Abstracts

This abstract has been invited to present during the Broadening Representation Inclusion and Diversity by Growing Equity (BRIDGE) poster session

Rationale: About a third of epilepsy patients are medically intractable and are forced to seek other alternatives. Surgical resection of the epileptogenic zone is irreversible and has variable outcomes. Electrical stimulation therapies have shown to have curative effects in suppressing seizures, as well as reducing overall seizure frequency. However, the optimization of stimulation parameters to maximize therapeutic effects remains a challenge. It is presently a trial-and-error process based on the patient’s response to therapy, which requires several months to reach a set of parameters with highest efficacy in reducing seizure frequency.
There is a need for a quantitative method that is able to track intrinsic epileptogenic network dynamics, its evolution towards seizure, and the influence of electrical stimulation on the network. This would serve as a valuable feature to track electrical stimulation efficacy. Here we propose a network analysis framework to classify brain-states involved in seizure genesis, track brain dynamics, and study the influence of electrical stimulation in the network.

Methods: The proposed framework was applied to SEEG recordings from 13 epilepsy patients during their presurgical evaluation visit in the EMU. The proposed algorithm uses a sliding window every 500 msec to create linear time-invariant models from SEEG signals. The linear models are stacked and analyzed to evaluate the nodal influence within the network. SEEG epochs from interictal and seizure activity were used to construct a network subspace. Network features extracted from the time-series are projected onto this network subspace to track brain dynamics. In this subspace, it is possible to visualize ictal/seizure zones and baseline/healthy zones. A multivariate-gaussian function was assigned to the data distribution, to track the current brain-state and its likelihood of entering the seizure state zone. Performance of our feature was compared against 21 state-of-the-art univariate and multivariate features. Performance was measured by maximizing separability between brain-states, interpretability within states, and robustness across subjects.

Results: Our results showed that we were able to identify source and sink nodes from SEEG recordings. The source-sink distribution across channels was used to define a baseline and an ictal brain-states involved in epileptogenesis. With our network subspace analysis framework, we can track brain-states in this feature space and show modulation towards seizure. Results showed that our network feature outperforms the rest of the features (19/21 with significance).

Conclusions: The proposed analytical algorithm provides a framework for tracking brain network dynamics that are sensitive to seizure genesis. This framework can be used by a model-based controller, in order to continuously track brain-states and deliver adaptive electrical stimuli to steer the brain-state from entering an ictal state, avoiding seizures altogether.

Funding: DE is supported by the HHMI Gilliam Fellowship. SS is supported by the NIH R21 NS103113.