An automatic warning system for epileptic seizures recorded on intracerebral EEGs
Introduction
Long term electroencephalogram (EEG) monitoring is required to analyze and characterize seizures and other epileptic activity. This can often be prolonged, lasting several days or weeks, and can become time consuming and expensive, as it is difficult to have enough personnel available to monitor patients at all times or to review all EEGs. As a result, automatic seizure detection systems have become an important component of EEG monitoring because they improve the cost of prolonged monitoring by reducing the amount of data that need to be reviewed (Gotman et al., 1997). In addition to detection systems, warning systems have also become increasingly valuable as detections of seizures at an early stage cannot only warn the patient that a seizure is occurring, but also alert medical staff, and allow them to perform behavioural testing to further asses which specific functions may be impaired as a result of a seizure and help them in localizing the source of the seizure activity.
The task of designing a reliable automatic seizure detection and onset warning system is difficult. The systems not only require high sensitivity, but low false detection rates and short detection delays as well. Although numerous attempts have been made, their performances are far from ideal and can use significant improvement. Currently, the method most commonly used clinically for scalp and depth electrodes is that of Gotman, 1982, Gotman, 1990 with an overall sensitivity of 75.8% for both scalp and depth electrodes and a false detection rate of 0.84/h for scalp EEG and 1.35/h for depth EEG. A newer detection and onset warning system, designed by Saab and Gotman (in press), for scalp electrodes is based on determining the seizure probability of a section of EEG using Bayesian formulation. Designed specifically for clinical use, this system features a user-tuneable threshold, which allows for trade-off between sensitivity, detection delay, and rate of false detection. Results indicate a sensitivity of 76%, a false detection rate of 0.34/h and a median detection delay of 10 s.
The premise for the seizure detection and onset warning system proposed here is the work described by Saab and Gotman (in press). However, it is specifically designed for use with intracerebral electrodes. Differences between the two algorithms come into effect because of the differences between the two types of recordings. Unlike scalp EEG recordings, intracerebral recordings are relatively free of technical artefacts but contain seizures of a greater variety of morphologies, and lower and higher frequencies. In addition, the intracerebral EEG has a lot of non-epileptic activity with prominent rhythmic characteristics.
Section snippets
Data selection
The depth EEG data used in this study were all collected using the Stellate Harmonie System for EEG monitoring from the Epilepsy Telemetry Unit at the Montreal Neurological Institute and Hospital. Once filtered between 0.5 and 70 Hz, data were sampled at 200 Hz. All patients had predominantly stainless steel depth electrodes of nine contacts surgically inserted inside the brain with contacts located 5 mm apart. Some patients also had epidural peg electrodes. Patients were not pre-screened, but
Results
This section will present the results obtained for the training and testing data. Results will also be presented for the Gotman (1990) system, which is commonly used.
Results will be presented on a per-patient basis and as averages for all patients, to avoid biasing by one particular patient. There were three measures of performance: false detections, delay times and sensitivity. False detections were events identified by the automatic detection system but not by the EEG experts. Since the
Discussion
The purpose of this study was to develop an online automatic seizure detection and warning system for clinical use in patients with intracerebral electrodes. The goal was to design a system with high sensitivity, low false detection rate, and low delay times resulting in early detections. Unlike past systems, which employed patient-specific algorithms, this work was designed to run independently of patient-specific seizure data and no patient training at run time. Furthermore, rather than
Acknowledgements
This work was supported by grant number MOP-10189 from the Canadian Institutes of Health Research. J. Gotman is a major shareholder of Stellate, a company that makes equipment for long-term monitoring in epilepsy.
References (18)
- et al.
Very high-frequency rhythmic activity during SEEG suppression in frontal lobe epilepsy
Electroencephalogr Clin Neurophysiol
(1992) Automatic recognition of epileptic seizures in the EEG
Electroenceph Clin Neurophysiol
(1982)Automatic recognition of epileptic seizures in the EEG
Electroenceph Clin Neurophysiol
(1990)- et al.
Automatic seizure detection in the newborn: Methods and initial evaluation
Electroenceph Clin Neurophysiol
(1997) An automated seizure monitoring system for patients with indwelling recording electrodes
Electroenceph Clin Neurophysiol
(1993)- et al.
Wavelet-based automatic seizure detection in intracerebral electroencephalogram
Clin Neurophysiol
(2003) - et al.
Anticipation of epileptic seizures from standard EEG recordings
Lancet
(2001) - et al.
How well can epileptic seizures be predicted? An evaluation of a nonlinear method
Brain
(2003) - et al.
Hippocampal and entorhinal cortex high-frequency oscillations (100–500 Hz) in human epileptic brain and in kainic acid-treated rats with chronic seizures
Epilepsia
(1999)