Abstracts

Source analyses of interictal and ictal scalp EEG often implicate involvement of multiple cerebral regions. Fast neuroelectric activity emitted from each region may be estimated on a macroscopic spatial scale using a local source estimator such as regional activity estimation (REGAE). From such derived time series, we aim to make causal inferences about (dys)functional connections between regions around the times of interictal or ictal onset discharges. We developed a new method to study cerebral connectivity and influences among regions involved with epileptogenicity. The general method has three main steps: (i) transform scalp EEG time series to brain regional activity time series via REGAE; (ii) for each region A and time t, calculate a set of one-sided time derivatives that characterize the dynamic state of A at time t; and (iii) for each pair of brain regions A and B, time t, and lag u, compute [italic]time-lagged causally predictive information[/italic], CPI[sub]AB[/sub](t,u), an information theoretic measure of the degree to which the states of A in a time window positioned at time t-u predict the states of B in a time window positioned at time t, taking into account the possibility of mediating influences (e.g., the mediation of a third region C). CPI[sub]AB[/sub](t,u) has linear and nonlinear variants (previously studied using simulated data with known influences at given lags). To evaluate this technique, we applied CPI analysis to scalp ictal EEG recorded from 57-channels (10-10 system) obtained on two seizures recorded on two consecutive days from a temporal lobe epilepsy patient. Starting with 80 regions of interest (ROIs) that collectively covered brain gray matter (obtained from an MRI segmentation), a subset of 8 temporal lobe ROIs were selected as the potential underlying generators based on the magnitude of their estimated activity. Linear CPI[sub]AB[/sub](t,u) was computed using 1 s time windows for all pairs of ROI candidates for both seizure onsets, and peak lead-lag relationships were examined across time. Results comprise a matrix of 28 graphs of CPI as a function of time (s) and lead-lag (ms). CPI[sub]AB[/sub](t,u) disclosed consistent magnitude and lead-lag relationships among the 8 temporal regions both within and between the two seizure onsets. In particular, 2 of the 8 candidate ROIs demonstrated more causal influence than the others (with leads of 5 to 10 ms). These two regions were located in the antero-medial and infero-lateral posterior temporal areas. CPI measures dynamic state predictability of one region with respect to another after accounting for causally confounding influences, and may provide clinically useful information about ictal rhythm onset and propagation sources. An inherent limitation is that regions unobservable by scalp EEG (such as deep brain structures) cannot be included as confounds. Larger series will be required for further evaluation of clinical utility. (Supported by NIBIB 1 R43 EB000614)