SNIC SUPR
Community detection in temporal brain networks
Dnr:

sens2019536

Type:

SNIC SENS

Principal Investigator:

Granit Kastrati

Affiliation:

Karolinska Institutet

Start Date:

2019-04-15

End Date:

2020-05-01

Primary Classification:

30105: Neurosciences

Allocation

Abstract

The human brain is a complex hierarchical and dynamic network giving rise to all that makes us unique in this world. In order to understand this system, it is necessary to undertand in what ways it is organised. Discovering the organisation of the brain, its sub-division into modules or communities, with its ability to integrate and segregate, enables an understanding of the functioning of the mind. Detecting communities in networks is popular in network theory. Proposals about new ways to detect communities in static and dynamic communities continue to appear. Community detection algorithms applied to complex networks require large computational power. The present work aims to apply novel community detection algorithms to brain networks from human subjects. This project therefore handles sensitive information. Specifically, this data comes from functional magnetic resonance imaging (fMRI) and represent blood-oxygen-level dependent responses (BOLD), an indirect measure of brain activity, by capturing differences in magnetic properties of oxygenated and deoxygenated haemoglobin. The importance of this is that with communities detected in the human brain, measures from network theory can be applied to reveal insights into the functioning of the brain. For this to be possible, the algorithms must capture true partitions of the brain, for network theoretic measures to be meaningful. The estimation about projected needs comes from attempts to run these analyses on Linux on my own computer. For one participant it takes >4 days and for the present study there are 300 participants. The data that I have now is ~800GB. This represent one way out of many in which one can detect communities. If one wants to do research on this and try several methods, then larger computational capacity is required.