Deep learning enables accurate brain tissue microstructure analysis based on clinically feasible diffusion magnetic...

Area of Science:

• Neuroimaging
• Biomedical Engineering
• Computational Neuroscience

Background:

• Diffusion magnetic resonance imaging (dMRI) enables non-invasive brain tissue microstructure assessment.
• Current model-based methods require extensive diffusion gradients, limiting clinical application due to time constraints.
• Deep learning (DL) shows potential for microstructure reconstruction with clinically feasible dMRI.

Purpose of the Study:

• To evaluate the reliability of DL for brain tissue microstructure analysis using clinically feasible dMRI.
• To determine if DL preserves subtle, disease- or age-related microstructural changes.
• To assess the clinical utility of DL-based microstructure reconstruction.

Main Methods:

• Reconstructed tissue microstructure from four brain dMRI datasets using only 12 diffusion gradients.
• Applied neurite orientation dispersion and density imaging (NODDI) and spherical mean technique (SMT) models.
• Validated DL reconstruction against established models on clinically feasible data.

Main Results:

• DL approaches accurately identified disease-related and age-dependent alterations in brain tissue microstructure.
• Reconstruction from limited diffusion gradients (12) using DL yielded reliable microstructural metrics.
• DL-based analysis demonstrated consistency across different dMRI datasets.

Conclusions:

• DL-based tissue microstructure reconstruction is reliable for clinically feasible dMRI.
• DL methods can accurately quantify subtle microstructural changes associated with disease and aging.
• This approach enhances the clinical feasibility of advanced brain tissue analysis.