Dynamic graphs make it possible to understand the evolution of complex systems along time. In this thesis, we look at their application to understand one of the most common
neurological disorder that affects around 1% of the population: epilepsy. A complete and objective characterization of the patient-specific dynamic graph describing this pathology
is crucial for optimal surgical treatment.
First, we propose to modify a measure of functional connectivity, the Phase-Locking-Value, in order to infer robust dynamic graph from the neurophysiological signals recorded during an epileptic seizure.
Then, new constrained tensor decomposition methods are proposed and applied to extract the principal features from the dynamic graph describing the pathology.
Finally, a clinical study is performed to compare the obtained features from the visual interpretation of a clinician, expert in neurophysiological signal interpretation.