Abstracts

Rationale: Epilepsy is a neurological disorder that affects over 60 million people worldwide and is characterized by sudden-onset seizures. The unpredictability of the seizures is one of epilepsy's most debilitating features. Effective treatment and management of epilepsy would be greatly improved if seizures could be detected as early as possible. However, early seizure detection is still largely an open problem due to the challenge of extracting a powerful statistic from neural measurements. Methods: We describe a new multivariate statistic that combines information theory and matrix analysis, and incorporates information from all recording electrodes at each time. The data analyzed were long intracranial electrocorticographic recordings (ECoG) from 3 human patients with epilepsy. These measurements contained 38 channels of simultaneously recorded data, spatial distributed over cortex and were recorded continuously over a period of 2-3 days. A series of time dependent connectivity matrices were formed by calculating the pairwise mutual information between all sites in the ECoG data before, during and after seizure. Each element of the connectivity matrix is a measure of pairwise dependence but at each time the entire multivariate matrix structure over all electrodes was analyzed. Singular value decomposition (SVD) was used to identify the leading order structure in these connectivity matrices by tracking the first singular vector at each time. The time evolution of the recordings was quantified by calculating the relative angles (inner product) between the first singular vectors at different times periods with the mean inter-ictal first singular vector.Results: We find the first singular vector has a smaller projection onto the inter-ictal vector during seizure, indicating the seizure state (ictal) vectors have a characteristic direction that is different from normal brain states (inter-ictal). The weighting of the entries in the singular vectors are a measure of the strength of connectivity of the corresponding electrodes. The change in direction of the first singular vector during seizure indicates the reorganization of connectivity during seizure relative to inter-ictal states. Conclusions: We conclude the direction of the first singular vector is capable of characterizing seizure activity and may be a statistic that can be used for early onset detection of seizures. Furthermore, the structure of the first singular vector during seizure can be used to identify the connectivity associated with seizure and may be used for seizure focus localization.