Comparison of statistical tests for group differences in brain functional networks.

Area of Science:

• Neuroimaging
• Statistical Genetics
• Ecology
• High-Dimensional Data Analysis

Background:

• Brain functional connectivity is analyzed using resting-state fMRI time series correlations, representing brain activity as networks.
• Existing research primarily focuses on estimating these functional brain networks.
• A critical, underexplored area is statistically testing for differences in brain networks between groups, such as patients versus controls.

Purpose of the Study:

• To evaluate and compare the performance of various state-of-the-art statistical tests for detecting group differences in brain functional connectivity networks.
• To identify robust and practical methods for analyzing group-level brain network variations.

Main Methods:

• Performance evaluation of multiple statistical tests, including Network Based Statistic (NBS), adaptive sum of powered score (aSPU), weighted adaptive sum of powered score (aSPUw), and Multivariate Matrix Distance Regression (MDMR).
• Utilized both simulated and real resting-state fMRI data for comprehensive method assessment.
• Assessed the impact of parameter choices, particularly the threshold parameter for NBS.

Main Results:

• Network Based Statistic (NBS) showed good performance but was sensitive to threshold parameter selection.
• Adaptive sum of powered score (aSPU) and its weighted version (aSPUw) demonstrated strong performance, ease of use, and adaptability for covariate adjustment.
• MDMR, when using Euclidean distance, is a specific instance of the aSPU test. aSPU and aSPUw often performed similarly but without a consistent winner.

Conclusions:

• The aSPU and aSPUw tests are recommended as powerful, user-friendly alternatives and complements to NBS for testing group differences in brain functional connectivity.
• NBS, aSPU, and aSPUw tests are collectively recommended for robust group difference analysis in brain connectivity studies.
• The choice of statistical test and parameter settings significantly impacts the detection of group differences in brain networks.