Evaluating bio-inspired approaches for advance prediction of epileptic seizures

Epilepsy is the most common neurological disorder, affecting between 0.6% and 0.8% of the global population. During an epileptic seizure, the onset of which tends to be sudden and without prior warning, sufferers are highly vulnerable to harm, and methods that might accurately predict seizure episodes in advance are clearly of value. Building on recent work by Costa et al, we compare and contrast the sensitivity, specificity and accuracy of a selection of algorithms that attempt to predict the onset of epileptic seizures on the basis of 14 features extracted from electroencephalograph (EEG) monitoring data. We focus on how predictability varies as a function of how far in advance we are trying to predict the seizure episode, and also consider feature selection issues. We find that, using either a multi-class support vector machine (MC-SVM) or an evolved neural network (EANN), reasonable specificity and sensitivity can be achieved for prediction 8-10 minutes in advance. Indications are that the EANN performance is preferable for advance prediction, however the results so far do not support this with statistical significance. Meanwhile, we find that with a well-chosen reduced feature set (using mutual information), promising results can be obtained with only 8 of the 14 features. Further analysis showed that the accumulated energy in the signal, the maximum Lyupanov exponent, as well as measures of high-frequency signal components measured over short term windows, seem most promising for future research into accurate advance prediction models.