Epileptic seizures are unpredictable events that may occur only rarely, hence the requirement for long-term video-EEG monitoring in the diagnosis of epilepsy. A clinical neurophysiologist can then periodically review the EEG recordings and analyze the seizures that may have occurred during the monitoring session. However, reviewing a continuous EEG recording lasting several days can be a very time-consuming process. In practice, the patient can indicate that a seizure occurs through the use of an alarm button, so that only the recording sections surrounding the press of the button need to be analyzed. Unfortunately, there are many cases where patients are not aware of the occurrence of their own seizures. An automated seizure detection system can thus be of great use in identifying EEG sections that need to be reviewed.
The main difficulty with this task lies in the wide variety of EEG patterns that can characterize a seizure, such as “low-amplitude desynchronization, polyspike activity, rhythmic waves at a wide variety of frequencies and amplitudes, and spikes and waves” (Gotman, J Clin Neurophysiol, 16(2), 130-140, 1999).
Methods for automatic detection of seizures may rely on the identification of various patterns such as an increase in amplitude, sustained rhythmic activity, or EEG flattening. Various algorithms have been developed based on spectral or wavelet features, amplitude relative to background activity, and spatial context. These features can then be combined in a decision tree, an artificial neural network, or a Bayesian framework to identify the occurrence of seizures.
It is crucial for seizure detection systems to have a high sensitivity to seizures, even if this results in a large number of false detections. Such systems can then be used to considerably reduce the amount of data that need to be reviewed; a neurologist can then easily discard false detections.
In contrast to seizure detection systems applied on long-term recordings, seizure warning systems have been developed to detect seizures in real time as soon as possible after their onset. A small detection delay could allow patients to take appropriate measures, such as sitting down to avoid injuries, even before they are themselves aware that a seizure has begun. It could also be possible to administer treatment such as electrical stimulation or drug injection to stop the evolution of the seizure. There are also many types of seizures for which very few clinical signs are observable, but for which a better diagnosis could be performed by questioning the patient early during the seizure. An early detection could also be useful for the purposes of an ictal SPECT scan, which should be performed as close as possible to the onset of the seizure. It would be more important for a seizure warning system to have a high specificity rather than a high sensitivity. Clinical staff would probably start ignoring warnings by the system if there were many false detections. We have developed adaptive seizure warning systems incorporating information about spatial context and clinical state of the patient to reduce the number of false detections.