Statistical inference in brain graphs using threshold‐free network‐based statistics

Abstract

The description of brain networks as graphs where nodes represent different brain regions and edges represent a measure of connectivity between a pair of nodes is an increasingly used approach in neuroimaging research. The development of powerful methods for edge-wise group-level statistical inference in brain graphs while controlling for multiple-testing associated false-positive rates, however, remains a difficult task. In this study, we use simulated data to assess the properties of threshold-free network-based statistics (TFNBS). The TFNBS combines threshold-free cluster enhancement, a method commonly used in voxel-wise statistical inference, and network-based statistic (NBS), which is frequently used for statistical analysis of brain graphs. Unlike the NBS, TFNBS generates edge-wise significance values and does not require the a priori definition of a hard cluster-defining threshold. Other test parameters, nonetheless, need to be set. We show that it is possible to find parameters that make TFNBS sensitive to strong and topologically clustered effects, while appropriately controlling false-positive rates. Our results show that the TFNBS is an adequate technique for the statistical assessment of brain graphs.

One Model to Rule Them All: An Integrative Approach to Matrix-Based Analyses in Neuroimaging Connectomics

“…The immediate challenges that the modeling frameworks in (2) and (3) (Zalesky et al, 2010;Baggio et al, 2018).…”

“…Even though each study usually reports only one of many potential results in the literature, the number of trials are most likely hidden from the publication. Nevertheless, evidence of arbitrariness in primary thresholding leading to different results can still be seen under NBS (Baggio et al, 2018). (3) Failure to capture the intricate relatedness among RPs.…”

“…To contain the arbitrariness involved in the primary permutation version implemented in NBS, a modified version of permutation testing for NBS has been recently proposed (Baggio et al, 2018). The modification, termed as thresholdfree network-based statistics (TFNBS), is an extension of its whole brain analysis counterpart, threshold-free cluster enhancement (TFCE) (Smith and Nichols, 2009), that takes an integrative consideration between signal strength and spatial relatedness.…”

An integrative Bayesian approach to matrix‐based analysis in neuroimaging

“…Understanding the correlation structure associated with multiple brain regions is a central goal in neuroscience, as it informs us of potential “functional groupings” and network structure (Pessoa, ; Baggio et al, ). The correlation structure can be conveniently captured in a matrix format that reveals the relationships among a set of brain regions, which could involve electroencephalogram sensors, electrophysiology recordings, calcium imaging data, or functional magnetic resonance imaging (FMRI) data, among others.…”

“…First, the irregular pattern of the correlation matrix P (m) is not explicitly accounted for by the GLMs (although, to some extent, they are implicitly treated in the process of correction for multiple testing, such as in the permutation‐based methods in NBS and FSL Randomize). The second challenge is the issues of multiplicity and arbitrary dichotomization involved: As there are a total of M models (Equation ) that correspond to the M RPs, it remains a daunting job to effectively and efficiently maintain an overall false‐positive rate (PFR) under the null hypothesis significance testing (NHST) framework (Baggio et al, ; Zalesky et al, ).…”