Anisotropic kernels for coordinate-based meta-analyses of neuroimaging studies

Area of Science:

• Neuroimaging
• Neuroscience
• Biostatistics

Background:

• Neuroimaging meta-analyses use kernels to create brain maps from study peaks.
• Current isotropic kernels assume uniform influence around peaks, which is biologically implausible.
• Anisotropic kernels offer a more realistic approach by accounting for spatial correlations.

Purpose of the Study:

• To introduce and validate anisotropic kernels for neuroimaging meta-analyses.
• To improve the accuracy of effect size and likelihood maps in studies.
• To provide new correlation templates for enhanced meta-analysis.

Main Methods:

• Developed anisotropic kernels considering spatial correlations between voxels.
• Created correlation templates for gray matter, white matter, CSF, and FA.
• Validated the method by recreating effect size maps from peak data.

Main Results:

• Anisotropic kernels significantly improved the recreation of effect size maps.
• The optimal degree of anisotropy and FWHM varied with data.
• Full anisotropy improved recreation independently of FWHM, suggesting its broad utility.

Conclusions:

• Anisotropic kernels are recommended for effect-size signed differential mapping (ES-SDM) and activation likelihood estimation (ALE).
• Freely available software and templates support the adoption of this improved methodology.
• Further research may focus on estimating optimal meta-analysis-specific parameters.