Comparing spatial null models for brain maps

Abstract

Technological and data sharing advances have led to a proliferation of high-resolution structural and functional maps of the brain. Modern neuroimaging research increasingly depends on identifying correspondences between the topographies of these maps; however, most standard methods for statistical inference fail to account for their spatial properties. Recently, multiple methods have been developed to generate null distributions that preserve the spatial autocorrelation of brain maps and yield more accurate statistical estimates. Here, we comprehensively assess the performance of ten such published null frameworks in controlling the family-wise error rate in statistical analyses of parcellated neuroimaging data. We apply each framework on two prototypical analyses: (1) testing the correspondence between brain maps (e.g., correlating two activation maps) and (2) testing the spatial distribution of a feature within a partition (e.g., quantifying the specificity of an activation map within an intrinsic functional network). In agreement with previous reports, we find that naive null models that do not preserve spatial autocorrelation consistently yield unrealistically liberal statistical estimates. Performance of spatially-constrained null models depended on research context; model performance was generally consistent when testing correspondences between brain maps, but considerably more variable when testing partition specificity. Throughout these analyses, we observe minimal impact of parcellation and parcel resolution on null model performance. Altogether, our results highlight the need for continued development and standardization of statistically-rigorous methods for comparing brain maps.

Topographic gradients of intrinsic dynamics across neocortex

“…To confirm this finding is not driven by spatial proximity, we repeated the analysis with distance-residualized values ( Mišić et al, 2014 ), finding a significant difference (two-tailed t -test; , ). We also repeated the analysis using a nonparametric label-permutation null model with preserved spatial autocorrelation (10,000 repetitions) ( Alexander-Bloch et al, 2018 ; Markello and Misic, 2020 ), again finding significantly greater within- compared to between-network temporal profile similarity (two-tailed; ; Figure 2d ). These results are consistent when applying the 17 network partition of intrinsic networks ( Yeo et al, 2011 ; Schaefer et al, 2018 ; Figure 2—figure supplement 1 ).…”

“…We use Spearman rank correlations ( ) throughout, as they do not assume a linear relationship among variables. Given the spatially autocorrelated nature of both hctsa features and other imaging features, we assess statistical significance with respect to nonparametric spatial autocorrelation-preserving null models ( Alexander-Bloch et al, 2018 ; Markello and Misic, 2020 ).…”

“…A consistent question in the present work is the topographic correlation between time-series features and other features of interest. To make inferences about these links, we implement a null model that systematically disrupts the relationship between two topographic maps but preserves their spatial autocorrelation ( Alexander-Bloch et al, 2018 ; Markello and Misic, 2020 ) (see also Burt et al, 2018 ; Burt et al, 2020 for an alternative approach). We first created a surface-based representation of the Cammoun atlas on the FreeSurfer fsaverage surface using the Connectome Mapper toolkit ( https://github.com/LTS5/cmp ; Daducci et al, 2012 ).…”

Learning function from structure in neuromorphic networks

“…(b) To determine whether memory capacity estimates depend on the underlying connectivity structure, a null distribution is constructed by randomly rewiring pairs of edges, while preserving network size, density, degree sequence and node-level intrinsic-network assignment (cyan in panel (a); referred throughout the text as rewired ) [86]. (c) To evaluate the extent to which the partition of the connectome into seven intrinsic networks is relevant for the task at hand, a null distribution is constructed by spherical projection and random rotation of the intrinsic network labels, preserving their spatial embedding and autocorrelation (yellow in panel (a); referred throughout the text as spin ) [1, 85].…”

Rapid processing and quantitative evaluation of multicontrast EPImix scans for adaptive multimodal imaging

“…Permutations were generated by randomly rotating a projection of ROI centroids on the (FreeSurfer) sphere, before mapping rotated ROIs to the nearest unrotated ones. Mirrored rotations were applied to the contralateral hemisphere, resulting in a permutation which controls for spatial autocorrelation and hemispheric symmetry of regions (Váša et al, 2018;Alexander-Bloch et al, 2018;Markello and Misic, 2020). P-values for the correlation between two regional maps were obtained by comparing the empirical value of Spearman's ρ to a null distribution of Spearman correlations, generated by correlating one of the empirical maps to a set of 10,000 spatially permuted versions of the other map; these "spin-test" Pvalues are referred to as P spin .…”