PaSTA: Fast parametric inference of significance for spatial associations between brain maps

Testing for spatial correlations between pairs of brain maps has emerged as a central task in neuroimaging studies. Determining the statistical significance of such correlations is challenging due to the presence of spatial autocorrelation in brain maps. Here we establish a novel parametric method, Parametric Spatial Test for Associations (PaSTA), to infer the significance of spatial associations between brain maps via covariance-variance modelling and effective degrees of freedom estimation. Our method is fast, reliable, and sensitive, enabling flexible significance testing of brain map correlations over arbitrary cortical surface and brain volumetric domains. We examine the sensitivity and specificity of PaSTA using simulated datasets with known ground truth and demonstrate its utility when applied to empirical brain maps. We extend PaSTA to approximately model modest nonstationarity in spatial autocorrelation and show that, our method yields improved false positive control and statistical power relative to existing approaches when brain maps are spatially heterogeneous.