One Class SVM for Predicting Brain State
Janaina Mourão-Miranda, David Hardoon, Joao Sato and Michael Brammer
In: 15th Annual Meeting Human Brain Mapping(2008).
One aim of brain imaging studies of visual perception is to characterize neural codes for abstract percepts such as faces, non-face body parts, places, and objects. A common approach is to apply the general linear model to functional magnetic resonance imaging (fMRI) data and test each voxel individually for a mean response difference between categories. A common interpretation is that a significant difference reflects a domain-specific category preference at that voxel. Category information may also be coded in patterns distributed across voxels, which may be studied using multivariate statistics. Indeed, such applications can predict better than chance which object exemplar, facial expression or facial identity a participant is viewing (Eger, et al, in press; Furl et al, 2007). Here, we explore the potential for kernel canonical correlation analysis (kCCA) to detect category-diagnostic information in fMRI responses to facial expressions and head poses.