Section: Partnerships and Cooperations

Regional Initiatives

Connectome, and large graph mining

Participant : Philippe Preux.

• Title: Connectome and epilepsy

• Type: No funding yet (self-funded project)

• Coordinator: Louise Tyvaert, Department of clinical neurophysiology, CHRU Lille, Université de Lille 2, France

• Others partners: Mostrare, Inria Lille

• Duration: Began in spring 2012

• Abstract: The long term goal of this collaboration is to investigate the use of machine learning tools to analyse connectomes, and possibly related EEG signals, to determine, for a given patient, the region of the brain from which originate epilepsy strokes. As a first step, we concentrate on connectome, that is a graph representation of the connectivity in the brain. We study the properties of these graphs from a formal point of view, and try to match these properties with brain activity, and brain disorders.

• Activity Report: being a multi-disciplinary project, the first thing was to understanding each others. Connectomes having been acquired at the hospital via MRI and image processing, the resulting graphs have been processed using a spatially regularized spectral clustering approach; we were able to recover well-known brain areas automatically. Indeed, one of the first issues to clarify is the relevance of the graph representation of these MRI data (connectomes), an issue unclear in the medicine community. These first results have been submitted for publication at the IEEE 2013 symposium on Bio-Imaging (ISBI'2013).