Abstracts

Rationale:
Understanding what drives the generation of seizures and finding markers of approaching seizure onset remain obstacles to seizure prediction and therapeutic or adaptive responses to an upcoming seizure event. Seizures likely result from aberrant network activity and synchronization. Previous studies have shown that persistent interictal changes in brain network connectivity estimated from intracranial electroencephalography (iEEG) are associated with the epileptogenic zone (EZ) and that measures of connectivity like betweenness centrality can predict surgical outcomes in resective epilepsy surgery. In this study, we use rapid network model estimation to show that significant changes in network structure peri-ictally can be used to predict seizure onset well before it occurs.
Method:
iEEG data from 5 temporal lobe epilepsy (TLE) patients from the iEEG.org database were included in this study. Criteria for inclusion were seizure freedom post-surgery (ILAE Class I) and that seizures were clearly defined and spatially limited, as defined by clinical notes. Recordings for each patient were common mode average referenced and bandpass filtered (0.5-250Hz). For each clinical seizure, the 2 minutes prior to seizure onset were divided into 5 second epochs. In each window, a multiple input, single output (MISO) state space model was estimated for each channel output with all other channels as inputs. The magnitude of the parameters describing the influence of each other channel on the given channel were used to infer a directed network graph of the relationship between all channels for the duration of the window. The structure and characteristics of the resulting networks were assessed using degree and betweenness centrality and were analyzed across seizures and patients. This procedure was repeated for interictal data as a non-seizure control.
Results:
Figure 1 shows the degree and betweenness centrality of the total network, EZ, and non-EZ channels at significant timepoints prior to seizure onset, aggregated across patients and seizures. Both degree and betweenness of EZ channels increase significantly as early as 30 seconds prior to seizure onset and these centrality metrics continually increase until seizure onset. Additionally, total network degree rises significantly in the window just prior to seizure onset as the number of connections between the EZ and surrounding channels grows dramatically. Together, these findings allow us to identify a rising likelihood of seizure occurrence up to a probable “point of no return”, prior to the actual onset of seizure. Networks estimated on interictal data showed low overall connectivity and no increase in connectivity over time.
Conclusion:
MISO model-based rapid network estimation was used to identify alterations in iEEG network characteristics prior to seizure onset. Significant differences in EZ connectivity as a seizure approaches allows for early prediction of seizure onset and enables therapeutic or protective interventions, which could improve patient safety and quality of life.
Funding:
:This work was supported by intramural research funding from the Department of Neurology and University of Connecticut School of Medicine.