MULTI-VOXEL PATTERN ANALYSIS OF fMRI DATA DURING SELF- AND OTHER-REFERENTIAL PROCESSING

Abstract

The study of brain activity in the processing of self-referential information, as compared to the processing of information related to other people, is based on the application of mass-univariate analysis, based on the assumption that activity in one region is independent of activity in other regions. Recently, there has been a growing interest in neuroimaging to investigate spatially distributed information using multivariate approaches such as multivoxel pattern analysis (MVPA). In this paper, we used MVPA to analyze fMRI data recorded during self-evaluation and evaluation of other people of varying proximity. In all pairwise classifications tested, the number of correct identifications was significantly higher than the level of random matches. Predictively significant structures were widely distributed over different brain regions and included areas of the visual, lateral prefrontal, and many other cortical areas in addition to the cortical midline structures that contributed the most. In the self-other classification, ventral areas of the medial prefrontal and cingulate cortex were the most informative for the self condition, whereas parietal and occipital medial areas were the most informative for the other condition. The combination of brain structures, which included the anterior cingulate cortex and both amygdalae, revealed by principal component analysis, correlated positively with the psychometric scale of sensitivity to reward, and negatively with neuroticism scales. Overall, the results show the fruitfulness of using machine learning methods to analyze data from such kinds of experiments.