Dynamic multi-channel EEG graph modeling for time-evolving brain network
Keywords: EEG, Neuro Science, Graph, Time Series
TL;DR: We propose a novel dynamic GNN approach for multi-channel EEG, supported by theoretical expressivity analysis
Abstract: We describe a novel dynamic graph neural network (GNN) approach for seizure detection and prediction from multi-channel Electroencephalography (EEG) data thet addresses several limitations of existing methods.
While deep learning models have achieved notable success in automating seizure detection, static graph-based methods fail to capture the evolving nature of brain networks, especially during seizure events.
To overcome this, we propose EvoBrain, which uses a time-then-graph strategy that first models the temporal dynamics of EEG signals and graphs, and then employs GNNs to learn evolving spatial EEG representations.
Our contributions include
(a) a theoretical analysis proving the expressivity advantage of time-then-graph over other approaches,
(b) a simple and efficient model that significantly improves AUROC and F1 scores compared with state-of-the-art methods, and
(c) the introduction of dynamic graph structures that better reflect transient changes in brain connectivity.
We evaluate our method on the challenging early seizure prediction task.
The results show improved performance, making EvoBrain a valuable tool for clinical applications.
The source code is available at:
https://anonymous.4open.science/r/EvoBrain-FBC5
Primary Area: applications to neuroscience & cognitive science
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 10428
Loading