Quantifying spatio-temporal dependencies in epileptic ECOG

There is evidence that the mechanisms leading to epileptic seizures can be understood by continuously tracking the ongoing spatio-temporal mappings in the brain. We propose to quantify the spatio-temporal dynamics by using a SOM-based similarity index (SI) measure. While it is shown that this measure is statistically as accurate as the original SI measure, it is also computationally faster and therefore applicable for real-time analyses. In order to quantify the changes of SI in electrode space and along time, a spectral clustering approach is employed by interpreting the SOM–SI values as affinity matrices. Preliminary analyses on spatial mappings of multivariate epileptic ECOG data are presented using the modified spectral clustering approach. Results involving two pairs of seizures of an epileptic patient suggest that patterns associated with channels’ spatio-temporal dynamics during the inter-ictal to pre-post ictal transition vary from seizure to seizure.