Spatial parcellations, spectral filtering, and connectivity measures in fMRI: Optimizing for discrimination

Area of Science:

• Neuroimaging and Cognitive Neuroscience
• Brain Connectivity Analysis

Background:

• Functional Connectivity (FC) analysis is crucial in fMRI for distinguishing brain states.
• Lack of comparative assessments hinders the selection of optimal FC calculation methods and result comparability.

Purpose of the Study:

• To assess the impact of methodological choices on the discriminability of brain states using fMRI.
• To provide a basis for consistent selection of approaches for estimating and analyzing FC.

Main Methods:

• Utilized a controlled dataset of continuous active states (visual and motor tasks) with localized FC changes.
• Tested various anatomical and functional parcellations (AAL, HCP, ICA) and dependency measures (amplitude, covariance, correlation, regularized partial correlation).
• Employed multivariate pattern analysis (MVPA) to evaluate feature discriminability under different temporal filtering conditions.

Main Results:

• Multidimensional functional parcellations outperformed anatomical atlases in discriminating states.
• Partial correlation, particularly with regularization, showed superior performance over simple correlation.
• Higher frequencies demonstrated significant discriminability, especially with ICA-based parcellations and specific dependency measures.

Conclusions:

• Methodological choices in FC analysis profoundly influence fMRI results.
• Functional parcellations and partial correlation offer optimized accuracy and interpretability.
• Frequency-specific analysis, particularly at higher frequencies, can enhance discriminability.