Skip to content

Latest commit

 

History

History
201 lines (163 loc) · 8.82 KB

dwi.md

File metadata and controls

201 lines (163 loc) · 8.82 KB

Diffusion Weighted Imaging (DWI)

Data

The Diffusion Magnetic Resonance Imaging (dMRI) data, which describes the movement of water within the brain, is a specialized 4D T2*-contrasted magnetic resonance sequence. This type of imaging is critical because models fitted to the dMRI data can summarize the orientation of water movement within the tissue, reflecting aspects like water moving parallel to (along) axons.

When collecting dMRI data, the sequence iterates along orientation rather than incrementing time along the 4th dimension.

Structure of dMRI Data and Associated Files

A dMRI scan consists of two kinds of volumes collected in the 4D sequence:

  1. Unweighted volumes ($\text{b}0$): These are images collected with no diffusion weighting applied.
  2. Weighted gradient volumes: These volumes have diffusion gradients applied, which makes them sensitive to the movement of water molecules along specific directions.

To correctly interpret the weighted gradient volumes, additional information describing the diffusion gradient applied during the acquisition must be stored separately. These supplementary files typically contain information about the magnitude and orientation of the gradients used.

Specifically, the required information is stored in two accompanying files, often adhering to conventions such as those used by analysis software like FSL:

  • Magnitude of the gradient applied ($\text{bvals}$ file): This file contains the magnitude of the diffusion weighting applied for each volume in the 4D scan. The magnitude ($\text{b}$-value) determines the strength of the diffusion weighting, and multiple different magnitudes (shells) can be collected in a modern sequence (e.g., multi-shell diffusion imaging).
  • Orientation of the gradient applied ($\text{bvecs}$ file): This file contains the orientation vector of the diffusion gradient applied for each volume. The orientation of this file relative to the image data file is critical for proper use of the scan.

The ability to fit increasingly complex models to the data depends on the number of shells ($\text{bvals}$) and directions ($\text{bvecs}$) collected in the sequence. These models represent properties of the underlying tissue by summarizing the difference in orientation from the baseline ($\text{b}0$) image.

Diffusion MRI in the PREVENT-AD Cohort

In the PREVENT-AD study, Diffusion Weighted Imaging (DWI) data were collected across both Phase 1 (Stage 1) and Phase 2 (Stage 2).

Stage 1/Phase 1 Acquisition (2011–2017):

  • The DWI sequence used in Phase 1 was a 2D EPI transversal sequence.
  • The acquisition parameters involved $\text{b}$-values of $ \text{ s}/\text{mm}^2$ with 1 ($\text{b}0$ image) and 64 directions. This indicates a single-shell protocol (excluding the $\text{b}0$ acquisition).

Stage 2/Phase 2 Transition (Post-2017):

  • When the MRI protocol was updated in Phase 2 (resuming 2019), the diffusion MRI acquisition changed from single shell to multi-shell.
  • The final raw data acquired are available in NIfTI file format.

Analytic Measures (Data Derivatives): The PREVENT-AD repository also shares analytical measures derived from the dMRI data, specifically focusing on Diffusion Tensor Imaging (DTI) measures:

  • Diffusion properties of white matter tracts are quantified using pipelines such as TractoFlow, RBXFlow, and Tractometry Flow.
  • Examples of DTI measures shared include:
    • Fractional Anisotropy (FA)
    • Mean Diffusivity (MD)
    • Axial Diffusivity (AD)
    • Radial Diffusivity (RD)

Furthermore, advanced analyses leveraging multi-shell data, such as Neurite Orientation Dispersion and Density Imaging (NODDI), can yield microstructure estimates like Neurite Density Index (NDI), Orientation Dispersion Index (ODI), and Isotropic Volume Fraction (ISOVF). These measures can be mapped to connectome edges using tractography pipelines.

Processing

The processing of Diffusion Weighted Imaging (DWI) data, which is essential for deriving metrics about water movement and microstructure within the white matter, involves several common and necessary steps, primarily focused on artifact correction, normalization, and modeling. These steps are often nearly identical across different standardized pipelines (such as TractoFlow, QSIPrep, or MRTrix3 Preprocessing).

Here are the common steps necessary for processing DWI data, followed by modeling and analysis steps:

I. Preprocessing and Quality Control

The initial stages focus on cleaning and correcting the raw 4D T2*-contrasted sequence data for noise and artifacts.

  1. Check Diffusion Gradient Orientation: It is crucial to check the orientation of the gradient vectors ($\text{bvecs}$) in conjunction with the diffusion data itself. The raw DWI data consists of unweighted volumes ($\text{b}0$) and weighted gradient volumes, and the associated files ($\text{bvals}$ and $\text{bvecs}$) containing the magnitude and orientation of the diffusion gradients are necessary for using the scan.

  2. Denoising (Remove Scanner Noise): This step involves removing unwanted scanner noise from the data. For example, the MRTrix3 pipeline includes a specific denoising command (dwidenoise) using MP-Distribution.

  3. Gibbs Ringing Artifact Removal: Gibbs ringing artifacts, which appear as characteristic ripples or ghosts in the image, must be removed.

  4. Top-up, Eddy Current, and Motion Correction: This step corrects for artifacts caused by susceptibility distortion, eddy currents (induced by rapidly switching magnetic gradients), and bulk head motion that occur during acquisition. This process often requires the input of two diffusion volumes with opposing phase-encode directions (e.g., AP/PA or RL/LR acquisitions, often referred to as "top-up").

  5. Bias Correction: The background bias field, which causes non-uniform intensity across the image, needs to be normalized or corrected.

  6. Rician Denoising: This step filters Rician noise, often using commands like mrcalc.

  7. Intensity Normalization (Optional): Image intensity can be optionally normalized.

  8. Brain Extraction and Masking: The brain tissue must be isolated from non-brain tissue (like skull and skin) to create a clean mask for subsequent processing.

  9. Registration and Reslicing: The preprocessed DWI data must be registered (aligned) to a standard anatomical space, such as AC-PC space. The data may also be optionally resliced to a new resolution.

II. Modeling and Derivation

Once preprocessed, the cleaned DWI data is used to fit models that summarize water movement and tissue properties.

  1. Estimate the Diffusion Tensor Model: The diffusion tensor model (DTI) is estimated by summarizing the differences in orientation from the baseline unweighted volume ($\text{b}0$). This model can be complex depending on the number of shells ($\text{bvals}$) and directions ($\text{bvecs}$) collected. This step yields measures like Fractional Anisotropy (FA), Mean Diffusivity (MD), Axial Diffusivity (AD), and Radial Diffusivity (RD).

  2. Estimate Advanced Microstructure Models: For multi-shell data (like that collected in PREVENT-AD Phase 2), more complex microstructure models, such as Neurite Orientation Dispersion and Density Imaging (NODDI), can be estimated. NODDI provides estimates like Neurite Density Index (NDI), Orientation Dispersion Index (ODI), and Isotropic Volume Fraction (ISOVF).

  3. Estimate Fiber Orientation Distribution (fODF) using CSD: Advanced analysis often involves estimating the Fiber Orientation Distribution (fODF) using methods like Constrained Spherical Deconvolution (CSD), which estimates the response function of the tissue to derive directional information.

  4. Tractography and Structural Connectome Estimation: Tractography algorithms use the orientation summaries (like the diffusion tensor or CSD fODFs) to follow a likely path of connectivity through the brain, reconstructing white matter bundles or streamlines. This can involve anatomically constrained techniques, such as Reproducible Anatomically Constrained Ensemble Tractography (RACE-Trac), which can involve steps like creating five tissue-type partial volume maps and estimating CSD models.

  5. Validation and Mapping (Optional/Advanced):

    • Linear Fascicle Evaluation (LiFE): This technique uses the tractography estimates to predict the diffusion signal and assigns weights to the streamlines based on how well they match the underlying raw diffusion data.
    • Mapping Properties to Edges: Once streamlines (tracts) are defined, other properties derived from the DWI data, such as FA, MD, or NODDI metrics (NDI, ODI, ISOVF), can be mapped onto these structural network edges to quantify microstructural integrity along specific white matter bundles.