We embed functional iEEG connectivity of epileptic patients into a low-dimensional Euclidean space using a spectral decomposition of networks (diffusion maps). Based on the Euclidean coordinates of embedded nodes, we introduce a biomarker that may serves as a tool for seizure forecasting.

Th emethod embeds multilayer networks into the Poincaré disk. Entirely data-driven, it produces a separate embedding for each layer, while accounting for the global multilayer structure. It deals with multilayer graphs with heterogeneous node sets (with a correspondence between node-layer pairs)

A Gaussian model in hyperbolic space is introduced to characterize the distribution of a given node’s embedded positions across layers. Applied to brain networks from multiple patients with temporal lobe epilepsy, the method produces a spatially coherent localization of the regions with lessions.

The paper is here:
journals.aps.org/prresearch/a...
Very proud of my Neurodynamics colleagues M Guillemaud, L. Cousyn & V. Navarro
ArXiv version here: arxiv.org/abs/2406.10184

Representations of brain networks in a hyperbolic geometry space can reveal a daily and reliable signature of the upcoming epileptic seizure(s), thus providing a promising biomarker for seizure forecasting.