Comparative Systematic Analysis of Gray Matter Biophysical Models on a Public Dataset

Area of Science:

• Neuroimaging
• Diffusion MRI (dMRI)
• Biophysical modeling

Background:

• Gray matter (GM) microstructure characterization is crucial in neuroimaging.
• Advanced diffusion MRI (dMRI) models like NEXI, SMEX, SANDI, and SANDIX aim to capture complex microstructural features.
• These models account for soma contributions and inter-compartment exchange in the dMRI signal.

Purpose of the Study:

• To provide a comparative evaluation of four leading GM diffusion models: NEXI, SMEX, SANDI, and SANDIX.
• To assess their performance in estimating cortical microstructure metrics.
• To understand the trade-offs between biological specificity, model complexity, and fitting robustness.

Main Methods:

• Utilized the Connectome Diffusion Microstructure Dataset (CDMD), a public in vivo human dataset with two diffusion times.
• Estimated cortical microstructure metrics in 26 healthy subjects using the Gray Matter Swiss Knife (GMSK) toolbox.
• Evaluated goodness of fit, anatomical patterns, and consistency with prior research.

Main Results:

• All four models produced GM parameter estimates consistent with previous studies.
• NEXI and SMEX demonstrated similar cortical anatomical patterns and regional distributions across diffusion times.
• NEXI exhibited the best goodness of fit, followed by SMEX, SANDIX, and SANDI, with SANDI showing high dependence on diffusion time and fitting algorithm.

Conclusions:

• Estimating exchange models from two diffusion times is feasible.
• Model selection involves balancing biological specificity, complexity, and fitting robustness for clinical and research applications.
• This comparative analysis informs the choice of dMRI models for GM microstructure research.