Detecting spatially high-resolved brain connectivity patterns: a multi-subject approach

• Applicant: Dr. Britta Pester
• Subject Area: Medical Physics, Biomedical Technology ; Epidemiology and Medical Biometry/Statistics
• Term: from 2016 to 2019
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 314806675

The main topic of this proposal is the development of an approach for the extension of a subject-wise analysis of high-dimensional brain networks towards an integrative group approach that enables a comparison amongst a large number of subjects.A comprehensive insight into brain processes requires both, the consideration of brain activity as well as an understanding of information flow (connectivity) between and within structures of the underlying neural system. Commonly this connectivity is indirectly estimated on the basis of statistical dependencies between recorded brain signals. The amount of considered interaction thereby increases quadratically with the number of recorded signals. Therefore, the connectivity analysis of spatially high-resolved data (such as fMRI) results in highly complex networks, comprising numerous voxels together with directed connections in-between. This large amount of analysis output precludes any (statistical) evaluation of the results and requires further processing. Here, connectivity-based segmentation of the networks can offer new insights into network patterns by identifying subsets of voxels (so-called modules) that exhibit a pronounced interaction within each module.However, a remaining issue in the course of this methodology is how to integrate multiple subjects or multiple groups of subjects. This is due to the problem that identified modules are not assigned to each other across subjects (correspondence problem). Thus, it is difficult to identify corresponding modules, and in many cases there are no corresponding modules.The aim of this project is to develop an approach that enables the functional segmentation of high-dimensional multi-subject data, representing a first step towards a highly resolved connectivity analysis within large cohorts. The basic idea is to rearrange the subject-specific networks into an integrative tensor and to subsequently decompose this tensor by means of parallel factor analysis. Then, the resulting spatial loadings can be translated into networks and segmented into modules that allow a comparison between the subjects.Thorough simulations will be utilized for a profound evaluation of the approach. To validate the methodology in practice, data from a resting state fMRI experiment will be used. Here, results from conventional analysis techniques (such as investigations concerning activity or low-dimensional connectivity analyses) serve for the assessment of the benefit obtained by the proposed methodology.

• DFG Programme: Research Grants