when estimating the model parameters. Smaller values use more of the data, but the procedure is slower.

msk: String ( input )

The segmentation can be masked by an image that conforms to the same space as the images to be segmented. If an image is selected, then it must match the image(s) voxel-for voxel, and have the same voxel-to-world mapping. Regions containing a value of zero in this image do not contribute when estimating the various parameters.

snMat: Any Type ( output )

snInvMat: Any Type ( output )

Technical information

Toolbox : Tools

User level : 2

Identifier : segment_SPM_noLinks

File name : brainvisa/toolboxes/tools/processes/spm/segment_SPM_noLinks.py

Supported file formats :

MRI_Nat :

GIS image, VIDA image, NIFTI-1 image, MINC image, gz compressed MINC image, DICOM image, TIFF image, XBM image, PBM image, PGM image, BMP image, XPM image, PPM image, gz compressed NIFTI-1 image, TIFF(.tif) image, ECAT i image, PNG image, JPEG image, MNG image, GIF image, SPM image, ECAT v image

MRI_Mni_tpmSeg :

GIS image, VIDA image, NIFTI-1 image, MINC image, gz compressed MINC image, DICOM image, TIFF image, XBM image, PBM image, PGM image, BMP image, XPM image, PPM image, gz compressed NIFTI-1 image, TIFF(.tif) image, ECAT i image, PNG image, JPEG image, MNG image, GIF image, SPM image, ECAT v image

grey_Nat :

GIS image, VIDA image, NIFTI-1 image, MINC image, gz compressed MINC image, DICOM image, TIFF image, XBM image, PBM image, PGM image, BMP image, XPM image, PPM image, gz compressed NIFTI-1 image, TIFF(.tif) image, ECAT i image, PNG image, JPEG image, MNG image, GIF image, SPM image, ECAT v image

grey_Mni :

GIS image, VIDA image, NIFTI-1 image, MINC image, gz compressed MINC image, DICOM image, TIFF image, XBM image, PBM image, PGM image, BMP image, XPM image, PPM image, gz compressed NIFTI-1 image, TIFF(.tif) image, ECAT i image, PNG image, JPEG image, MNG image, GIF image, SPM image, ECAT v image

white_Nat :

GIS image, VIDA image, NIFTI-1 image, MINC image, gz compressed MINC image, DICOM image, TIFF image, XBM image, PBM image, PGM image, BMP image, XPM image, PPM image, gz compressed NIFTI-1 image, TIFF(.tif) image, ECAT i image, PNG image, JPEG image, MNG image, GIF image, SPM image, ECAT v image

csf_Nat :

GIS image, VIDA image, NIFTI-1 image, MINC image, gz compressed MINC image, DICOM image, TIFF image, XBM image, PBM image, PGM image, BMP image, XPM image, PPM image, gz compressed NIFTI-1 image, TIFF(.tif) image, ECAT i image, PNG image, JPEG image, MNG image, GIF image, SPM image, ECAT v image

biasCorrected :

GIS image, VIDA image, NIFTI-1 image, MINC image, gz compressed MINC image, DICOM image, TIFF image, XBM image, PBM image, PGM image, BMP image, XPM image, PPM image, gz compressed NIFTI-1 image, TIFF(.tif) image, ECAT i image, PNG image, JPEG image, MNG image, GIF image, SPM image, ECAT v image

snMat :

Matlab file

snInvMat :

Matlab file