On the Parcellation of Functional Magnetic Resonance Images

Abstract

Functional Magnetic Resonance Imaging (fMRI) is one of the techniques for measuring activities in the brain and it has been demonstrated to have a high potential in clinical application. However, fMRI is limited by some of the contradictory results reported by different studies. One of the possible reasons for this contradiction is the lack of standard and acceptable methods of analyzing fMRI data. Analysis of fMRI data in studies focusing on brain connectivity normally requires the definition of region of interest. This is normally done using regions of interest drawn on high resolution anatomical images. The use of anatomical images implies using structural information, thereby losing any functional information that could improve the analysis of fMRI data. In this article, we present the framework for a region of interest definition for fMRI using structural and functional information. Contrary to existing approaches, the proposed method will also consider the use of network properties. The method uses a bottom-up approach as it starts with structural information, then include functional information before it finally includes network properties. We hypothesize that the use of multiple information in defining the regions of interests in fMRI data will produce a more accurate, more reproducible and more trusted results than the use of structural information only. It is hoped that the use of the proposed model will lead to improved analysis of fMRI brain data, hence increasing its diagnostic potential.