Multivariate analysis of SEEG signals during seizure

Epilepsy is a neurological disorder that affects tens of millions of people every year and is characterized by sudden-onset seizures which are often associated with physical convulsions. Effective treatment and management of epilepsy would be greatly improved if convulsions could be caught quickly through early seizure detection. However, this is still a largely open problem due to the challenge of finding a robust statistic from the neural measurements. This paper suggests a new multivariate statistic by combining spectral techniques with matrix theory. Specifically, stereoelectroencephalography (SEEG) data was used to generate a series of coherence connectivity matrices which were then examined using singular value decomposition. Tracking the relative angles of the first singular vectors generated from this data provides an effective way of defining the most dominant characteristics of the SEEG during the normal, the pre-ictal, and the ictal states. This paper indicates that the first singular vector has a characteristic direction indicative of the seizure state and illustrates a data analysis method that incorporates all neural data as opposed to a small selection of channels.