A data-driven controllability analysis of cortico-cortical evoked potentials

Several authors have used network control theory to model the brain as a controllable network, Gu et al. (2015). However, previous work focused almost exclusively on model-based control techniques, an approach that is severely weakened by the need to correctly identify the underlying network connectivity (adjacency matrix). This limitation can be overcome by using the data-driven control approach, where explicit knowledge of the underlying network is not required, and can be bypassed using the observations of the system under perturbations. Baggio et al. (2021) introduced a solid way to apply data-driven control techniques to complex networks, showing their advantages over model-based methods. In this thesis, we first present model-based control techniques. Then, we review the data-driven techniques using Baggio's framework, we validate them on synthetic networks and the Kuramoto model and compare them with model-based techniques in a theoretical setup. Finally, data-driven control techniques are used on a dataset by Parmigiani et al. (2022), comprising cortico-cortical evoked potentials (CCEPs) in epileptic patients. By using stereoencephalographic (sEEG) stimulus information (specifically intracranial electroencephalography (iEEG)) and the high-density encephalographic (hdEEG) recordings, we characterize the underlying control properties of the individual brains. These properties are first examined through the controllability matrix and the Gramian matrix estimated purely from the data, and then used to characterize the reachable subsets and the control inputs needed to achieve chosen target states in different areas of the brain. Our analysis aims to assess the possibility of employing data-driven methods for human brain data analysis without explicit knowledge of the underlying system's dynamics, hence broadening our understanding of brain dynamics. Moreover, this thesis aims to answer questions such as which states could be reached by using the available stimulus space, how easily they can be reached, and how these patterns relate to intrinsic patterns observed during spontaneous dynamics.