Abstracts

Rationale: Today millions of scalp electroencephalography (EEG) are performed each year to aid in the diagnosis and treatment of epilepsy. Despite nearly 100 years of utilizing routine EEGs, epileptologists still primarily evaluate scalp EEGs by visual inspection. The diagnosis of epilepsy depends on a comprehensive clinical history, neurological examination, and axillary studies including scalp EEG. However, the sensitivity of a routine 30-minute scalp EEG for detecting abnormalities indicating epilepsy, such as interictal (between seizure) epileptiform discharges (IEDs) varies from 29-55%. Overreliance on and misinterpretation of routine EEGs has been found to be a significant factor in the nearly 30% misdiagnosis rate of epilepsy, contributing to approximately $1 billion to $6.4 billion in annual costs in the US.

Methods: We are developing a novel EEG algorithm that predicts if a patient is epileptic when being monitored “at rest” and does not require IEDs or other abnormalities to be present. The key innovations are (i) the use of a dynamic network model (DNM) and (ii) our “source sink” hypothesis that states when an epilepsy patient is not having a seizure, it is because the seizure focus/foci is being inhibited by neighboring regions. The novel algorithm identifies two groups of nodes from the EEG data: those that are continuously inhibiting their neighbors (“sources”) and the inhibited nodes themselves (“sinks”), the latter which correspond to the seizure focus/foci. Specifically, patient specific DNMs are estimated from EEG data and their connectivity properties reveal strong sources and sinks in the network when a patient has epilepsy, and no strong sources or sinks otherwise.

This approach captures how nodes in a network dynamically influence each other, while existing clinical approaches rely on reading the EEG with the naked eye.

Results: We tested our algorithm on 24 patients evaluated at the Johns Hopkins Epilepsy monitoring unit (EMU). The diagnosis was based on an EMU admission which recorded the patients’ habitual clinical events. Our algorithm was tested on each patient’s first EEG obtained at JHU. Of those patients, 8 had epilepsy with IEDs, 8 had epilepsy with no IEDs, and 8 did not have epilepsy. Our conjecture was that a source-sink phenomenon occurring inside the brain could be identified by a significant source-sink activation (above a threshold) on the scalp; and source-sink activation below the threshold at the scalp for non-epilepsy subjects (see Figure 1). We found significant source-sink activation for 7/8 epilepsy patients with no IEDs in their scalp EEG in the 30-minute window used to diagnose epilepsy (sensitivity of 87.5%). In contrast, the 7/8 individuals without epilepsy showed heatmaps with ss-activation below the threshold (87.5% specificity). All epilepsy patients with IEDs have ss-activation above the threshold (see Figure 1 bottom right).

Conclusions: Our approach has the potential to dramatically change clinical practice, reducing social and economic costs associated with misdiagnosis of epilepsy.

Funding: Please list any funding that was received in support of this abstract.: This work was supported by the Maryland Innovation Initiative.