HAITCH: Motion Correction

HAITCH is a robust framework for distortion and motion correction in fetal multi-shell diffusion-weighted MRI. It is an integral part of the FEDI toolbox, enabling high-quality preprocessing of challenging fetal dMRI datasets, and facilitating accurate downstream analyses such as tractography and microstructural modeling.

For full methodological and validation details, refer to the following publication:

Snoussi et al., 2025 – HAITCH: A framework for distortion and motion correction in fetal multi-shell diffusion-weighted MRI

HAITCH Pipeline Steps

The HAITCH pipeline consists of 10 main processing steps that progressively correct for noise, artifacts, distortion, and motion in fetal dMRI data:

  1. DWI Denoising (GSVS): Removes noise using the Generalized Spatially Varying Smoothing (GSVS) estimator

  2. Gibbs Ringing Artifact Removal: Eliminates Gibbs ringing artifacts using the method by Kellner et al. (2016)

  3. Rician Bias Correction: Corrects Rician noise bias, particularly important for low SNR fetal data

  4. Fetal Brain Extraction: Segments the fetal brain from surrounding maternal tissue using deep learning methods

  5. Split, Crop, and Skull Stripping: Processes the segmented brain data and creates masks

  6. Slice-wise Distortion Correction: Corrects for slice-specific distortions

  7. B1 Field Bias Correction: Corrects for B1 field inhomogeneities

  8. Motion Correction: Iterative process using 3D-SHORE modeling. This step includes: - Outlier detection (slice-wise and voxel-wise weighting) - SHORE-based signal prediction - Volume-to-volume registration - B-vector rotation - Iterative refinement (typically 6 epochs)

  9. Registration to T2W Atlas: Registers the motion-corrected dMRI data to T2-weighted anatomical space and fetal brain atlas

  10. Tensor/FOD Estimation and Tractography: Estimates diffusion tensors or fiber orientation distributions (FODs) and performs tractography

Accessing the HAITCH Pipeline Code

The HAITCH pipeline is available in two versions:

HAITCH 1.0 (Original Version)

  • Location: FEDI/pipelines/HAITCH1.0/

  • Uses original Python scripts from the src/ directory

  • Manual iteration loop for motion correction

  • Fully tested and stable

HAITCH 2.0 (Updated Version)

  • Location: FEDI/pipelines/HAITCH2.0/

  • Uses the new fedi_* command-line tools

  • Simplified STEP 8 using fedi_dmri_moco (single command instead of manual loop)

  • Cleaner implementation with better maintainability

Both versions produce identical results and can be used interchangeably. The main pipeline script is dMRI_HAITCH.sh, which requires a configuration file as input.

For detailed usage instructions and configuration options, see the README.md files in each pipeline directory.

Sampling Schemes

The diffusion-weighted MRI sampling schemes used in the HAITCH validation study are available in the FEDI/sampling_scheme/ directory. These include:

  • HAITCH_scheme_dual_echo_sequence.dvs - Dual-echo sequence sampling scheme

  • HAITCH_scheme_siemens_product_sequence.dvs - Siemens product sequence sampling scheme

Results and Visualizations

The following figures illustrate the effectiveness of HAITCH in real fetal diffusion MRI datasets.

../_images/Figure5.png

Figure 1. Two examples fetal dMRI scans before and after motion correction.

../_images/Figure6.png

Figure 2. Estimated motion parameters and slice weights for Subject B. Peaks in motion correlate with low slice weights.

../_images/Figure9.png

Figure 3. Impact of HAITCH on the quality of NODDI parameter maps.