Composite Hypothesis Testing Via Support Vector Regression

Nagesh Adluru¹, Richard J. Davidson¹, Andrew L. Alexander¹

Hypothesis testing is one of the most commonly performed statistical analyses in neuroimaging studies. Often the goal is to study the effects of variables such as physiological, behavioral and/or clinical on a variety of neuroimaging measurements e.g., cortical thickness, bold-activation, microstructural properties. A most common approach to test the significance of these effects is to project the measured neuroimaging data onto the linear models defined using the variables of interest via ordinary least squares. In this paper we propose that using a more robust projection can increase sensitivity to the effects and hence can also control for confounding parameters.