Browse by author
Lookup NU author(s): Tao Ding, Dr Tom NyeORCiD, Professor Yujiang WangORCiD
This work is licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0).
© 2025 The Authors. EEG (electroencephalogram) records brain electrical activity and is a vital clinical tool in the diagnosis and treatment of epilepsy. Time series of covariance matrices between EEG channels for patients suffering from epilepsy, obtained from an open-source dataset, are analysed. The aim is two-fold: to develop a model with interpretable parameters for different possible modes of EEG dynamics, and to explore the extent to which modelling results are affected by the choice of geometry imposed on the space of covariance matrices. The space of full-rank covariance matrices of fixed dimension forms a smooth manifold, and any statistical analysis inherently depends on the choice of metric or Riemannian structure on this manifold. The model specifies a distribution for the tangent direction vector at any time point, combining an autoregressive term, a mean reverting term and a form of Gaussian noise. Parameter inference is performed by maximum likelihood estimation, and we compare modelling results obtained using the standard Euclidean geometry and the affine invariant geometry on covariance matrices. The findings reveal distinct dynamics between epileptic seizures and interictal periods (between seizures), with interictal series characterized by strong mean reversion and absence of autoregression, while seizures exhibit significant autoregressive components with weaker mean reversion. The fitted models are also used to measure seizure dissimilarity within and between patients.
Author(s): Ding T, Nye TMW, Wang Y
Publication type: Article
Publication status: Published
Journal: Computational Statistics and Data Analysis
Year: 2025
Volume: 209
Print publication date: 01/09/2025
Online publication date: 07/03/2025
Acceptance date: 26/02/2025
Date deposited: 09/04/2025
ISSN (print): 0167-9473
ISSN (electronic): 1872-7352
Publisher: Elsevier BV
URL: https://doi.org/10.1016/j.csda.2025.108168
DOI: 10.1016/j.csda.2025.108168
Altmetrics provided by Altmetric
