Classification Inference for Event-related fMRI Data

Abstract

This study presents methods, results and conclusions used and drawn from an attempt to cluster fMRI brain activation patterns in an unsupervised manner in order to retrieve a categorization in a set of nouns denoting objects. Over the last few years, a lot of work has been done on the classification of neural responses patterns corresponding to the presentation of different objects or concepts, sometimes called brain reading. Here, we try to infer a categorization from such a set of neural response patterns based only on the similarities between these patterns. By assigning each neural response pattern for the presentation of an object to a cluster in a multidimensional space, a categorization in the set patterns can be induced. Furthermore, clustering methods are used to address the question which number of clusters is most plausibly present in the pattern set. As it turns out, an experiment design with long inter trial intervals and multiple stimulus presentations is critical for such an approach to be successful.