Abstracts

The measurement of EEG and fMRI is increasingly being used in the study of epilepsy, but the haemodynamic response to epileptiform activity remains incompletely characterised. In particular, it is not known whether the haemodynamic consequences of prolonged bursts of interictal spikes are equivalent to those of short bursts and to those provoked by normal brain function, or to what extent the assumptions of the general linear model (GLM), commonly used in fMRI studies, are upheld. Fourteen patients were selected from a database of ninety-one patients on the basis that bursts of focal or generalised interictal epileptiform activity of different durations were observed on the scalp EEG recorded during fMRI scanning. Echo-planar fMRI images were acquired in one of two 1.5T MR scanners (Vision and Sonata, Siemens, Germany; voxels 5x5x5mm, 25 slices, TE=50ms, TR=3s, flip angle 90[ordm]). EEG data were recorded with an EMR32 amplifier (Schwarzer, Germany) and 21 Ag/AgCl electrodes.
Nineteen data sets underwent statistical analysis, as 5 patients had more than one type of spikes. To determine whether inclusion of the event durations in the model resulted in higher statistical values, two maps were created by either including or ignoring the event durations.
To test the assumptions of the GLM, a region of interest of five voxels was defined for each data set around the voxel with the highest t value from either of the two maps. The haemodynamic response functions (HRFs) for bursts of different durations (i.e. 0-1 seconds, 2-3 seconds etc) were estimated and their amplitudes and latencies compared with the expected values from the GLM. In fifteen data sets at least one cluster was significant in both maps and in thirteen there was a mean increase in t value when durations were included. The mean increase across all data sets was 14.5% in peak t value and 29.5% in volume when including the duration of the events.
There were some consistent differences between the amplitudes of the measured HRFs and the GLM. In eight data sets the HRF for events of 0-1 seconds could be reliably estimated and the measured response was always larger than predicted. In the two data sets with the widest range of event durations, the measured amplitude increased with event duration without the plateau that was expected from the GLM. In the majority of data sets, the statistical significance of areas of activation increased when burst durations were included in the model. Analysing the fitted HRFs leads to some evidence of non-linearity, but is generally consistent with the observation that including the duration of the bursts leads to a more accurate model than considering each burst as an instantaneous event. (Supported by CIHR grant MOP 38079. CGB was funded by a CIHR doctoral research award. EK was funded by a Preston Robb Fellowship from the MNI.)