nipype.interfaces.camino.odf module

LinRecon

Link to code

Bases: StdOutCommandLine

Wrapped executable: linrecon.

Runs a linear transformation in each voxel.

Reads a linear transformation from the matrix file assuming the imaging scheme specified in the scheme file. Performs the linear transformation on the data in every voxel and outputs the result to the standard output. The output in every voxel is actually:

[exit code, ln(S(0)), p1, ..., pR]

where p1, …, pR are the parameters of the reconstruction. Possible exit codes are:

    1. No problems.

    1. Bad data replaced by substitution of zero.

The matrix must be R by N+M where N+M is the number of measurements and R is the number of parameters of the reconstruction. The matrix file contains binary double-precision floats. The matrix elements are stored row by row.

Example

First run QBallMX and create a linear transform matrix using Spherical Harmonics (sh).

>>> import nipype.interfaces.camino as cam
>>> qballmx = cam.QBallMX()
>>> qballmx.inputs.scheme_file = 'A.scheme'
>>> qballmx.inputs.basistype = 'sh'
>>> qballmx.inputs.order = 4
>>> qballmx.run()

Then run it over each voxel using LinRecon

>>> qballcoeffs = cam.LinRecon()
>>> qballcoeffs.inputs.in_file = 'SubjectA.Bfloat'
>>> qballcoeffs.inputs.scheme_file = 'A.scheme'
>>> qballcoeffs.inputs.qball_mat = 'A_qmat.Bdouble'
>>> qballcoeffs.inputs.normalize = True
>>> qballcoeffs.run()
Mandatory Inputs:

  • in_file (a pathlike object or string representing an existing file) – Voxel-order data filename. Maps to a command-line argument: %s (position: 1).

  • qball_mat (a pathlike object or string representing an existing file) – Linear transformation matrix. Maps to a command-line argument: %s (position: 3).

  • scheme_file (a pathlike object or string representing an existing file) – Specifies the scheme file for the diffusion MRI data. Maps to a command-line argument: %s (position: 2).

Optional Inputs:

  • args (a string) – Additional parameters to the command. Maps to a command-line argument: %s.

  • bgmask (a pathlike object or string representing an existing file) – Background mask. Maps to a command-line argument: -bgmask %s.

  • environ (a dictionary with keys which are a bytes or None or a value of class ‘str’ and with values which are a bytes or None or a value of class ‘str’) – Environment variables. (Nipype default value: {})

  • log (a boolean) – Transform the log measurements rather than the measurements themselves. Maps to a command-line argument: -log.

  • normalize (a boolean) – Normalize the measurements and discard the zero measurements before the linear transform. Maps to a command-line argument: -normalize.

  • out_file (a pathlike object or string representing a file) – Maps to a command-line argument: > %s (position: -1).

Outputs:

recon_data (a pathlike object or string representing an existing file) – Transformed data.

MESD

Link to code

Bases: StdOutCommandLine

Wrapped executable: mesd.

MESD is a general program for maximum entropy spherical deconvolution. It also runs PASMRI, which is a special case of spherical deconvolution. The input data must be in voxel order.

The format of the output in each voxel is: { exitcode, ln(A^star(0)), lambda_0, lambda_1, …, lambda_N }

The exitcode contains the results of three tests. The first test thresholds the maximum relative error between the numerical integrals computed at con- vergence and those computed using a larger test point set; if the error is greater than a threshold the exitcode is increased from zero to one as a warning; if it is greater than a larger threshold the exitcode is increased to two to suggest failure. The second test thresholds the predicted error in numerical integrals computed using the test point set; if the predicted error is greater than a threshold the exitcode is increased by 10. The third test thresholds the RMS error between the measurements and their predictions from the fitted deconvolution; if the errors are greater than a threshold, the exit code is increased by 100. An exitcode of 112 means that all three tests were failed and the result is likely to be unreliable. If all is well the exitcode is zero. Results are often still reliable even if one or two of the tests are failed.

Other possible exitcodes are:

  • 5 - The optimization failed to converge

  • -1 - Background

  • -100 - Something wrong in the MRI data, e.g. negative or zero measurements, so that the optimization could not run.

The standard MESD implementation is computationally demanding, particularly as the number of measurements increases (computation is approximately O(N^2), where N is the number of measurements). There are two ways to obtain significant computational speed-up:

i) Turn off error checks and use a small point set for computing numerical integrals in the algorithm by adding the flag -fastmesd. Sakaie CDMRI 2008 shows that using the smallest point set (-basepointset 0) with no error checks usually has only a minor effect on the output of the algorithm, but provides a major reduction in computation time. You can increase the point set size using -basepointset with an argument higher than 0, which may produce better results in some voxels, but will increase computation time, which approximately doubles every time the point set index increases by 1.

ii) Reduce the complexity of the maximum entropy encoding using -mepointset <X>. By default <X> = N, the number of measurements, and is the number of parameters in the max. ent. representation of the output function, ie the number of lambda parameters, as described in Jansons and Alexander Inverse Problems 2003. However, we can represent the function using less components and <X> here specifies the number of lambda parameters. To obtain speed-up, set <X> < N; complexity become O(<X>^2) rather than O(N^2). Note that <X> must be chosen so that the camino/PointSets directory contains a point set with that number of elements. When -mepointset decreases, the numerical integration checks make less and less of a difference and smaller point sets for numerical integration (see -basepointset) become adequate. So when <X> is low -fastmesd is worth using to get even more speed-up.

The choice of <X> is a parameter of the technique. Too low and you lose angular resoloution; too high and you see no computational benefit and may even suffer from overfitting. Empirically, we have found that <X>=16 often gives good results and good speed up, but it is worth trying a few values a comparing performance. The reduced encoding is described in the following ISMRM abstract: Sweet and Alexander “Reduced Encoding Persistent Angular Structure” 572 ISMRM 2010.

Example

Run MESD on every voxel of the data file SubjectA.Bfloat using the PASMRI kernel.

>>> import nipype.interfaces.camino as cam
>>> mesd = cam.MESD()
>>> mesd.inputs.in_file = 'SubjectA.Bfloat'
>>> mesd.inputs.scheme_file = 'A.scheme'
>>> mesd.inputs.inverter = 'PAS'
>>> mesd.inputs.inverter_param = 1.4
>>> mesd.run()
Mandatory Inputs:

  • in_file (a pathlike object or string representing an existing file) – Voxel-order data filename. Maps to a command-line argument: -inputfile %s (position: 1).

  • inverter (‘SPIKE’ or ‘PAS’) – The inversion index specifies the type of inversion to perform on the data. The currently available choices are:

    Inverter name

    Inverter parameters

    SPIKE

    bd (b-value x diffusivity along the fibre.)

    PAS

    r

    Maps to a command-line argument: -filter %s (position: 2).

  • inverter_param (a float) – Parameter associated with the inverter. Cf. inverter description formore information. Maps to a command-line argument: %f (position: 3).

  • scheme_file (a pathlike object or string representing an existing file) – Specifies the scheme file for the diffusion MRI data. Maps to a command-line argument: -schemefile %s.

Optional Inputs:

  • args (a string) – Additional parameters to the command. Maps to a command-line argument: %s.

  • bgmask (a pathlike object or string representing an existing file) – Background mask. Maps to a command-line argument: -bgmask %s.

  • environ (a dictionary with keys which are a bytes or None or a value of class ‘str’ and with values which are a bytes or None or a value of class ‘str’) – Environment variables. (Nipype default value: {})

  • fastmesd (a boolean) – Turns off numerical integration checks and fixes the integration point set size at that ofthe index specified by -basepointset.. Maps to a command-line argument: -fastmesd. Requires inputs: mepointset.

  • inputdatatype (‘float’ or ‘char’ or ‘short’ or ‘int’ or ‘long’ or ‘double’) – Specifies the data type of the input file: “char”, “short”, “int”, “long”,”float” or “double”. The input file must have BIG-ENDIAN ordering.By default, the input type is “float”. Maps to a command-line argument: -inputdatatype %s.

  • mepointset (an integer) – Use a set of directions other than those in the scheme file for the deconvolution kernel.The number refers to the number of directions on the unit sphere. For example, “-mepointset 54” uses the directions in “camino/PointSets/Elec054.txt”. Maps to a command-line argument: -mepointset %d.

  • out_file (a pathlike object or string representing a file) – Maps to a command-line argument: > %s (position: -1).

Outputs:

mesd_data (a pathlike object or string representing an existing file) – MESD data.

QBallMX

Link to code

Bases: StdOutCommandLine

Wrapped executable: qballmx.

Generates a reconstruction matrix for Q-Ball. Used in LinRecon with the same scheme file to reconstruct data.

Examples

To create a linear transform matrix using Spherical Harmonics (sh).

>>> import nipype.interfaces.camino as cam
>>> qballmx = cam.QBallMX()
>>> qballmx.inputs.scheme_file = 'A.scheme'
>>> qballmx.inputs.basistype = 'sh'
>>> qballmx.inputs.order = 6
>>> qballmx.run()

To create a linear transform matrix using Radial Basis Functions (rbf). This command uses the default setting of rbf sigma = 0.2618 (15 degrees), data smoothing sigma = 0.1309 (7.5 degrees), rbf pointset 246

>>> import nipype.interfaces.camino as cam
>>> qballmx = cam.QBallMX()
>>> qballmx.inputs.scheme_file = 'A.scheme'
>>> qballmx.run()

The linear transform matrix from any of these two examples can then be run over each voxel using LinRecon

>>> qballcoeffs = cam.LinRecon()
>>> qballcoeffs.inputs.in_file = 'SubjectA.Bfloat'
>>> qballcoeffs.inputs.scheme_file = 'A.scheme'
>>> qballcoeffs.inputs.qball_mat = 'A_qmat.Bdouble'
>>> qballcoeffs.inputs.normalize = True
>>> qballcoeffs.inputs.bgmask = 'brain_mask.nii'
>>> qballcoeffs.run()
Mandatory Inputs:

scheme_file (a pathlike object or string representing an existing file) – Specifies the scheme file for the diffusion MRI data. Maps to a command-line argument: -schemefile %s.

Optional Inputs:

  • args (a string) – Additional parameters to the command. Maps to a command-line argument: %s.

  • basistype (‘rbf’ or ‘sh’) – Basis function type. “rbf” to use radial basis functions “sh” to use spherical harmonics. Maps to a command-line argument: -basistype %s. (Nipype default value: rbf)

  • environ (a dictionary with keys which are a bytes or None or a value of class ‘str’ and with values which are a bytes or None or a value of class ‘str’) – Environment variables. (Nipype default value: {})

  • order (an integer) – Specific to sh. Maximum order of the spherical harmonic series. Default is 4. Maps to a command-line argument: -order %d.

  • out_file (a pathlike object or string representing a file) – Maps to a command-line argument: > %s (position: -1).

  • rbfpointset (an integer) – Specific to rbf. Sets the number of radial basis functions to use. The value specified must be present in the Pointsets directory. The default value is 246. Maps to a command-line argument: -rbfpointset %d.

  • rbfsigma (a float) – Specific to rbf. Sets the width of the interpolating basis functions. The default value is 0.2618 (15 degrees). Maps to a command-line argument: -rbfsigma %f.

  • smoothingsigma (a float) – Specific to rbf. Sets the width of the smoothing basis functions. The default value is 0.1309 (7.5 degrees). Maps to a command-line argument: -smoothingsigma %f.

Outputs:

qmat (a pathlike object or string representing an existing file) – Q-Ball reconstruction matrix.

SFPeaks

Link to code

Bases: StdOutCommandLine

Wrapped executable: sfpeaks.

Finds the peaks of spherical functions.

This utility reads coefficients of the spherical functions and outputs a list of peak directions of the function. It computes the value of the function at each of a set of sample points. Then it finds local maxima by finding all points at which the function is larger than for any other point within a fixed search radius (the default is 0.4). The utility then uses Powell’s algorithm to optimize the position of each local maximum. Finally the utility removes duplicates and tiny peaks with function value smaller than some threshold, which is the mean of the function plus some number of standard deviations. By default the program checks for con- sistency with a second set of starting points, but skips the optimization step. To speed up execution, you can turn off the con- sistency check by setting the noconsistencycheck flag to True.

By default, the utility constructs a set of sample points by randomly rotating a unit icosahedron repeatedly (the default is 1000 times, which produces a set of 6000 points) and concatenating the lists of vertices. The ‘pointset = <index>’ attribute can tell the utility to use an evenly distributed set of points (index 0 gives 1082 points, 1 gives 1922, 2 gives 4322, 3 gives 8672, 4 gives 15872, 5 gives 32762, 6 gives 72032), which is quicker, because you can get away with fewer points. We estimate that you can use a factor of 2.5 less evenly distributed points than randomly distributed points and still expect similar performance levels.

The output for each voxel is:

  • exitcode (inherited from the input data).

  • ln(A(0))

  • number of peaks found.

  • flag for consistency with a repeated run (number of directions is the same and the directions are the same to within a threshold.)

  • mean(f).

  • std(f).

  • direction 1 (x, y, z, f, H00, H01, H10, H11).

  • direction 2 (x, y, z, f, H00, H01, H10, H11).

  • direction 3 (x, y, z, f, H00, H01, H10, H11).

H is the Hessian of f at the peak. It is the matrix:

[d^2f/ds^2 d^2f/dsdt]
[d^2f/dtds d^2f/dt^2]
= [H00 H01]
  [H10 H11]

where s and t are orthogonal coordinates local to the peak.

By default the maximum number of peak directions output in each voxel is three. If less than three directions are found, zeros are output for later directions. The peaks are ordered by the value of the function at the peak. If more than the maximum number of directions are found only the strongest ones are output. The maximum number can be changed setting the ‘numpds’ attribute.

The utility can read various kinds of spherical function, but must be told what kind of function is input using the ‘inputmodel’ attribute. The description of the ‘inputmodel’ attribute lists additional information required by SFPeaks for each input model.

Example

First run QBallMX and create a linear transform matrix using Spherical Harmonics (sh).

>>> import nipype.interfaces.camino as cam
>>> sf_peaks = cam.SFPeaks()
>>> sf_peaks.inputs.in_file = 'A_recon_params.Bdouble'
>>> sf_peaks.inputs.inputmodel = 'sh'
>>> sf_peaks.inputs.order = 4
>>> sf_peaks.inputs.density = 100
>>> sf_peaks.inputs.searchradius = 1.0
>>> sf_peaks.run()
Mandatory Inputs:

  • in_file (a pathlike object or string representing an existing file) – Voxel-order data of spherical functions. Maps to a command-line argument: -inputfile %s.

  • inputmodel (‘sh’ or ‘maxent’ or ‘rbf’) – Type of functions input via in_file. Currently supported options are: sh - Spherical harmonic series. Specify the maximum order of the SH series with the “order” attribute if different from the default of 4. maxent - Maximum entropy representations output by MESD. The reconstruction directions input to MESD must be specified. By default this is the same set of gradient directions (excluding zero gradients) in the scheme file, so specify the “schemefile” attribute unless the “mepointset” attribute was set in MESD. rbf - Sums of radial basis functions. Specify the pointset with the attribute “rbfpointset” if different from the default. See QBallMX. Maps to a command-line argument: -inputmodel %s.

Optional Inputs:

  • args (a string) – Additional parameters to the command. Maps to a command-line argument: %s.

  • density (an integer) – The number of randomly rotated icosahedra to use in constructing the set of points for random sampling in the peak finding algorithm. Default is 1000, which works well for very spiky maxent functions. For other types of function, it is reasonable to set the density much lower and increase the search radius slightly, which speeds up the computation. Maps to a command-line argument: -density %d.

  • environ (a dictionary with keys which are a bytes or None or a value of class ‘str’ and with values which are a bytes or None or a value of class ‘str’) – Environment variables. (Nipype default value: {})

  • mepointset (an integer) – Use a set of directions other than those in the scheme file for the deconvolution kernel. The number refers to the number of directions on the unit sphere. For example, “mepointset = 54” uses the directions in “camino/PointSets/Elec054.txt” Use this option only if you told MESD to use a custom set of directions with the same option. Otherwise, specify the scheme file with the “schemefile” attribute. Maps to a command-line argument: -mepointset %d.

  • noconsistencycheck (a boolean) – Turns off the consistency check. The output shows all consistencies as true. Maps to a command-line argument: -noconsistencycheck.

  • numpds (an integer) – The largest number of peak directions to output in each voxel. Maps to a command-line argument: -numpds %d.

  • order (an integer) – Specific to sh. Maximum order of the spherical harmonic series. Maps to a command-line argument: -order %d.

  • out_file (a pathlike object or string representing a file) – Maps to a command-line argument: > %s (position: -1).

  • pdthresh (a float) – Base threshold on the actual peak direction strength divided by the mean of the function. The default is 1.0 (the peak must be equal or greater than the mean). Maps to a command-line argument: -pdthresh %f.

  • pointset (an integer) – To sample using an evenly distributed set of points instead. The integer can be 0, 1, …, 7. Index 0 gives 1082 points, 1 gives 1922, 2 gives 3002, 3 gives 4322, 4 gives 5882, 5 gives 8672, 6 gives 12002, 7 gives 15872. Maps to a command-line argument: -pointset %d.

  • rbfpointset (an integer) – Specific to rbf. Sets the number of radial basis functions to use. The value specified must be present in the Pointsets directory. The default value is 246. Maps to a command-line argument: -rbfpointset %d.

  • scheme_file (a pathlike object or string representing an existing file) – Specific to maxent. Specifies the scheme file. Maps to a command-line argument: %s.

  • searchradius (a float) – The search radius in the peak finding algorithm. The default is 0.4 (cf. “density”). Maps to a command-line argument: -searchradius %f.

  • stdsfrommean (a float) – This is the number of standard deviations of the function to be added to the “pdthresh” attribute in the peak directions pruning. Maps to a command-line argument: -stdsfrommean %f.

Outputs:

peaks (a pathlike object or string representing an existing file) – Peaks of the spherical functions.