Quantitative analysis of EEG for the detection and localization of seizures in patients with medically refractory epilepsy

Visual interpretation of the EEG associated with an epileptic seizure (ictal EEG) can be cumbersome and sometimes misleading, due to rapid seizure spread, limited spatial resolution and the fact that signals of interest are often obscured by wideband noise and/or artifacts. The development of useful quantitative tools to assist the clinician presents a significant technical challenge requiring the application of modern signal processing methodology. This thesis describes the design and testing of new analytic methods for the detection and localization of both scalp- and intracranially-recorded seizures. The methods were designed with the aim of providing the clinician with tools that: (1) improve the noninvasive evaluation of surgical candidates, thereby reducing the need for intracranial EEG monitoring; (2) enhance the planning of appropriate electrode placement for intracranial EEG monitoring; and (3) assist in localizing and tracking the spread patterns of intracranial seizure records.; Time-frequency representations (TFRs) based on the Fast Fourier Transform (FFT) are often inadequate for analysis of seizures, which vary rapidly over time and may be heavily contaminated by artifacts. Using simulated and actual ictal EEGs, the performance of several alternative TFRs was studied. The exponential distribution (ED) was shown to be superior to the other TFRs and was subsequently used to evaluate band-limited spectral measures of the ictal EEG with respect to a preictal baseline. Based on this, decision rules for scoring scalp- and intracranially-recorded seizures were designed and tested on sets of ictal EEGs from patients with verifiably localized seizure foci, and with the same seizures being scored by an expert reader. For scalp-recorded seizures the quantitative method was able to perform significantly better than the expert in terms of correct lateralization, while the expert performed better in terms of correct localization of seizures. For intracranially-recorded seizures, performance was excellent in terms of correct lateralization and localization of seizures. In an effort to improve performance of scoring scalp-recorded seizures, a wavelet-based approach that utilizes the time-scale representation (TSR) in a detection method similar to the TFR-based approach was then designed and tested on the same ictal EEGs. The resulting wavelet-based method allowed a further substantial improvement in computer scoring relative to expert scoring.