Patient Un-Specific Detection of Epileptic Seizures Through Changes in Variance

Despite much progress and research, fully reliable computer based epileptic seizure detection in EEG recordings is still elusive. This paper outlines a new strategy toward seizure detection. It is proposed that it is not the precise nature of a statistic that is important, but rather its variance over time. Using this, algorithms are presented that are able to successfully identify 97.6% of seizures from over 170 hours of recording and 15 different patients. False positives remain high, but virtually no pre-processing has been applied to the raw data and it is expected that this can be improved with further work