DWI_Denoise
===========

Denoising of the DWI with DIPY (MP-PCA or Patch2Self). Use it first in the pipeline, on the raw images, before any interpolation and before GIBBS\_Correct

:Command: ``dmriprep``
:Module version: 0.1
:Module type: ``prep``
:Attributes: ``deform_image``
:Notes: `DWI_Denoise README <https://github.com/NIRALUser/DTIPlayground/blob/master/dtiplayground/dmri/preprocessing/modules/DWI_Denoise/README.md>`_
:Source: `dtiplayground/dmri/preprocessing/modules/DWI_Denoise <https://github.com/NIRALUser/DTIPlayground/blob/master/dtiplayground/dmri/preprocessing/modules/DWI_Denoise>`_

Protocol options
----------------

``method``
    Denoising method

    :Type: ``list``
    :Default: ``mppca``
    :Choices:
        - ``mppca`` -- Marchenko-Pastur PCA of local patches (Veraart et al. 2016), the method of MRtrix dwidenoise. Also writes the noise level map (DWI\_noise\_sigma)
        - ``patch2self`` -- Self-supervised denoising, each volume predicted from the other volumes (Fadnavis et al. 2020); b-values up to 50 are the b=0 volumes (as in QSIPrep). Also writes the RMS of the removed signal (DWI\_noise\_residual)


``patchRadius``
    Radius of the patches of MP-PCA in voxels (2 is a 5x5x5 patch). 0 chooses the smallest patch with at least as many voxels as there are volumes

    :Type: ``integer``
    :Default: ``0``


``patch2selfModel``
    Regression model of Patch2Self

    :Type: ``list``
    :Default: ``ols``
    :Choices:
        - ``ols`` -- Ordinary least squares
        - ``ridge`` -- Ridge regression
        - ``lasso`` -- Lasso regression