AimsSegmentationEM
    ------------------

Segmentation by Expectation-Maximization.


Options :

-i | --input <file name (read only): { Volume of DOUBLE, Volume of FLOAT,
    Volume of RGB, Volume of RGBA, Volume of S16, Volume of S32, Volume of S8,
    Volume of U16, Volume of U32, Volume of U8 }>  Input image.

[ -c | --channel <S32> ]
    Channel to use if color image: 0 [red], 1 [green], 2 [blue], 3 [alpha], 4
    [norm] (default)

[ -p | --input-prior-map <string> ]
    Class probability image (classes are in the T dimension)

[ -cm | --input-clique-matrix <string> ]
    MRF clique matrix (default: diag(1))

[ -m | --input-mask <string> ]
    Mask in which to compute the classification

[ -bf | --input-bias-field <string> ]
    Input multiplicative bias field

[ -o | --output <string> ]
    Output segmentation.

[ -op | --output-probabilities <string> ]
    Output probabilites

[ -onb | --output-no-bias <string> ]
    Output bias corrected image

[ -obf | --output-bias-field <string> ]
    Output multiplicative bias field

[ -n | --n-classes <S32> ]
    Number of classes (only necessary in the most basic case with no parameter
    provided)

[ --mu <vector of FLOAT> ]
    Initial mean values for the normal distributions.

[ --sigma <vector of FLOAT> ]
    Initial standard deviation values for the normal distributions.

[ --alpha <vector of FLOAT> ]
    Initial values for the stationary prior. They only make sense if no prior
    map is provided. If prior type is non-stationary, priors will be
    initialized with these values but their updated values won't stay
    stationary.

[ --stop-crit <string> ]
    Stop criterion for Log-Likelihood Sum:
    - g[ain] (default)
    - a[bs diff]

[ --stop-value <FLOAT> ]
    Stop criterion value (default: 0.01)

[ --iter-min <S32> ]
    Minimum number of iterations (default: 2)

[ --iter-max <S32> ]
    Maximum number of iterations (default: 100/nclasses)

[ --init-normal <string> ]
    Initialization mode for the normal distribution. It only makes sense if no
    mu/sigma and no prior map was provided.
    - r[andom] (default)
    - f[rom prior]
    - k[means]
    - kmeans++
    - p[rovided]

[ --prior-type <string> ]
    Force prior type:
    - s[tationary] (default if no prior map provided)
    - n[on-stationary]
    - m[ixed] (default if a prior map is provided)

[ --parameterize-prior <boolean> ]
    Make the prior a parameter in the EM algorithm. i.e. its values will be
    updated at each iteration with its maximum-likelihood estimation. If the
    prior is mixed, only the global class proportion, no the non-stationary
    prior, is updated. (default: true)

[ --smooth <FLOAT> ]
    Probability image smoothing sigma (no smoothing: 0, default: biggest voxel
    size)

[ --log <boolean> ]
    Log-transform intensities before EM (default: true)

[ --denoising <boolean> ]
    Enable denoising during EM (default: true)

[ --bias <string> ]
    Perform bias correction:
    - s[pline]
    - l[ow pass] (default)
    - n[one]

[ --bias-grid <vector of FLOAT> ]
    Bias grid (default: 10 10 10)

[ --bias-unit <string> ]
    Bias grid unit:
    - m[m]
    - v[oxel]
    - %[ of mri dimensions] (default)

[ --rf <string> ]
    Use of random fields for spatial regularization:
    - i[ter] (at each step of the EM algorithm) (default)
    - e[nd]  (at the end of a classical mixture model)
    - n[one] (no spatial regularization)

[ --rf-type <string> ]
    Class estimation method for the random field mode:
    - m[arkov] (clique potentials are kept as is and can be null) (default)
    - g[ibbs] (clique potential are exp transformed and are assured to be
    positive).

[ --rf-order <S32> ]
    Connectivity order for random field: 
    - 1 (2d: 4 / 3d: 6) (default)
    - 2 (2d: 8 / 3d: 18)
    - 3 (3d: 26)

[ --rf-eval <string> ]
    Class p method for the random field mode:
    - m[mp] (Maximum Marginal Probabilities) (default)
    - i[cm] (Iterated Conditional Mode).

[ --icm-iter-min <S32> ]
    Minimum number of ICM iterations (default: 1)

[ --icm-iter-max <S32> ]
    Maximum number of ICM iterations (default: 5)

[ --beta <FLOAT> ]
    Modulation of the neighborhood priors used in random fields. (default: 1)
    It consists in applying a beta power to the clique matrix before the
    eventual log transform (with gibbs fields). If beta == 0, all clique
    potentials are equal. If beta == 1, they are kept as is. In the gibbs case
    with the default diag(1) clique matrix, this is equivalent to Besag's beta
    parameter (in 'On the Statistical Analysis of Dirty Pictures', Julian Besag,
    1986)

[ --labels <vector of S32> ]
    List of labels corresponding to probability indices. (default: same). If
    several indices possess the same labels, their probabilities are merged.

[ --ignore <vector of S32> ]
    List of labels to exclude from the prior map.

[ --exclude <vector of S32> ]
    List of labels to ignore when estimating bias field.

[ -w | --write-inter-files <string> ]
    Debug directory (if set, write intermediate files) (default: false)

[ --output_4d_volumes <S32> ]
    Write 4D images as several 3D files (0 = series of 3D volumes, 1 = one 4D
    volume, no value = use .aimsrc configuration files)

[ --debugLevel | -d <S32> ]
    Set debug information level (default = 0)

[ --verbose | -v [ <S32> ] ]
    Set verbosity level (default = 0, without value = 1)

[ -h | --help <boolean> ]
    show help message

[ --version <boolean> ]
    show Cartograph version

[ --info <boolean> ]
    show libraries information (install, plugins, paths, etc.)

[ --optionsfile <string> ]
    Read additional commandline options from the specified file (one switch or
    value per line)