Improving the sensitivity of cluster-based statistics for functional magnetic resonance imaging data

Area of Science:

• Neuroimaging Analysis
• Statistical Neuroscience
• Brain Imaging Techniques

Background:

• High-dimensional neuroimaging data presents challenges for statistical tests, impacting sensitivity and validity.
• Existing cluster-based statistics in functional magnetic resonance imaging (fMRI) analysis may lack optimal sensitivity to detect subtle effect patterns.

Purpose of the Study:

• To enhance the sensitivity of cluster-based statistics in fMRI data analysis.
• To introduce novel cluster definitions and a min(p) method for combining test statistics.
• To validate these methods using simulations and real task fMRI data.

Main Methods:

• Development of novel cluster definitions to optimize sensitivity to plausible effect patterns.
• Implementation of the min(p) method to combine test statistics with varying sensitivity profiles.
• Utilizing the randomization inference framework for statistical validation.

Main Results:

• The proposed methods effectively control the false-alarm rate in fMRI data.
• Sensitivity profiles of cluster-based statistics are shown to vary with cluster defining thresholds and definitions.
• The min(p) method achieved up to a fivefold increase in sensitivity compared to existing fMRI analysis methods, maintaining spatial specificity.

Conclusions:

• The combination of novel cluster definitions and the min(p) method offers substantial improvements in fMRI statistical sensitivity.
• These innovations provide a more powerful and accurate approach to analyzing neuroimaging data.
• The enhanced sensitivity does not compromise the spatial specificity of the statistical inference in fMRI.