Enhancing the reproducibility of group analysis with randomized brain parcellations

Area of Science:

• Neuroimaging and Computational Neuroscience
• Brain Organization and Variability Analysis

Background:

• Group analyses in neuroimaging compare brain organization variability with behavioral or genetic factors.
• Current voxel-based methods lack stability and sensitivity, leading to irreproducible results.
• Assessing risk factors for brain diseases requires robust analytical techniques.

Purpose of the Study:

• To introduce a novel approach for detecting active voxels in brain images.
• To overcome the limitations of standard voxel-based analysis schemes.
• To improve the sensitivity, recovery, and reproducibility of neuroimaging group analyses.

Main Methods:

• A new method detects active voxels based on consensus across multiple random brain image parcellations.
• A permutation test is employed to control the false positive risk.
• The approach was validated using both synthetic and real neuroimaging data.

Main Results:

• The proposed method demonstrated higher sensitivity, better recovery, and improved reproducibility compared to standard methods.
• A significant association was detected between a genetic variant near the COMT gene and a left thalamus region.
• This association was identified in a functional magnetic resonance imaging (fMRI) contrast within an imaging-genetic study.

Conclusions:

• The novel consensus-based parcellation and permutation testing approach offers a more robust alternative to standard voxel-based analyses.
• This method enhances the reliability of findings in neuroimaging group studies, particularly in imaging-genetics.
• The study successfully identified a specific genetic influence on brain structure in the left thalamus.