Abstracts

Quantitative analyses of intracranial EEG recordings from epileptic patients with temporal lobe epilepsy indicate that ictal and preictal states can be distinguished from seizure-free states for the applications of detection and prediction of seizures (IEEE Transactions on Biomedical Engineering, 50(5): 549-558, 2003; Lancet Neurology, 1(1):22-30, 2002). These findings suggested that it is possible to develop an implantable device for diagnostic and therapeutic purposes. In this study, we propose a new measure of signal regularity, pattern-match regularity score (PMRS), for the detection of EEG state changes, especially seizures. The measure is based on the estimation of signal pattern similarity. A major advantage of this measure is the ability to interpret it in both stochastic and chaotic models. This study tests the hypothesis that PMRS can distinguish state changes in intracranial EEG recordings. Intracranial EEG recordings obtained from 6 patients with a total of 81 medically intractable partial seizures were analyzed to test the hypothesis. PMRS was calculated for each EEG channel for each sequential 10.24-second non-overlapping data segment. The algorithm involves state space reconstruction, search for the pattern matched state vectors, and the estimation of pattern-match probabilities. The paired-T statistic was employed for each 10-minute sliding overlapping window to test the mean difference of PMRS values between two electrode sites. Electrode pairs were considered not entrained during any 10-minute period if the mean PMRS values were significantly different (p[lt]0.05). The PMRS and T-index curves were generated for the 1-hour time interval before and after each seizure. Significant changes observed in both PMRS and T-index curves were used for the detection of epileptic state changes in EEG recordings. Significant decrease of PMRS values during the ictal periods was observed in 91.4% of seizures. 87.7% of the preictal periods were detected by the presence of entrainment transition (gradual decrease in T-index values), and 87.7% of the seizures showed the postictal disentrainment with a rapid increase of T-index values after the end of a seizure. The results suggest that epileptic state changes can be detected by pattern-match regularity statistical analysis of EEG recordings from intracranial electrodes. Thus, it may be possible to predict and detect a seizure with this measure for clinical applications. (Supported by NIH Grant RO1EB002089 and Department of Veterans Affairs)