Epilepsy

Roughly 30% of epilepsy patients do not achieve alleviation of seizures using medications. Patients with drug resitant epilepsy become candidates for surgical resection of brain tissue suspected to be the seizure onset zone (SOZ). Intracranial EEG (iEEG) data is collected from these patients to localize the seizure onset zone, which is done by manually or semi-automatically scanning through hours of iEEG recordings. iEEG data presents several challenges for the development of automated techniques such as high-dimensionality, non-stationarity, and heterogeneity of SOZ across patients.

We have discovered iEEG biomarkers and developed ML methods to addresses two important clinicals challenges: (i) SOZ localization, and (ii) predicting seizure clustering. We observed that relative entropy, a bivariate iEEG feature which represents the similarity in brain activity between pairs of iEEG electrodes, is a suitable biomarker for the above tasks. Relative entropy also enables abstracting streams of iEEG data into time-evolving graphs - rendering the data suitable for application of graph theoretical techniques.

Relevant publications

• K. V. Saboo, Y. Cao, V. Kremen, V. Sladky, N. M. Gregg, P. M. Arnold, P. J. Karoly, D. R. Freestone, M. J. Cook, G. A. Worrell, R. K. Iyer (2023). Individualized seizure cluster prediction using machine learning and chronic ambulatory intracranial EEG. IEEE Transactions on NanoBioscience.

• K. V. Saboo*, I. Balzekas*, V. Kremen, Y. Varatharajah, M. T. Kucewicz, R. K. Iyer, G. A. Worrell (2021). Leveraging electrophysiologic correlates of word encoding to map seizure onset zone in focal epilepsy: Task-dependent changes in epileptiform activity, spectral features, and functional connectivity. Epilepsia.

• Y. Varatharajah, M.J. Chong, K. V. Saboo, B. M. Berry, B. Brinkmann, G. A. Worrell, R. K. Iyer (2017). EEG-GRAPH: A factor graph-based model for capturing spatial, temporal, and observational relationships in electroencephalograms. Advances in Neural Information Processing Systems (NeurIPS).