Seizure detection in adult ICU patients based on changes in EEG synchronization likelihood

Seizures are common in Intensive Care Unit (ICU) patients, and may increase neuronal injury.

To explore the possible value of synchronization likelihood (SL) for the automatic detection of seizures in adult ICU patients.

We included EEGs from ICU patients with a variety of diagnoses. The gold standard for further analyses was the consensus judgment of three clinical neurophysiologists who classified 150 scalp EEG epochs as "definitely epileptiform," "definitely non epileptiform," or "uncertain." SL estimates the statistical interdependencies between two time series, such as two EEG channels. We computed the average synchronization by calculating the SL between one channel and every other channel, and taking the mean of these values.

The mean SL in the 38 "definitely epileptiform" epochs ranged from 0.095 to 0.386 (mean 0.189; SD 0.066). In the 34 "definitely nonepileptiform" epochs the mean SL ranged from 0.087 to 0.158 (mean 0.115; SD 0.016; p < 0.0005). The area under the ROC curve was 0.812 (95% Confidence Interval 0.725 to 0.898).

The mean SL may distinguish between seizure and nonseizure epochs, and may prove helpful to monitor epileptic activity in ICU patients.