Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SpaTeoGL: Spatiotemporal Graph Learning for Interpretable Seizure Onset Zone Analysis from Intracranial EEG

Elham Rostami, Aref Einizade, Taous-Meriem Laleg-Kirati

TL;DR

The paper tackles SOZ localization in iEEG by introducing SpaTeoGL, a spatiotemporal graph learning framework that jointly learns window-level spatial graphs $\mathbf{L}^{(s)}_m$ and a temporal graph $\mathbf{L}^{(t)}$ under a smooth graph signal processing model. The optimization uses alternating block coordinate descent with convergence guarantees, and the objective combines spatial smoothness $\mathrm{tr}(\mathbf{X}_m^\top \mathbf{L}^{(s)}_m \mathbf{X}_m)$, temporal smoothness $\mathrm{tr}(\tilde{\mathbf{X}}^\top \mathbf{L}^{(t)} \tilde{\mathbf{X}})$, and Frobenius regularization $\beta \|\cdot\|_F^2$ on Laplacians, with constraints to valid Laplacian sets. Empirical results on a multicenter iEEG dataset show SpaTeoGL is competitive with a strong HVG+LR baseline for SOZ detection and superior for non-SOZ discrimination, while providing interpretable insights into onset–propagation regimes and spatial localization around clinically reported SOZ electrodes. The work demonstrates that joint spatiotemporal graph representations can enhance presurgical evaluation by offering quantifiable, interpretable network dynamics of seizures.

Abstract

Accurate localization of the seizure onset zone (SOZ) from intracranial EEG (iEEG) is essential for epilepsy surgery but is challenged by complex spatiotemporal seizure dynamics. We propose SpaTeoGL, a spatiotemporal graph learning framework for interpretable seizure network analysis. SpaTeoGL jointly learns window-level spatial graphs capturing interactions among iEEG electrodes and a temporal graph linking time windows based on similarity of their spatial structure. The method is formulated within a smooth graph signal processing framework and solved via an alternating block coordinate descent algorithm with convergence guarantees. Experiments on a multicenter iEEG dataset with successful surgical outcomes show that SpaTeoGL is competitive with a baseline based on horizontal visibility graphs and logistic regression, while improving non-SOZ identification and providing interpretable insights into seizure onset and propagation dynamics.

SpaTeoGL: Spatiotemporal Graph Learning for Interpretable Seizure Onset Zone Analysis from Intracranial EEG

TL;DR

The paper tackles SOZ localization in iEEG by introducing SpaTeoGL, a spatiotemporal graph learning framework that jointly learns window-level spatial graphs and a temporal graph under a smooth graph signal processing model. The optimization uses alternating block coordinate descent with convergence guarantees, and the objective combines spatial smoothness , temporal smoothness , and Frobenius regularization on Laplacians, with constraints to valid Laplacian sets. Empirical results on a multicenter iEEG dataset show SpaTeoGL is competitive with a strong HVG+LR baseline for SOZ detection and superior for non-SOZ discrimination, while providing interpretable insights into onset–propagation regimes and spatial localization around clinically reported SOZ electrodes. The work demonstrates that joint spatiotemporal graph representations can enhance presurgical evaluation by offering quantifiable, interpretable network dynamics of seizures.

Abstract

Accurate localization of the seizure onset zone (SOZ) from intracranial EEG (iEEG) is essential for epilepsy surgery but is challenged by complex spatiotemporal seizure dynamics. We propose SpaTeoGL, a spatiotemporal graph learning framework for interpretable seizure network analysis. SpaTeoGL jointly learns window-level spatial graphs capturing interactions among iEEG electrodes and a temporal graph linking time windows based on similarity of their spatial structure. The method is formulated within a smooth graph signal processing framework and solved via an alternating block coordinate descent algorithm with convergence guarantees. Experiments on a multicenter iEEG dataset with successful surgical outcomes show that SpaTeoGL is competitive with a baseline based on horizontal visibility graphs and logistic regression, while improving non-SOZ identification and providing interpretable insights into seizure onset and propagation dynamics.
Paper Structure (7 sections, 1 theorem, 5 equations, 4 figures, 1 table, 1 algorithm)

This paper contains 7 sections, 1 theorem, 5 equations, 4 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. SpaTeoGL: Spatiotemporal Graph Learning
  3. Experimental Protocol and Results
  4. Q1: Quantitative comparison with baseline
  5. Q2: Temporal graph learning for revealing temporally coherent regimes (onset vs. propagation)
  6. Q3: Spatial graph learning to localize clinically reported SOZ electrodes
  7. Conclusion

Key Result

Theorem 1

Assume $\beta>0$ and that each block subproblem eq:spatial_subprob and eq:temporal_subprob is solved exactly over nonempty closed convex sets $\mathcal{L}^{(s)}$ and $\mathcal{L}^{(t)}$. Then, the objective in eq:joint is non-increasing across SpaTeoGL iterations. Moreover, every limit point of the

Figures (4)

  • Figure 1: Illustration of the SpaTeoGL method in an example setting. Here, the multi-electrode recorder data $\mathbf{X}\in\mathbb{R}^{N\times K}$ with $N=4$ is devided into $M=4$ windows of $\{\mathbf{X}_m\in\mathbb{R}^{N\times K_m}\}_{m=1}^M$. Therefore, the SpaTepGL learns $M=4$ spatial graphs $\{\mathcal{G}^{(s)}_m\}_{m=1}^M$ representing the connections between electrodes in each window, and one temporal graph $\mathcal{G}^{(T)}$ (with red edges in the figure) showing the connection between $M=4$ windows.
  • Figure 2: Comparison of Normalized Interpolated Histogram for two methods (HVG+LR and SpaTeoGL) in classifying SOZ (left) and non-SOZ (right) electrodes. While no statistically significant difference was found for SOZ electrodes (p = 0.6404), a significant difference was observed for non-SOZ electrodes (p = 0.0048), with SpaTeoGL showing superior performance.
  • Figure 3: Temporal graph and bar plot for each spatial graph
  • Figure : (a)

Theorems & Definitions (2)

  • Theorem 1
  • proof