Interpretable High-Dimensional Inference via Score Maximization with an Application in Neuroimaging

• Activity Number: 659
• Type: Contributed
• Date/Time: Thursday, August 4, 2016 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Imaging
• Author(s): Simon Vandekar* and Philip Reiss and Russell Shinohara
• Companies: University of Pennsylvania and New York University and University of Pennsylvania
• Keywords: Score Test ; Score maximization ; neuroimaging ; big data

Abstract

In the fields of neuroimaging and genetics there is interest in testing the association of a single categorical or continuous outcome with a very high-dimensional imaging or genetic variable. Oftentimes several summary measures of the high-dimensional variable are created to sequentially test and localize the association with the outcome. In some cases the results for the summary measures are significant, but subsequent tests used to localized differences are underpowered and do not indicate regions that are significantly associated with the outcome. We propose a modification of Rao's score test based on maximizing the score statistic in a linear subspace of the parameter space. If the test rejects the null hypothesis, then we provide methods to perform inference on the scores in the high-dimensional space by projecting the scores to the subspace where the score test was performed. This allows for inference in the high-dimensional space on the same degrees of freedom as the score test. We illustrate the method using cortical thickness data from the Alzheimer's Disease Neuroimaging Initiative.