Diffusion MRI microstructure models with in vivo human brain Connectome data: results from a multi-group comparison.

Area of Science:

• Biomedical Imaging
• Neuroimaging
• Diffusion MRI

Background:

• Numerous mathematical models exist for diffusion-weighted (DW) magnetic resonance imaging (MRI) signals.
• Existing model comparisons are limited to specific subclasses, lacking a comprehensive performance evaluation across diverse model types.

Purpose of the Study:

• To quantitatively compare various diffusion-weighted MRI models using in vivo human brain data.
• To assess model performance in predicting DW signals for unseen diffusion gradients and b-values.

Main Methods:

• Organized the 'White Matter Modeling Challenge' at ISBI 2015.
• Utilized data from the Connectome scanner with high gradient strengths and diverse diffusion times.
• Focused on signal prediction accuracy in the genu of the corpus callosum and the fornix.

Main Results:

• Tissue models, on average, outperformed signal models in predicting unseen DW MRI data.
• Non-Gaussian noise models showed minimal improvement in prediction accuracy compared to Gaussian models.
• Data preprocessing (outlier removal) and signal-predicting strategies (bootstrapping, cross-validation) enhanced model fitting.

Conclusions:

• Tissue models are a promising direction for next-generation quantitative MRI techniques.
• Signal outlier removal and cross-validation are beneficial for robust diffusion MRI model fitting.
• The challenge established a benchmark for diffusion MRI model comparison, with data available for future research.