B0937
Title: Robust and reproducible group-level neuroimage analysis in R with the pbj package
Authors: Simon Vandekar - Vanderbilt University (United States) [presenting]
Kaidi Kang - Vanderbilt University (United States)
Neil Woodward - Vanderbilt University Medical Center (United States)
Anna Huang - Vanderbilt University Medical Center (United States)
Maureen McHugo - University of Colorado Anschutz Medical Campus (United States)
Shawn Garbett - Vanderbilt University Medical Center (United States)
Jeremy Stephens - Vanderbilt University Medical Center (United States)
Russell Shinohara - University of Pennsylvania (United States)
Armin Schwartzman - University of California, San Diego (United States)
Jeffrey Blume - University of Virginia (United States)
Abstract: Recent simulation studies have identified group-level neuroimaging statistical inference methods that make minimal assumptions about the data and consistently control error rates for cluster and voxel-wise inference. These include permutation and bootstrap procedures that can use sandwich covariance estimators to robustly account for spatial correlation in the images. Until now, there was no software available to implement these robust analyses within R. We present the pbj R package, a validated tool to perform group-level neuroimage analysis completely within R. pbj can be combined with tools available through Neuroconductor and CRAN to perform reproducible and interactive analyses. The theory and validation of the pbj software implementation are presented. A brief tutorial of neuroimage analysis using the pbj package is given.