DWI distortion correction using dwifslpreproc

The dwifslpreproc script, responsible for performing general pre-processing of DWI series, has been completely re-designed as part of the MRtrix3 3.0_RC1 update. Although the ‘guts’ of the script are completely new, the fundamental operation - eddy current-induced distortion correction, motion correction, and (optionally) susceptibility-induced distortion correction, using FSL’s eddy / topup / applytopup tools, remains the same. While the user interface remains reasonably similar to that provided previously (examples to come), they are slightly different.

The major benefit of the new design is that MRtrix3 is now capable of not only capturing the relevant phase encoding information from DICOM headers, but also using that information within dwifslpreproc to internally generate the necessary phase encoding table files in order to run these FSL tools. This comes with a number of benefits:

  • It makes it possible to acquire and process a wider range of DWI acquisition designs, without requiring that the user laboriously manually construct the phase encoding tables that these FSL tools require.

  • It means that automated pre-processing pipelines (e.g. these two works-in-progress) can be applied to provided data without requiring manual intervention to specify this information.

  • Over time, as MRtrix 0.3.16 code is used to import DICOMs (and hence capture the phase encoding information) and the relevant code is thoroughly tested, there will be less onus on users to track and specify the type of phase encoding acquisition performed.

Note

Although the dwifslpreproc script is provided as part of MRtrix3 in the hope that users will find it useful, the major image processing steps undertaken by this script are still performed using tools developed at FMRIB and provided as part of FSL. It is therefore essential that the appropriate references be cited whenever this script is used!

The dwifslpreproc script now has four major ‘modes’ of operation, that can be selected at the command-line using the -rpe_* options. Note that exactly one of these options *must* be provided. The following are example use cases; specific parameters, file names etc. may need to be altered to reflect your particular data.

  1. No variation in phase encoding

All DWI volumes are acquired with precisely the same phase encoding direction and EPI readout time, and no additional spin-echo b=0 images are acquired over and above those interleaved within the DWI acquisition. It is therefore impossible to estimate the inhomogeneity field using topup, and eddy will perform motion and eddy current correction only.

Example DICOM image data:

002_-_DWI_phaseAP/

Old usage (i.e. prior to MRtrix 0.3.16):

$ dwifslpreproc AP 002_-_DWI_phaseAP/ dwi_preprocessed.mif -rpe_none

New usage:

$ dwifslpreproc 002_-_DWI_phaseAP/ dwi_preprocessed.mif -rpe_none -pe_dir AP [ -readout_time 0.1 ]

Note that here (and in subsequent examples), providing the EPI readout time manually is optional (if omitted, the ‘sane’ default of 0.1s will be assumed). The precise scaling of this parameter is not expected to influence results provided that the readout time is equivalent for all b=0 / DWI volumes.

  1. Reversed phase encode b=0 pair(s)

All DWI volumes are acquired with precisely the same phase encoding direction and EPI readout time. In addition, one or more pairs of spin-echo b=0 EPI volumes are provided, where half of these volumes have the same phase encoding direction and readout time as the DWIs, and the other half have precisely the opposite phase encoding direction (but the same readout time). These additional images are therefore used to estimate the inhomogeneity field, but do not form part of the output DWI series.

Example DICOM image data:

002_-_ep2dse_phaseAP/
003_-_ep2dse_phasePA/
004_-_DWI_phaseAP/

Old usage (i.e. prior to MRtrix 0.3.16):

$ dwifslpreproc AP 004_-_DWI_phaseAP/ dwi_preprocessed.mif -rpe_pair 002_-_ep2dse_phaseAP/ 003_-_ep2dse_phasePA/

New usage:

$ mrcat 002_-_ep2dse_phaseAP/ 003_-_ep2dse_phasePA/ b0s.mif -axis 3
$ dwifslpreproc 004_-_DWI_phaseAP/ dwi_preprocessed.mif -pe_dir AP -rpe_pair -se_epi b0s.mif [ -readout_time 0.1 ]
  1. Reversed phase encoding for all DWIs

For all diffusion gradient directions & b-values, two image volumes are obtained, with the opposite phase encoding direction with respect to one another. This allows for the combination of the two volumes corresponding to each unique diffusion gradient direction & strength into a single volume, where the relative compression / expansion of signal between the two volumes is exploited.

Example DICOM image data:

002_-_DWI_64dir_phaseLR/
003_-_DWI_64dir_phaseRL/

Old usage (i.e. prior to MRtrix 0.3.16):

$ dwifslpreproc LR 002_-_DWI_64dir_phaseLR/ dwi_preprocessed.mif -rpe_all 003_-_DWI_64dir_phaseRL/

New usage:

$ mrcat 002_-_DWI_64dir_phaseLR/ 003_-_DWI_64dir_phaseRL/ all_DWIs.mif -axis 3
$ dwifslpreproc all_DWIs.mif dwi_preprocessed.mif -pe_dir LR -rpe_all [ -readout_time 0.1 ]

Note that in this particular example, the dwifslpreproc script will in fact extract the b=0 volumes from the input DWIs and use those to estimate the inhomogeneity field with topup. If additional b=0 images are also acquired, and it is desired to instead use those images to estimate the inhomogeneity field only, the -se_epi option can be used.

  1. Arbitrary phase encoding acquisition

In cases where either:

  • An up-to-date version of MRtrix3 has been used to convert from DICOM, such that phase encoding information is embedded in the image header; or:

  • Image data of unknown origin are to be processed by an automated pipeline without user intervention, and therefore phase encoding information must be provided using data files associated with the input images (such as JSON files in the BIDS standard),

it is possible for the dwifslpreproc script to automatically determine the appropriate steps to perform based on the phase encoding configuration of the image data presented to it.

Usage:

$ mrcat <all_input_DWIs> all_dwis.mif -axis 3
$ mrcat <all_extra_b=0_volumes> all_b0s.mif -axis 3   (optional)
$ dwifslpreproc all_dwis.mif dwi_preprocessed.mif -rpe_header [ -se_epi all_b0s.mif ]

Warning

With regards to Option 4 (using phase encoding information as it is stored in the header), note that this functionality is preliminary and should not be trusted blindly. It is impossible for us to check and test all possible usage scenarios. Furthermore, if this information is imported or exported to/from the image header, this requires reorientation due to the way in which MRtrix3 handles image orientations internally, which introduces additional mechanisms by which the tracking of phase encoding orientations may go awry. Results should therefore be checked manually if using / testing this mechanism.

When one of the options 1-3 are used, internally the dwifslpreproc script generates the effective phase encoding table given the user’s images and command-line input; this is what is passed to topup / applytopup / eddy. If one of these options is used, but there is actually phase encoding information found within the image header(s), the script will compare the user’s phase encoding specification against the header contents, and produce a warning if it detects a mismatch (since either the phase encoding design is not what you think it is, or the import of phase encoding information from DICOM is awry; either warrants further investigation).

Configuring FSL commands topup and eddy

The two principal FSL commands with which the MRtrix3 dwifslpreproc script interfaces - topup and eddy - themselves have a wide range of command-line options that may be utilised to alter their behaviour. dwifslpreproc does not provide any command-line options to these commands over and above those absolutely necessary to perform the processing operations; as such, the default parameters / behaviour of these commands as configured in FSL will be utilised. As such, any non-default parameters or behaviour of these commands desired by the user (e.g. eddy outlier replacement) must be specified explicitly. This is achieved by using the dwifslpreproc command-line options -topup_options and -eddy_options, which accept as input a string that will then be propagated down to the topup or eddy command invocation respectively.

For information on the command-line options available within these FSL commands, consult the internal help page of the command installed on your system, and/or the online FSL documentation:

Note

Due to a quirk in the Python library argparse that is utilised by MRtrix3 for command-line parsing, in some instances, some contents of a string provided at the command-line encased in double-quotes may get lost during parsing. To mitigate this, we suggest that whatever text data are provided to topup or eddy via the -topup_options or -eddy_options command-line options contain an additional space at the beginning or end of the string, like thus:

dwifslpreproc ... -eddy_options " --repol" ...

Using eddy’s slice-to-volume motion correction capability

As of September 2017, FSL’s eddy tool has the capability of not only estimating and correcting motion between DWI volumes, but also motion within volumes. Details of this method can be found within the relevant publication. MRtrix3 is capable of supporting this underlying eddy functionality within the wrapping dwifslpreproc script. Below are a few relevant details to assist users in getting this working:

  • At time of writing, only the CUDA version of the eddy executable provides the slice-to-volume correction capability. Therefore, this version must be installed on your system, and CUDA itself must be appropriately set up. Note that with MRtrix3 version 3.0_RC3, presence of the CUDA version of eddy will be automatically detected within your PATH by dwifslpreproc, and this version will be executed in preference to the OpenMP version.

  • eddy’s slice-to-volume correction is triggered by the presence of the --mporder=# command-line option. Therefore, to activate this behaviour, the contents of the -eddy_options command-line option passed to dwifslpreproc must contain this entry.

  • The timing of acquisition of each slice must be known in order to perform slice-to-volume correction. This is provided to eddy via the command-line option --slspec=, where a text file is provided that defines the order in which the slices are acquired within each volume. In dwifslpreproc, there are two ways in which this information can be provided:

    • If you include the string --slspec=path/to/file.txt within the contents of the -eddy_options command-line option, then dwifslpreproc will copy the file to which you have provided the path into the temporary directory created by the script, such that eddy appropriately locates that file.

    • If DICOM conversion & all subsequent processing is performed solely using MRtrix3 commands, and Header key-value pairs are preserved, then where possible, MRtrix3 will store fields “SliceEncodingDirection” and “SliceTiming” based on DICOM information (note that the naming of these fields is consistent with the BIDS specification). dwifslpreproc will then use these fields to internally generate the “slspec” file required by eddy without user intervention; as long as the user does not provide the --slspec= option within -eddy_options.