Data-driven classification of tissue water populations by massively multidimensional diffusion-relaxation correlation...

Area of Science:

• Biomedical Imaging
• Neuroimaging
• Biophysics

Background:

• Massively multidimensional diffusion-relaxation correlation MRI analyzes water populations for detailed tissue microstructure information.
• Current methods manually bin diffusion-relaxation data, oversimplifying complex tissue fractions and underutilizing parameters.

Purpose of the Study:

• To implement an unsupervised clustering approach for automatic classification of white matter (WM), gray matter (GM), and free water (FW).
• To explore additional water populations using the full D(ω)-R1-R2 distributions.
• To validate the approach on ex vivo and in vivo rat brain and in vivo human brain data.

Main Methods:

• Applied unsupervised clustering to D(ω)-R1-R2 distributions from diffusion-relaxation MRI data.
• Utilized data from ex vivo and in vivo rat brains and in vivo human brains.
• Compared clustering results with histological myelin and Nissl stainings.

Main Results:

• Unsupervised clustering successfully separated WM, GM, and FW across different protocols and species.
• Identified distinct water populations, including one localized in high cell density regions (dentate gyrus, cerebellum) with high frequency-dependent diffusion.
• Demonstrated that clustering reveals tissue complexity beyond traditional segmentation without parameter assumptions.

Conclusions:

• Unsupervised clustering of diffusion-relaxation MRI data offers a powerful, assumption-free method for analyzing tissue microstructure.
• This approach can be extended to other body parts like the prostate and breast for cancer research.
• Characterizing clusters by diffusion and relaxation properties enhances understanding of subtle pathological changes in cellular fractions.