Parsing spatiotemporal dynamical stability in ECoG during seizure onset, propagation, and termination
Understanding brain dynamics in epilepsy is critical for establishing rigorous control objectives that enable new therapeutic methods to mitigate seizure occurrence. In multichannel electrocorticography (ECoG) recordings acquired in 21 subjects during a total of 94 seizures, we apply dynamical systems stability analysis to assess the balance versus imbalance of seizure dynamics across different timescales and brain regions. Specifically, we consider a sliding time window multivariate autoregressive linear approximation of the data captured by the ECoG channels, where eigendecomposition of the estimated matrix of coefficients describes the contribution of different regions to the spatiotemporal process (eigenvectors) associated with a particular timescale (eigenvalues). Interestingly, we observe a pattern of eigenvalue evolution and slowly changing (or approximately time-invariant) eigenvectors across both seizures and subjects. The seizure-onset is marked by an increase in high frequency spatial information to which a few regions contribute for a long period. By contrast, the seizure termination is characterized by a sudden, small time period change in dynamics to which many regions contribute. As the seizure terminates, the relatively stable ictal dynamics rapidly transition into the post-ictal regime, marked by relatively fast-damping oscillations. Our methodology offers a subject-specific characterization of the spatiotemporal behavior of the seizure, providing new insights into the dynamic patterns and functional interactions between brain regions that occur over different timescales. More generally, our approach informs the development of engineering objectives that can be used to deploy new control strategies to prevent seizure evolution or to hasten seizure termination.
Code (0)
등록된 구현이 없습니다.
Similar Papers 제목 키워드 기반
KoopGen: Koopman Generator Networks for Representing and Predicting Dynamical Systems with Continuous Spectra
Representing and predicting high-dimensional and spatiotemporally chaotic dynamical systems remains a fundamental challenge in dynamical systems and machine learning. Although data-driven models can achieve accurate shor…
Integrating Human Parsing and Pose Network for Human Action Recognition
Human skeletons and RGB sequences are both widely-adopted input modalities for human action recognition. However, skeletons lack appearance features and color data suffer large amount of irrelevant depiction. To address …
Action RecognitionHuman ParsingAction Recognition by Hierarchical Mid-level Action Elements
Realistic videos of human actions exhibit rich spatiotemporal structures at multiple levels of granularity: an action can always be decomposed into multiple finer-grained elements in both space and time. To capture this …
Action ParsingAction RecognitionClusteringTemporal Action LocalizationDynamically Encoded Actions Based on Spacetime Saliency
Human actions typically occur over a well localized extent in both space and time. Similarly, as typically captured in video, human actions have small spatiotemporal support in image space. This paper capitalizes on thes…
Action RecognitionTemporal Action LocalizationSpatiotemporal Observer Design for Predictive Learning of High-Dimensional Data
Although deep learning-based methods have shown great success in spatiotemporal predictive learning, the framework of those models is designed mainly by intuition. How to make spatiotemporal forecasting with theoretical …
Deep Learning