Machine learning for brain networks : task identification based on BOLD fMRI signals

Details

Supervisors

Delvenne, Jean-Charles

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] : ingénieur civil en mathématiques appliquées, à finalité spécialisée

Abstract

Data has become one of the main issues of the 21st century and their potential contribution to society is huge. One of the most interesting sources of data could come from our brain. It is composed of about one hundred millions of neurons each connected to other neurons by up to fifteen thousand synapses and forming a gigantic network responsible for the treatment of signals coming from sensors from all over our body and leading our actions in everyday life. Detailed access to brain activity could help us understand the most important organ of the human body, which is still the subject of lots of interrogations. The last decades saw the emergence of neuroimaging techniques capable to give us more and more precise brain networks representations. The introduction of these data into a machine learning framework takes all its interest. Classification of brain networks can help us understand some brain disorder responsible for pathologies like Alzheimer or schizophrenia, but also brain behaviour when we perform different actions. In this thesis, we will use functional magnetic resonance imaging data from the human connectome project to classify resting-state fMRI and seven other task fMRI from these hundred subjects with three different methods : (i) using functional connectomes (ii) using raw time series of the fMRI and (iii) using autocorrelation of these time series.