Estimation of diffusion properties in three-way fiber crossings without overfitting

Area of Science:

• Neuroimaging
• Biophysics
• Computational Neuroscience

Background:

• Diffusion-weighted magnetic resonance imaging (dMRI) is crucial for assessing white matter structural integrity.
• Accurate quantification of diffusion properties is essential, especially in complex fiber geometries like crossings.
• Existing methods may struggle with unbiased estimation in intricate neural pathways.

Purpose of the Study:

• To develop and validate a novel rank-2 tensor model selection framework for precise diffusion property estimation.
• To accurately quantify diffusion in simple and complex white matter fiber geometries, including three-way crossings.
• To improve the reliability and sensitivity of diffusion tensor imaging (DTI) analysis.

Main Methods:

• Utilized a maximum a-posteriori (MAP) estimator with a constrained triple tensor model.
• Implemented a prior to maximize divergence between tensor principal orientations, preventing model degeneracy.
• Introduced an information complexity measure (ICOMP-TKLD) for robust model selection (single, dual, or triple tensor).

Main Results:

• The MAP estimator with the proposed prior significantly reduced parameter spread and enhanced precision.
• Accurate estimation of fractional anisotropy (FA) was achieved down to ~40° fiber angles.
• Reliable estimation of volume fractions (0.2-0.8) and accurate inference of neuro-anatomy in complex crossings were demonstrated.
• The ICOMP-TKLD model selection favored single-fiber configurations in simple regions, unlike FSL's ball-and-stick approach.

Conclusions:

• The MAP estimator with prior improves parameter estimation precision in dMRI without bias.
• The ICOMP-TKLD model selection effectively balances goodness-of-fit and information complexity, preventing overfitting.
• This framework enhances the sensitivity of statistical analyses in diffusion tensor MRI, particularly for complex white matter structures.