Epilepsy is a neurological condition characterized by recurrent unprovoked seizures whose cause is primarily unknown. Drug-resistant epilepsy, accounting for one-third of known cases, can be remediated by neurostimulation and surgical resection. Identifying regions within the brain responsible for triggering seizures, commonly called seizure onset zones (SOZs), is crucial for effective treatment. Machine learning (ML) and deep learning (DL) algorithms for SOZ identification using intracranial EEG features have improved seizure location, making them a valuable tool for faster analysis of a large amount of data. These algorithms, however, often require feature extraction that heavily relies on domain knowledge, and they fail to map brain connectivity by locating Regions of Interest (ROIs). The fundamental network-like structure of the brain is often ignored in these models.

PhD student Sai Sanjay Balaji (Electrical Engineering), in a project called “Graph convolution network (GCN) for seizure onset zone identification from iEEG,” is aiming to develop an end-to-end graph neural network model for identifying SOZs. Using the intracranial EEG time-series, the model learns the graphical representation of directed functional connectivity and performs subsequent node classification to locate SOZs using multi-layer graph convolution networks (GCN). The main advantages of the proposed model over prior ML and DL models are: feature extraction is not required as the graph adjacency is learned during model training; graph representation can better interpret the directed functional connectivity from the spatio-temporal iEEG data; and analysis using topological graph features facilitates identification of potential SOZs along with actual SOZs. Future work on time-evolving connectivity using attention mechanisms can characterize seizure propagation in the human brain, thereby improving personalized epilepsy treatment plans. Further validation by individual datasets can pave the way for closed-loop neurostimulation through a unified seizure prediction and localization model.

Some funding for this project was provided by a 2023 University of Minnesota Informatics Institute MnDRIVE PhD Graduate Assistantship. The UMII MnDRIVE Graduate Assistantship program supports U of M PhD candidates pursuing research at the intersection of informatics and any of the five MnDRIVE areas:

  • Robotics
  • Global Food
  • Environment
  • Brain Conditions
  • Cancer Clinical Trials

This project is part of the Brain Conditions MnDRIVE area. See the complete list of the UMII Graduate Assistantships for 2023.

 

flow chart showing multi-layer graph convolution networks