Subject-Specific Structural Parcellations Based on Randomized AB-divergences.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Image Analysis

Background:

• Brain parcellation is crucial for dimensionality reduction in connectome creation.
• Standard methods involve registering individual scans to a template, which can introduce registration errors affecting biomarker quality.
• Diffusion MRI (dMRI) provides structural connectivity information vital for brain mapping.

Purpose of the Study:

• To mitigate registration errors in brain parcellation and connectome construction.
• To develop a hybrid approach that refines anatomical priors using dMRI-derived structural connectivity.
• To generate coherent structural parcels in native subject space with cross-population interpretability.

Main Methods:

• Registering a population-wide anatomical prior to individual dMRI scans.
• Generating voxel-wise connectivity signatures from dMRI data.
• Re-parcellating the brain based on connectivity signature similarity using AB-divergences and graph-based methods, while constraining parcel number.

Main Results:

• The proposed hybrid method generates highly coherent structural parcels in native subject space.
• The approach maintains interpretability and correspondences across the population.
• Results indicate the production of more coherent parcels and stronger connectomes compared to original anatomical priors.

Conclusions:

• The hybrid brain parcellation method effectively refines anatomical priors using dMRI connectivity.
• This approach offers a robust solution to registration errors in connectome analysis.
• The method enhances the quality and coherence of brain parcels and resulting connectomes.