A Novel Method for Automatic Detection and Visualization of Non-Convulsive Seizures in ICU (P04.016)

Abstract

Objective: To develop a method for automatically detecting non-convulsive seizures in ICU. Background Seizures are found in at least 8% of ICU patients; of these, 50-100% have only non-convulsive seizures detectable only by continuous EEG monitoring. Non-convulsive seizure duration and delay to treatment have been shown to be associated with increased mortality [1]. Seizure detection and management is problematic: a skilled electroencephalographer is not always available, generating a need for an automated seizure detection system. Design/Methods: Using a database of 179 ICU patients undergoing continuous EEG (148d) monitoring, we have developed a method that quantifies seizure-related evolution in a two-dimensional feature space consisting of EEG frequency and power. The method was evaluated in an independent data set of 53 ICU and epilepsy monitoring unit (EMU) patients (50d) [2]. All data were analyzed by an experienced electroencephalographer (BY), who marked the onsets and offsets of ten first seizures in each recording using commonly accepted seizure criteria [1]. Seizure detection was classified successful, if it took place within 15 minutes of the first marked seizure. Results: 10/19 seizure patients of the development data set were detected (false positive rates (FPR): median 0.00/hour, mean 0.34/hour). In the evaluation data set 6/8 seizure patients were detected (FPRs: 0.06/hour, 0.52/hour). Undetected patients were either in status epilepticus (SE) or had no seizures with evolutionary characteristics. Most false detections were caused by EMG. Conclusions: The method demonstrates good potential in detecting seizures with evolution, which further can be visualized in 2-D plane, facilitating seizure recognition at the bedside. To cover the full range of seizure disorders, the method needs to be integrated with methods suitable for spike detection and SE monitoring, such as WSE [3]. Further work will focus also on reducing FPR. Supported by: GE Healthcare, Finland OY. Disclosure: Dr. Tanner has received research support from GE Healthcare. Dr. Sarkela has received research support from GE Healthcare. Dr. Tolonen has received personal compensation for activities with GE Healthcare as an employee. Dr. Norton has nothing to disclose. Dr. Davies-Schinkel has nothing to disclose. Dr. Sharpe has nothing to disclose. Dr. Young has nothing to disclose.