TL;DR
This paper introduces a novel diffusion MRI technique, COMMIT<sub>AxSize</sub>, that can non-invasively estimate axon diameter index at the individual streamline level, offering a more precise and bundle-specific measure than previous voxel-wise methods. This advancement promises to improve our understanding of white matter tract organization and its relation to brain function and disease.
What Problem Does This Paper Address?
The brain's white matter is composed of intricate pathways (tracts or bundles) of axons, which vary in diameter. This diameter is crucial as it directly influences nerve impulse conduction velocity, playing a key role in how different brain regions communicate. Diffusion-weighted Magnetic Resonance Imaging (DW-MRI) is a powerful non-invasive tool for studying these pathways in vivo. However, existing DW-MRI methods for estimating axon diameter provide "voxel-wise" measurements. This means that for any given voxel (a small 3D cube of brain tissue), the estimated axon diameter is an average influenced by all the different axon bundles passing through it. This "partial volume effect" is a significant limitation, especially in areas where multiple tracts cross or fan out, leading to inaccurate and biased estimates for individual bundles. The problem is that we need a way to measure axon diameter for each specific white matter tract, not just an average for a small region, to truly understand their functional properties and how they are organized.
Methods
The core innovation of this paper is the COMMITAxSize approach, which integrates tractography with a microstructure model. Tractography is the standard DW-MRI technique used to reconstruct the likely paths of white matter bundles. COMMIT<sub>AxSize</sub> builds upon the existing Convex Optimization Modeling for Microstructure Informed Tractography (COMMIT) framework.
Here's how it works:
- Microstructure Model: They employ a "Cylinder-Zeppelin-Ball" model. This model represents the DW-MRI signal originating from:
- Cylinders: These represent the axons themselves, with a distribution of diameters. The method specifically models 12 different diameter bins ranging from 1.5 to 7 µm.
- Zeppelins: These represent the extra-axonal space, accounting for its anisotropic diffusion properties.
- Balls: These represent isotropic compartments like cerebrospinal fluid.
- Streamline-Level Estimation: Instead of estimating parameters for each voxel independently, COMMIT<sub>AxSize</sub> assigns contributions to the signal from each reconstructed streamline. The framework aims to express the entire DW-MRI signal in a voxel as a linear combination of signals from all streamlines passing through it, each with its own estimated axon diameter distribution.
- Joint Estimation: The method jointly estimates the microstructure properties (like axon diameter distribution) and the macroscopic organization (streamline trajectories) of the white matter. This is achieved by solving an inverse problem using non-negative least squares, effectively "unmixing" the signal from individual streamlines.
- Bundle-Specific ADI: The output is a bundle-specific Axon Diameter Index (sADI) for each streamline. To get a measure for an entire bundle, they average the sADI of all streamlines belonging to that bundle.
This approach is a significant departure from previous voxel-wise methods, which would average these streamline-level estimates across all voxels in a bundle, inheriting the voxel-wise biases.
Datasets
The study utilized two types of data:
- In vivo Human Data: Acquired from 3 healthy volunteers using a Siemens Connectom 3T MRI system. The acquisition protocol is detailed in Table 1, featuring high b-values and gradient strengths suitable for microstructure imaging.
- Numerical Phantom Data: A synthetic dataset was generated to simulate a crossing fiber configuration with known ground-truth axon diameters. This phantom was used to rigorously test the method's robustness to noise, dictionary sampling, and fiber dispersion.
Key Insights and Ideas
- Streamlines as the Unit of Analysis: The fundamental shift from voxel-wise to streamline-level analysis is the paper's core contribution. This allows for the characterization of individual white matter tracts, overcoming the limitations of partial volume effects in crossing regions.
- Robustness to Noise and Dispersion: The numerical simulations demonstrated that COMMIT<sub>AxSize</sub> provides more consistent and accurate estimates of axon diameter, especially in low signal-to-noise ratio (SNR) conditions and when dealing with complex fiber configurations (dispersion).
- Histological Validation: The method's estimates for the corpus callosum and posterior limb of the internal capsule showed good agreement with existing histological data, providing crucial validation for the in vivo findings.
- Mapping Axon Diameter to Cortical Areas: The study successfully visualized how axon diameter varies across different cortical projections, revealing a hierarchy where motor and sensory areas tend to be associated with larger axon diameters, consistent with known neurophysiology.
Key Results
- Improved Accuracy: In numerical simulations, COMMIT<sub>AxSize</sub> provided more accurate estimates of mean axon diameter compared to the voxel-wise ActiveAxAMICO method, particularly at lower SNR. For example, at SNR=50, COMMIT<sub>AxSize</sub> estimated an average sADI of 2.90 µm and 4.01 µm for the two simulated bundles, closely matching the ground truth of 2.7 µm and 4.0 µm.
- Robustness: The method showed better robustness to noise and fiber dispersion compared to voxel-wise approaches.
- Histological Correlation: The in vivo sADI estimates for the corpus callosum correlated well with human histological measurements, and for the posterior limb of the internal capsule, they correlated with monkey histological data.
- Hierarchical Organization: Visualization of sADI projected onto the pial surface revealed that streamlines connecting to primary motor and somatosensory cortices generally had larger sADI values.
Limitations and Future Work
- Resolution Limit: The authors acknowledge that current DW-MRI technology has a resolution limit for detecting very small axon diameters (around 2.0 µm). However, they note that larger axons still show a detectable contrast.
- Dictionary Optimization: The choice of the number of diameter bins (12 in this study) and fixed diffusivity values are areas for future exploration to potentially improve accuracy.
- Tractography Algorithm Dependence: The accuracy of the streamline-based estimates can be influenced by the tractography algorithm used. Future work will explore different tractography methods.
- Diseased Brain: The authors suggest that while the method may be inaccurate in brains with focal lesions, it could be applicable to developmental disorders, psychiatric disorders, and neurological diseases.
- Dynamic Axon Properties: Future studies could investigate the impact of dynamic axon properties like undulations and microscopic orientation dispersion on the estimates.
- Advanced Models: Further research will focus on optimizing the microstructure forward model and dictionary parameters for even greater sensitivity and accuracy.
Summary updated