fMRI-based data-driven brain parcellation using independent component analysis

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Brain Mapping

Background:

• Functional magnetic resonance imaging (fMRI) studies necessitate robust methods for parcellating the brain into regions of interest (ROIs).
• Current parcellation approaches rely on standardized anatomical atlases (e.g., Montreal Neurological Institute - MNI) or individual functional activity patterns (e.g., Personode software).

Purpose of the Study:

• To introduce and evaluate the Independent Component Analysis (ICA)-based Parcellation Algorithm (IPA) for creating individualized and group-level brain parcellations.
• To assess the spatial consistency and functional homogeneity of ROIs generated by the IPA in a hypertension study cohort.

Main Methods:

• The IPA algorithm utilizes independent components (ICs) derived from group ICA (gICA) to construct ROIs.
• Individualized parcellations were generated by regressing ICs across all subjects, alongside a gICA-derived parcellation.
• Spatial consistency was quantified using Dice Similarity Coefficients (DSCs), and functional homogeneity was assessed via mean Pearson correlation.

Main Results:

• Individualized parcellations generated by IPA demonstrated a mean DSC of 0.69 ± 0.14, indicating good spatial consistency.
• Functional homogeneity for individualized IPA parcellations averaged 0.30 ± 0.14, while gICA-derived parcellations showed 0.38 ± 0.15.
• Comparison with Personode showed IPA's individualized parcellations had higher DSC (0.69 vs. 0.43) and homogeneity (0.30 vs. 0.28).

Conclusions:

• The IPA method provides more reliable ROI definition and superior functional homogeneity compared to existing techniques like Personode and the MNI atlas.
• The IPA demonstrates significant promise as an advanced parcellation technique for enhancing fMRI data analysis.
• IPA-generated parcellations offer improved spatial consistency and functional homogeneity, crucial for accurate interpretation of fMRI findings.