Thresholding of statistical maps in functional neuroimaging using the false discovery rate

Area of Science:

• Neuroimaging analysis
• Statistical methods in neuroscience
• Brain imaging data interpretation

Background:

• Determining appropriate thresholds for voxelwise statistics in neuroimaging is a significant challenge.
• Standard multiple hypothesis testing corrections (e.g., Bonferroni) are often too conservative for neuroimaging data, leading to reduced sensitivity.
• The high number of voxelwise tests necessitates robust statistical approaches to control error rates effectively.

Purpose of the Study:

• To introduce and evaluate statistical procedures for controlling the false discovery rate (FDR) within the neuroimaging literature.
• To demonstrate the utility of FDR controlling procedures for analyzing voxelwise statistical tests in neuroimaging.
• To provide a more sensitive method for identifying statistically significant findings in brain imaging studies.

Main Methods:

• Application of statistical procedures designed to control the expected proportion of falsely rejected hypotheses (FDR).
• Simultaneous analysis of all voxelwise test statistics to determine significance.
• Validation through Monte Carlo simulations and analysis of functional magnetic resonance imaging (fMRI) data from two experimental paradigms.

Main Results:

• FDR controlling procedures effectively manage the rate of false positives in voxelwise neuroimaging analyses.
• The proposed methods demonstrate greater sensitivity compared to traditional approaches like Bonferroni correction.
• Successful application of FDR control to real fMRI data, identifying significant regions in simple experimental designs.

Conclusions:

• False discovery rate (FDR) controlling procedures represent a powerful and sensitive statistical tool for neuroimaging research.
• These methods offer an objective and effective way to set thresholds for voxelwise statistics, enhancing the interpretability of neuroimaging findings.
• The integration of FDR control into neuroimaging analysis workflows can lead to more reliable and meaningful discoveries in brain science.