A0424
Title: Improving statistical power of multi-modal associations via de-variation
Authors: Ruyi Pan - University of Toronto (Canada)
Yinqiu He - University of Wisconsin - Madison (United States)
Jun Young Park - University of Toronto (Canada) [presenting]
Abstract: Understanding the interplay between different modalities of brain MRI data is crucial for unravelling the complexities of brain structure and function. Existing statistical association tests for two random vectors are often limited in fully capturing dependencies between modalities, particularly by overlooking correlation structures within each modality, leading to the potential loss of statistical power. A novel approach termed de-variation is proposed to address this limitation. De-variation is a simple yet effective preprocessing method that leverages a penalized low-rank factor model to capture within-modality dependencies. Theoretical analyses and simulation studies show (i) its powerful performance when within-modality correlations impact signal-to-noise ratios and (ii) its robustness when these are absent. De-variation is then applied to brain imaging-driven phenotypes (IDPs) derived from functional, structural, and diffusion MRI from the UK Biobank to show its promising performance.
