EEG Network Topology in Juvenile Myoclonic Epilepsy
Abstract number :
1.511
Submission category :
3. Neurophysiology / 3G. Computational Analysis & Modeling of EEG
Year :
2023
Submission ID :
1315
Source :
www.aesnet.org
Presentation date :
12/2/2023 12:00:00 AM
Published date :
Authors :
Presenting Author: Felipe Branco de Paiva, MD – UW-Madison
Moo Chung, PhD – Biostatistics and Medical Informatics – UW-Madison; Aaron Struck, MD – Neurology – UW-Madison
Rationale: When examining epilepsy as a network disorder, it is conventional to apply a graph-theoretical approach to brain graphs derived from time-series data. However, in standard graph theory-based network analysis, metrics depend on the choice of threshold applied to weighted edges. The choice of threshold has the potential to alter the final results, which can jeopardize the result’s biological interpretation and hinder reproducibility. Persistent homology is a branch of algebraic topology that offers a novel solution to this multiscale analysis challenge – instead of examining networks at one fixed scale, it identifies persistent topological features that are robust over different scales. Our previous work has shown how graph filtration – an analysis tool derived from persistent homology – can be used to detect significant differences in network topology between healthy controls (HCs) and temporal lobe epilepsy patients when applied to functional MRI-derived networks. In the present work, we applied persistent homology to our growing EEG database of juvenile myoclonic epilepsy (JME) patients and HCs to determine whether the network topology can also be used as a biomarker of JME.
Methods: Resting-state high-density EEG from 100 subjects were used for the analysis. Datasets were source reconstructed using sLORETA. The source estimates were projected on a brain template and averaged into 100 regions of interest (ROIs). The functional brain network for each subject was modeled as an undirected graph, with each ROI representing one node and the magnitude of the wavelet coherence between ROIs representing the edges. Finally, we applied graph filtration to each network by gradually increasing the edge threshold while keeping track of certain topological features. Graph filtration generates two curves per subject – the Betti-0 curve keeps track of the number of connected components while the Betti-1 curve keeps track of the number of independent cycles. These curves were compared between every possible pairwise combination of subjects using the Wasserstein distance – the probabilistic version of optimal transport, used here to measure the topological discrepancy between persistent diagrams. The sum of all pairwise distances between both groups was divided by the sum of all pairwise distances within groups which defines the ratio statistic. To assess the statistical significance of the ratio statistic, we used a non-parametric permutation test.
Results: The ratio statistic was significant (Figure 1) for wavelet coherence in the alpha band (8-12 Hz) when comparing all HCs (n = 42) with JME patients (n = 42) matched for age and sex. Results in the theta (4-8 Hz) and beta (13-30 Hz) bands were not significant.
Conclusions: Our results indicate that the alpha band EEG-derived network topology can robustly differentiate between HCs and JME patients at the group level. Our innovative method further highlights the importance of the alpha band in the pathophysiology of JME.
Funding: This study was supported by NIH R01NS111022 and an AES Postdoctoral Research Fellowship.
Neurophysiology