API changes

Here we provide information about functions or classes that have been removed, renamed or are deprecated (not recommended) during different release circles.

DIPY 1.7.0 changes


  • Change in dipy.denoise.localpca, function genpca can use fewer images than patch voxels.

  • Change in dipy.denoise.pca_noise_estimate, function pca_noise_estimate has new argument images_as_samples

DIPY 1.6.0 changes

DIPY 1.5.0 changes


  • FURY minimum version is 0.8.0

  • Distutils has been dropped

  • dipy.io.bvectxt module is deprecated and will be removed


  • The default option in the command line for Patch2Self ‘ridge’ -> ‘ols’


  • Change in dipy.tracking.pmf
    • The parent class PmfGen has new mandatory parameter sphere. The sphere vertices correspond to the spherical distribution of the pmf values.

    • The parent class PmfGen has new function get_pmf_value(point, xyz) which return the pmf value at location point and orientation xyz.


  • The deprecated from dipy.segment.metric import ResampleFeature was removed and replaced by from dipy.segment.featurespeed import ResampleFeature.

DIPY 1.4.1 changes


  • The name of the argument for the number of cores/threads has been standardized to:
    • num_threads for OpenMP parallelization.

    • num_processes for parallelization using multiprocessing package.

  • Change in the parallelization logic when using OpenMP:
    • If num_threads = None the value of OMP_NUM_THREADS environment variable is used. If it is not set then all available threads are used.

    • If num_threads > 0 that number is used as the number of threads.

    • If num_threads < 0 the maximum between 1 and num_cpu_cores - |num_threads + 1| is selected. If -1 then all available threads are used.

    • If num_threads = 0 an error is raised.

  • Change in the parallelization logic when using multiprocessing package:
    • The same as with OpenMP with the difference that num_processes = None uses all cores directly.


  • Change in DirectionGetters:
    • The deprecated dipy.direction.closest_peak_direction_getter.BaseDirectionGetter was removed and replaced by dipy.direction.closest_peak_direction_getter.BasePmfDirectionGetter.

    • The deprecated dipy.reconst.EuDXDirectionGetter was removed and replaced by dipy.reconst.eudx_direction_getter.EuDXDirectionGetter.

DIPY 1.4.0 changes

  • Migration from Tavis to Azure

DIPY 1.3.0 changes

  • new dependency added: tqdm


  • The argument interp of the method dipy.align.imaffine.AffineMap.transform has been renamed interpolation.

  • The argument interp of the method dipy.align.imaffine.AffineMap.transform_inverse has been renamed interpolation.


  • The tissue segmentation method dipy.segment.TissueClassifierHMRF now checks the tolerance-based stopping criterion at every iteration (previously it was only checked every 10th iteration). This may result in earlier termination of iterations than with previous releases.

DIPY 1.2.0 changes


The dipy.reconst.csdeconv.auto_response has been renamed dipy.reconst.csdeconv.auto_response_ssst.

The dipy.reconst.csdeconv.response_from_mask has been renamed dipy.reconst.csdeconv.response_from_mask_ssst.

The dipy.sims.voxel.multi_shell_fiber_response has been moved to dipy.reconst.mcsd.multi_shell_fiber_response.


In prior releases, for users with SciPy < 1.5, a memory overlap bug occurs in multi_median, causing an overly smooth output. This has now been fixed, regardless of the user’s installed SciPy version. Users of this function via median_otsu thresholding should check the output of their image processing pipelines after the 1.2.0 release to make sure thresholding is still operating as expected (if not, try readjusting the median_radius parameter).


The dipy.reconst.peak_direction_getter.EuDXDirectionGetter has been renamed dipy.reconst.eudx_direction_getter.EuDXDirectionGetter.

The command line dipy_track_local has been renamed dipy_track.


The dipy.core.gradients.unique_bvals has been renamed dipy.core.gradients.unique_bvals_magnitude.


  • Use window.Scene() instead of window.Renderer().

  • Use scene.clear() instead of window.rm_all(scene).

  • Use scene.clear() instead of window.clear(scene).

DIPY 1.1.1 changes


img.get_data() is deprecated since Nibabel 3.0.0. Using np.asanyarray(img.dataobj) instead of img.get_data().


dipy.io.streamlines.StatefulTractogram can be created by another one.


dipy_nlmeans command lines have been renamed dipy_denoise_nlmeans.


get_data has been deprecated by Nibabel and replaced by get_fdata. This modification has been applied to all the codebase. The default datatype is now float64.

DIPY 1.0.0 changes

Some of the changes introduced in the 1.0 release will break backward compatibility with previous versions. This release is compatible with Python 3.5+


The spherical harmonics bases mrtrix and fibernav have been renamed to tournier07 and descoteaux07 after the deprecation cycle started in the 0.15 release.

We changed dipy.data.default_sphere from symmetric724 to repulsion724 which is more evenly distributed.


The API of dipy.segment.mask.median_otsu has changed in the following ways: if you are providing a 4D volume, vol_idx is now a required argument. The order of parameters has also changed.

Tractogram loading and saving

The API of dipy.io.streamlines.load_tractogram and dipy.io.streamlines.save_tractogram has changed in the following ways: When loading trk, tck, vtk, fib, or dpy) a reference nifti file is needed to guarantee proper spatial transformation handling.

Spatial transformation handling

Functions from dipy.tracking.streamlines were modified to enforce the affine parameter and uniform docstrings. deform_streamlines select_by_rois, orient_by_rois, _extract_vals and values_from_volume.

Functions from dipy.tracking.utils were modified to enforce the affine parameter and uniform docstring. density_map connectivity_matrix, seeds_from_mask, random_seeds_from_mask, target, target_line_based, near_roi, length and path_length were all modified.

The function affine_for_trackvis, move_streamlines, flexi_tvis_affine and get_flexi_tvis_affine were deleted.

Functions from dipy.tracking.life were modified to enforce the affine parameter and uniform docstring. voxel2streamline, setup and fit from class FiberModel were all modified.

afq_profile from dipy.stats.analysis was modified similarly.


  • dipy.sims.voxel.SingleTensor has been replaced by dipy.sims.voxel.single_tensor

  • dipy.sims.voxel.MultiTensor has been replaced by dipy.sims.voxel.multi_tensor

  • dipy.sims.voxel.SticksAndBall has been replaced by dipy.sims.voxel.sticks_and_ball


All interpolation functions have been moved to a new module name dipy.core.interpolation


The voxel_size parameter has been removed from the following function:

  • dipy.tracking.utils.connectivity_matrix

  • dipy.tracking.utils.density_map

  • dipy.tracking.utils.stremline_mapping

  • dipy.tracking._util._mapping_to_voxel

The dipy.reconst.peak_direction_getter.PeaksAndMetricsDirectionGetter has been renamed dipy.reconst.peak_direction_getter.EuDXDirectionGetter.

The LocalTracking and ParticleFilteringTracking functions were moved from dipy.tracking.local.localtracking to dipy.tracking.local_tracking. They now need to be imported from dipy.tracking.local_tracking.

  • functions argument tissue_classifier were renamed stopping_criterion

The TissueClassifier were renamed StoppingCriterion and moved from dipy.tracking.local.tissue_classifier to dipy.tracking.stopping_criterion. They now need to be imported from dipy.tracking.stopping_criterion.

  • TissueClassifier -> StoppingCriterion

  • BinaryTissueClassifier -> BinaryStoppingCriterion

  • ThresholdTissueClassifier -> ThresholdStoppingCriterion

  • ConstrainedTissueClassifier -> AnatomicalStoppingCriterion

  • ActTissueClassifier -> ActStoppingCriterion

  • CmcTissueClassifier -> CmcStoppingCriterion

The dipy.tracking.local.tissue_classifier.TissueClass was renamed dipy.tracking.stopping_criterion.StreamlineStatus.

The EuDX tracking function has been removed. EuDX tractography can be performed using dipy.tracking.local_tracking using dipy.reconst.peak_direction_getter.EuDXDirectionGetter.


dipy.io.trackvis has been removed. Use dipy.io.streamline instead.


  • dipy.external package has been removed.

  • dipy.fixes package has been removed.

  • dipy.segment.quickbundes module has been removed.

  • dipy.reconst.peaks module has been removed.

  • Compatibility with Python 2.7 has been removed.

DIPY 0.16 Changes


Welcome to the new module dipy.viz.stats. This module will be used to integrate various analyses.


  • New option to adjust the number of threads for SLR in Recobundles

  • The tracking algorithm excludes the stop point inside the mask during the tracking process.


  • Replacement of Nose by Pytest

DIPY 0.15 Changes


load_tck and save_tck from dipy.io.streamline have been added. They are highly recommended for managing streamlines.

Gradient Table

The default value of b0_thresold has been changed(from 0 to 50). This change can impact your algorithm. If you want to assure that your code runs in exactly the same manner as before, please initialize your gradient table with the keyword argument b0_threshold set to 0.


dipy.viz.fvtk module has been removed. Use dipy.viz.* instead. This implies the following important changes: - Use from dipy.viz import window, actor instead of from dipy.viz import fvtk`. - Use ``window.Renderer() instead of fvtk.ren(). - All available actors are in dipy.viz.actor instead of dipy.fvtk.actor. - UI elements are available in dipy.viz.ui.

dipy.viz depends on the FURY package. To learn more about FURY, go to https://fury.gl

DIPY 0.14 Changes


dipy.io.trackvis module is deprecated. Use dipy.io.streamline instead. Furthermore, load_trk and save_trk from dipy.io.streamline is highly recommended for managing streamlines. When you create streamlines, you should use from dipy.tracking.streamlines import Streamlines. This new object uses much less memory and it is easier to process.


dipy.viz.fvtk module is deprecated. Use dipy.viz.* instead. This implies the following important changes: - Use from dipy.viz import window, actor instead of from dipy.viz import fvtk`. - Use ``window.Renderer() instead of fvtk.ren(). - All available actors are in dipy.viz.actor instead of dipy.fvtk.actor. - UI elements are available in dipy.viz.ui.

DIPY 0.13 Changes

No major API changes.


dipy.io.trackvis module will be deprecated on release 0.14. Use dipy.io.streamline instead. dipy.viz.fvtk module will be deprecated on release 0.14. Use dipy.viz.ui instead.

DIPY 0.12 Changes

Dropped support for Python 2.6*

It has been 6 years since the release of Python 2.7, and multiple other versions have been released since. As far as we know, DIPY still works well on Python 2.6, but we no longer test on this version, and we recommend that users upgrade to Python 2.7 or newer to use DIPY.


probabilistic_direction_getter.ProbabilisticDirectionGetter input parameters have changed. Now the optional parameter pmf_threshold=0.1 (previously fixed to 0.0) removes directions with probability lower than pmf_threshold from the probability mass function (pmf) when selecting the tracking direction.


The default of DKI model fitting was changed from “OLS” to “WLS”.

The default max_kurtosis of the functions axial_kurtosis, mean_kurtosis, radial_kurotis was changed from 3 to 10.


Prefer using the UI elements in dipy.viz.ui rather than dipy.viz.widgets.


Use the module nibabel.streamlines for saving trk files and not nibabel.trackvis. Requires upgrading to nibabel 2+.

DIPY 0.10 Changes

New visualization module

fvtk.slicer input parameters have changed. Now the slicer function is more powerful and supports RGB images too. See tutorial viz_slice.py for more information.

Interpolation The default behavior of the function core.sphere.interp_rbf has changed. The default smoothing parameter is now set to 0.1 (previously 0). In addition, the default norm is now angle (was previously euclidean_norm). Note that the use of euclidean_norm is discouraged, and this norm will be deprecated in the 0.11 release cycle.


The following utility functions from vector_fields module were renamed:

warp_2d_affine is now transform_2d_affine warp_2d_affine_nn is now transform_2d_affine_nn warp_3d_affine is now transform_3d_affine warp_3d_affine_nn is now transform_3d_affine_nn

DIPY 0.9 Changes

GQI integration length

The calculation of integration length in GQI2 now matches the calculation in the ‘standard’ method. Using values of 1-1.3 for either is recommended (see docs and references therein).

DIPY 0.8 Changes


The module peaks is now available from dipy.direction and it can still be accessed from dipy.reconst but it will be completely removed in version 0.10.


The function resample from dipy.align.aniso2iso is deprecated. Please, use instead reslice from dipy.align.reslice. The module aniso2iso will be completely removed in version 0.10.

Changes between 0.7.1 and 0.6


The function peaks_from_model is now available from dipy.reconst.peaks . Please replace all imports like:

from dipy.reconst.odf import peaks_from_model


from dipy.reconst.peaks import peaks_from_model


The function target from dipy.tracking.utils now takes an affine transform instead of a voxel sizes array. Please update all code using target in a way similar to this:

img = nib.load(anat)
voxel_dim = img.header['pixdim'][1:4]
streamlines = utils.target(streamlines, img.get_data(), voxel_dim)

to something similar to:

img = nib.load(anat)
streamlines = utils.target(streamlines, img.get_data(), img.affine)