Diffusion MRI (dMRI)🔗

---

Diffusion-Weighted Imaging (DWI) measures the diffusion of water molecules in tissue and is used to model white matter microstructure and structural connectivity in the central nervous system. Full protocols, sequence installation, and operation instructions are available via HBCD Study MRI Protocols. In brief, the DWI protocol acquires roughly 140 diffusion-weighted echo planar images at four b-values (diffusion-weighting) between 0 and 3000 s/mm2 (12-13 minutes total acquisition time). For raw image acquisition, a minimum of 60% of the diffusion-weighted volumes are required to be collected for the acquisition to be deemed successful.

Diffusion Pulse Sequence Timings🔗

Scanner model	Gmax (mT/m)	δ (ms)	Δ (ms)
Philips Achieva	54.3	19.81	42.82
Philips MR7700	59.7	17.47	44.36
Philips R2D2	59.8	17.5	44.45
Siemens Prisma	73.7	14.2	42.97
Siemens Prisma Fit	73.7	14.2	42.97
GE MR750	49.6	21.8	43.1

Processing🔗

Diffusion MRI data are preprocessed using QSIPrep, which performs motion and distortion correction, MP-PCA denoising, co-registration to T1w images, spatial normalization (ANTs), and tissue segmentation (Cieslak et al. 2021, Cieslak et al. 2025). Preprocessed QSIPrep outputs are passed to QSIRecon, which runs curated reconstruction workflows, including ODF/FOD reconstruction, tractography, Fixel estimation, and regional connectivity. The diffusion encoding provided via the multiple QSIRecon derivative folders enable the estimation of multiple diffusion MRI models to create the derived data, outlined below. Full details of HBCD diffusion MRI processing and methods are described in Cieslak et al. 2025.

Pipeline	Folder	Description
QSIPrep	qsiprep/	Preprocessed diffusion data, transforms, QC metrics & reports
QSIRecon	qsirecon/	QSIRecon workflow logs and configuration files
QSIRecon-DSIStudio	qsirecon-DSIStudio/	DSI Studio DTI reconstruction & tractography
QSIRecon-DIPYDKI	qsirecon-DIPYDKI/	DIPY Diffusion kurtosis (DKI) and tensor-derived maps
QSIRecon-TORTOISE	qsirecon-TORTOISE_model-MAPMRI/	TORTOISE MAP-MRI and scalar maps
	qsirecon-TORTOISE_model-tensor/	TORTOISE Tensor fits and scalar maps

Derivatives🔗

QSIPrep🔗

hbcd/
└── derivatives/
    └── qsiprep/
        └── sub-[ID]/
            ├── log/
            └── ses-[V0X]/
                ├── anat/
                │   # Transforms
                │   ├── *_from-{ACPC_to-anat|anat_to-ACPC}_mode-image_xfm.mat
                │   ├── *_from-{ACPC_to-MNIInfant+1|MNIInfant+1_to-ACPC}_mode-image_xfm.h5
                │   ├── *_from-orig_to-anat_mode-image_xfm.txt
                │
                │   # Structural outputs (ACPC space)
                │   ├── *_space-ACPC_desc-preproc_T2w.nii.gz (+JSON)
                │   ├── *_space-ACPC_desc-{aseg_dseg|brain_mask}.nii.gz
                │   └── *_space-ACPC_dseg.nii.gz
                │
                ├── dwi/
                │   # QC & confounds
                │   ├── *_desc-confounds_timeseries.tsv
                │   ├── *_desc-{image|pepolar}_qc.tsv
                │   ├── *_space-ACPC_desc-slice_qc.json
                │
                │   # Preprocessed data
                │   ├── *_space-ACPC_desc-preproc_dwi.nii.gz (+JSON)
                │   ├── *_space-ACPC_desc-preproc_dwi.{bval|bvec|b|b_table.txt}
                │   ├── *_space-ACPC_dwiref.nii.gz
                │
                │   # Masks & maps
                │   ├── *_space-ACPC_desc-brain_mask.nii.gz
                │   └── *_space-ACPC_model-eddy_stat-cnr_dwimap.nii.gz (+JSON)
                │
                ├── figures/
                └── sub-[ID]_ses-[V0X].html
 How To Read File Trees →

QSIRecon Details🔗

Diffusion Tensor Imaging (DTI)🔗

DSI Studio models diffusion with a 3D Gaussian distribution of water displacements. Key outputs include fractional anisotropy (FA), i.e. anisotropic diffusion (typically higher in white matter bundles with dense, parallel fibers) and mean diffusivity (MD), i.e. directionally averaged apparent diffusion coefficient (inversely related to cellular membrane density) (Basser 1994).

hbcd/
└── derivatives/
    ├── qsirecon-DSIStudio/
    │   └── sub-[ID]/
    │       └── ses-[V0X]/
    │           ├── dwi/
    │           │   ├── *_space-ACPC_bundles-DSIStudio_{scalar|tdi}stats.tsv
    │           │   ├── *_space-ACPC_model-gqi_bundle-{BUNDLE}_streamlines.tck.gz
    │           │   ├── *_space-ACPC_model-gqi_bundlestats.csv
    │           │   ├── *_space-ACPC_model-gqi_dwimap.fib.gz
    │           │   ├── *_space-ACPC_model-gqi_dwimap.fib.gz.icbm152_adult.map.gz
    │           │   ├── *_space-ACPC_model-gqi_param-{gfa|iso|qa}_dwimap.nii.gz
    │           │   ├── *_space-{ACPC|MNIInfant+1}_model-rdi_param-{rd1|rd2}_dwimap.nii.gz
    │           │   └── *_space-{ACPC|MNIInfant+1}_model-tensor_param-{DTI-PARAM}_dwimap.nii.gz
    │           ├── figures/*
    │           └── sub-[ID]_ses-[V0X].html

DTI-PARAM: ad, fa, ha, md, rd, txx, txy, txz, tyy, tyz, tzz
BUNDLE: see full list

Diffusion Kurtosis Imaging (DKI)🔗

DKI extends DTI to capture non-Gaussian diffusion. The main metric is mean kurtosis (MK), which is more sensitive to complex or restricted diffusion and often higher in dense white matter (Jensen 2005).

    ├── qsirecon-DIPYDKI/
    │   └── sub-[ID]/
    │       └── ses-[V0X]/
    │           ├── dwi/
    │           │   # DIPY DKI
    │           │   ├── *_space-ACPC_bundles-DSIStudio_scalarstats.tsv
    │           │   ├── *_space-{ACPC|MNIInfant+1}_model-dki_param-{DKI-PARAM}_dwimap.nii.gz
    │           │   └── *_space-{ACPC|MNIInfant+1}_model-tensor_param-fa_dwimap.nii.gz
    │           ├── figures/*
    │           └── sub-[ID]_ses-[V0X].html

DKI-PARAM: ad, ak, kfa, md, mk, mkt, rd, rk

Mean Apparent Propagator MRI (MAP-MRI)🔗

MAP-MRI Extends DTI by estimating the full spatial probability distribution (propagator) of water diffusion without assuming Gaussian distribution. This enables quantification of non-Gaussian diffusion and more accurate measures of directionality and anisotropy (Özarslan 2013).

    ├── qsirecon-TORTOISE_model-MAPMRI/
    │   └── sub-[ID]/
    │       └── ses-[V0X]/
    │           ├── dwi/
    │           │   ├── *_space-ACPC_bundles-DSIStudio_scalarstats.tsv
    │           │   ├── *_space-{ACPC|MNIInfant+1}_model-mapmri_param-{MAPMRI}_dwimap.nii.gz
    │           │   └── *_space-{ACPC|MNIInfant+1}_model-tensor_param-{TENSOR}_dwimap.nii.gz
    │           ├── figures/*
    │           └── sub-[ID]_ses-[V0X].html

MAPMRI: ng, ngpar, ngperp, pa, path, rtap, rtop, rtpp

Metric	Description
Propagator Anisotropy (PA)	Quantifies anisotropy by computing the dissimilarity of the full MAP-MRI propagator from its fully isotropic counterpart. More accurate than FA.
Non-Gaussianity (NG)	Quantifies deviation from Gaussian diffusion. NG measures overall deviation, NGpar along the primary diffusion axis (fiber direction in white matter), and NGperp perpendicular to it (often related to restriction).
Return To Origin Probability (RTOP)	Probability that a water molecule returns to its starting point. Low in unrestricted diffusion (large cells), high in restricted diffusion (small or impermeable cells). Inversely related to pore volume.
Return To Axis Probability (RTAP)	Probability that a water molecule returns to the principal diffusion axis (primary eigenvector).
Return To Plane Probability (RTPP)	Reciprocal of mean cylinder length and inversely proportional to axial diffusivity; Related to diffusion taking place within coherently oriented cylinders.

QSIRecon-TORTOISE Tensor🔗

    └── qsirecon-TORTOISE_model-tensor/
        └── sub-[ID]/
            └── ses-[V0X]/
                └── dwi/
                    ├── *_space-ACPC_bundles-DSIStudio_scalarstats.tsv
                    └── *_space-MNIInfant+1_model-tensor_param-{TENSOR}_dwimap.nii.g

TENSOR: ad, am, fa, li, rd

---

Quality Control Summary Statistics🔗

Automated QC for processed diffusion data is fairly robust, with metrics provided in sub-[ID]_ses-[V0X]_space-ACPC_desc-image_qc.tsv within the QSIPrep derivatives (see QSIPrep documentation for details). Below are distributions of automated QC metrics from HBCD visits V02 and V03. Higher Neighboring DWI Correlation (NDC; closer to 1) and Contrast-to-Noise Ratio (CNR) indicate better image quality. NDC can also be used as a covariate in analyses to account for QC variation.

Left: NDC calculated pre- and post-processing for each vendor using combined AP/PA scans Right: Shell-wise CNR calculated by Eddy. We do not provide exclusion threshold recommendations because all data passed preliminary QC. However, NDC and CNR are useful covariates when analyzing other derivatives.

---