TL;DR
This paper introduces Bingham-NODDI, a novel diffusion MRI technique that extends the existing NODDI framework to accurately map the anisotropic orientation dispersion of neurites (axons and dendrites) in the brain. This advancement allows for a more detailed understanding of brain microstructure, potentially serving as a sensitive marker for normal development, aging, and neurological disorders.
What Problem Does This Paper Address?
Understanding the intricate structure of neurons, specifically the orientation and dispersion of their projections (neurites), is crucial for comprehending brain function, development, and disease. While diffusion MRI has become a powerful non-invasive tool for probing brain microstructure, existing methods like NODDI (Neurite Orientation Dispersion and Density Imaging) primarily model isotropic orientation dispersion. However, neurites in the brain often exhibit anisotropic dispersion – meaning they spread out more in certain directions than others, a phenomenon observed in histological studies due to fanning and bending of fibers. This anisotropic dispersion is widespread and can reflect subtle microstructural changes that isotropic models might miss. Therefore, there's a need for a diffusion MRI technique that can accurately quantify this anisotropic orientation dispersion in vivo.
Methods
The core innovation of this paper is the extension of the NODDI model by replacing the Watson distribution (used for isotropic dispersion) with the Bingham distribution. The Bingham distribution is a more flexible model that can capture anisotropic orientation dispersion.
Here's a breakdown of the approach:
- Bingham Distribution: This is a spherical analogue of a 2D Gaussian distribution, allowing for different degrees of dispersion along different axes. It introduces two new parameters (κ and β) that quantify the extent and anisotropy of neurite orientation dispersion.
- Bingham-NODDI Model: The authors integrate the Bingham distribution into the existing NODDI framework. This involves re-parameterizing the intra-neurite signal model to account for the anisotropic orientation distribution.
- New Indices: Based on the Bingham distribution parameters, they derive new indices to quantify:
- Overall Dispersion Index (ODI_Tot): Measures the total spread of neurite orientations.
- Dispersion Anisotropy Index (DAB): Specifically quantifies the degree of anisotropy in neurite orientation dispersion.
- Primary and Secondary Dispersion Indices (ODI_P, ODI_S): Measure dispersion along specific axes related to the dominant orientation.
- Model Fitting: The authors adapt the existing NODDI toolbox to incorporate the Bingham distribution and use a two-stage fitting procedure (grid search followed by Gauss-Newton optimization) to estimate the model parameters.
Datasets
The study utilized two types of datasets:
- Synthetic Data: Diffusion MRI signals were synthesized using a multi-compartment model that incorporated the Bingham distribution. This allowed for controlled evaluation of parameter estimation accuracy and precision under ideal conditions, with varying neurite densities, axon radii, and orientation distributions.
- In Vivo Data: Diffusion MRI data from one healthy volunteer acquired using a rich four-shell HARDI (High Angular Resolution Diffusion Imaging) protocol was used for in vivo validation. This data was previously used in other NODDI studies.
Key Insights and Ideas
- Anisotropic Dispersion is Widespread: The study confirms that anisotropic orientation dispersion is a common feature in the brain, particularly in white matter tracts like the corpus callosum and corona radiata.
- Bingham-NODDI Captures Anisotropy: The Bingham distribution provides a significantly better fit to data exhibiting anisotropic neurite dispersion compared to the Watson distribution used in standard NODDI.
- Separating Dispersion and Anisotropy: The new indices (ODI_Tot and DAB) allow researchers to disentangle the overall level of neurite dispersion from its anisotropic nature, offering a more nuanced view of microstructural changes.
- Clinical Feasibility: The proposed Bingham-NODDI model can be estimated using clinically feasible, multi-shell diffusion MRI protocols, making it suitable for translation to clinical settings.
- Robustness of Standard NODDI: For applications where anisotropic dispersion is not of interest, the original Watson-NODDI model remains robust and provides accurate estimates of other microstructural parameters.
Key Results
- Accurate Parameter Estimation: Synthetic data experiments showed that Bingham-NODDI can accurately and precisely estimate its parameters, with estimation errors generally low and dependent on neurite density (Vin) and the underlying orientation distribution.
- In Vivo Feasibility: Parameter maps generated from in vivo data showed sensible distributions of neurite morphology, with higher dispersion and anisotropy observed in expected white matter regions.
- Model Comparison (BIC): Bayesian Information Criterion (BIC) analysis revealed that Bingham-NODDI provides a better fit to the data than Watson-NODDI in a significant proportion of voxels, particularly in white matter regions with expected fanning/bending configurations.
- Protocol Efficiency: A two-shell HARDI protocol (N1) was found to be sufficient for estimating Bingham-NODDI parameters, similar to Watson-NODDI, highlighting its clinical practicality. However, estimating anisotropy-specific parameters (DAB, μ₂) was found to be more challenging and less robust to reduced sampling.
Limitations and Future Work
- Crossing Fibers: Bingham-NODDI, like standard NODDI, does not explicitly model crossing fibers. While it can capture regions with high dispersion and anisotropy, future work aims to integrate multi-fiber models with the Bingham distribution for a more comprehensive characterization.
- Parameter Estimation Challenges: Estimating anisotropy-specific parameters (DAB, μ₂) is more challenging and less robust to reduced sampling compared to other parameters. Increasing angular resolution in acquisition protocols could improve this.
- Fixed Diffusivities: The model assumes fixed diffusivities for intra-neurite and CSF compartments, which could introduce bias. Relaxing these assumptions with more efficient imaging protocols is a future direction.
- Fitting Procedure: The current fitting procedure can sometimes get stuck in local minima, particularly in regions with partial volume effects. Optimizing the fitting procedure, possibly by using Watson-NODDI fits as starting points, is suggested.
- Clinical Validation: The authors emphasize the need for clinical studies to demonstrate the utility of Bingham-NODDI metrics in diagnosing and prognosing neurological diseases.
- Histological Validation: Further validation using histological data is planned to confirm the accuracy of the estimated parameters in describing underlying microstructure.
Summary updated