Data requirements¶

DELTA-SVD analyses multi-directional diffusion MRI that has already been preprocessed. It does not perform image preprocessing itself. The guidance below describes acquisitions and preprocessing that make data well suited to the pipeline. It reflects current best practice rather than a single mandated protocol — sensible deviations are fine, but these recommendations are a safe default.

Required input¶

Each timepoint is described by four files. The DWI must already be preprocessed (see Preprocessing below); DELTA-SVD does not preprocess it for you.

• DWI image — a 4D, preprocessed diffusion-weighted NIfTI (.nii.gz or .nii).
• b-values — an FSL-format bval file matching the DWI.
• b-vectors — an FSL-format bvec file matching the DWI.
• Brain mask — a binary brain mask NIfTI (.nii.gz or .nii) in DWI space (aligned with the DWI image).

A longitudinal run takes one such set per timepoint. For how to pass these on the command line (and the naming convention that lets the bval/bvec/mask be inferred from the DWI path), see Usage.

Acquisition¶

• Field strength — 3T is recommended.
• Resolution and directions — a recommended minimum of 2 mm isotropic voxels and at least 20 diffusion-weighting directions. Higher resolution and more directions improve the tensor estimates, at the cost of scan time.
• b-value — include at least one shell suitable for diffusion-tensor fitting, around b = 1000 s/mm². By default DELTA-SVD fits the tensor on b-values in the range 800–1200 s/mm² (together with the b ≈ 0 volumes) and automatically selects those shell(s) from the data, so multi-shell acquisitions are fine — the appropriate shell is picked for you. A different range can be set with --bRange (see Advanced usage).
• Direction scheme — sample the diffusion directions over the whole sphere rather than a half-sphere (a custom whole-sphere direction set is recommended). Full-sphere sampling gives eddy-current and motion correction (FSL eddy) better-conditioned data to work with.
• Reversed phase encoding — acquire at least one b ≈ 0 volume (or a short series) with reversed phase-encoding direction, so that susceptibility-induced distortions can be corrected (FSL topup).

Preprocessing¶

DELTA-SVD expects preprocessed input, so apply a state-of-the-art pipeline before running it:

• Visual QC of the raw data — inspect for artefacts and gross problems before preprocessing.
• Conversion — convert from DICOM with a tool that preserves the metadata required for preprocessing (phase-encoding direction, total readout time, gradient table), for example dcm2niix with BIDS sidecars. If you intend to continue with MRtrix3, converting with mrconvert to the .mif format is a convenient alternative, as it embeds this metadata directly in the image file.
• Core steps — denoising (e.g. MP-PCA, via MRtrix3's dwidenoise) and Gibbs-ringing removal (MRtrix3's mrdegibbs), followed by susceptibility-distortion correction (FSL topup) and eddy-current/motion correction (FSL eddy).
• Brain mask — produce a DWI-space brain mask; DELTA-SVD takes it as input.