Peak p-values and false discovery rate inference in neuroimaging

Area of Science:

• Neuroimaging
• Statistical Analysis
• Brain Mapping

Background:

• Peak detection is crucial for localizing results in neuroimage analysis.
• Current SPM12 procedures use a pre-threshold and false discovery rate (FDR) for inference.
• The pre-threshold is undesirable, and FDR interpretation is problematic without a defined null hypothesis.

Purpose of the Study:

• To develop a novel peak detection method for neuroimage analysis.
• To provide a peak height distribution for Gaussian error fields that eliminates the need for a pre-threshold.
• To establish a signal-plus-noise model for controlled and interpretable FDR in peak analysis.

Main Methods:

• Developed a peak height distribution for smooth Gaussian error fields.
• Introduced a signal-plus-noise model for statistical inference.
• Provided Matlab code as an SPM extension for calculating p-values.

Main Results:

• The new method removes the need for a screening pre-threshold in peak detection.
• The proposed model allows for proper control and interpretation of FDR for peaks.
• Exact peak height distribution enables accurate p-value calculation.

Conclusions:

• The study offers an improved approach to peak detection in neuroimage analysis.
• The new method enhances the reliability and interpretability of statistical inference.
• Accessible Matlab code facilitates the application of these advancements in SPM software.