CMStatistics 2022: Start Registration
View Submission - CMStatistics
B1563
Title: Hypothesis testing and spatial localization of associations in multi-modal neuroimaging studies Authors:  Sarah Weinstein - Temple University (United States) [presenting]
Russell Shinohara - University of Pennsylvania (United States)
Jun Young Park - University of Toronto (Canada)
Abstract: Many neuroimaging-based studies of neurodevelopment and mental health involve collecting and integrating measures of both brain structure and function. In such studies, we are often interested in conducting hypothesis tests of associations between these different modalities and evaluating whether these associations differ across subgroups. Until recently, statistical methods in this area have primarily evaluated global associations--for example, testing whether associations exist throughout the entire brain or within pre-defined anatomical subregions or functional networks. We propose a new method for spatial localization of inter-modal associations. We first adjust for the underlying spatial autocorrelation structure in each imaging modality to address heterogeneity between modalities. Second, we use clusterwise inference to leverage spatial information and construct an interpretable map of spatially enhanced test statistics. Finally, we use permutation for inference to ensure type I error and family-wise error rate control. Through simulation studies using multi-modal neuroimaging data from the Philadelphia Neurodevelopmental Cohort, we illustrate our methods' statistical power, interpretability, and ability to replicate findings even in small-sample settings.