Abstracts

Rationale: Simlutaneous EEG-fMRI recording of seizures suggest it may be a useful technique to predict the ictal onset zone. However, scalp EEG used to model events in this technique may be insensitive to seizure related neuronal activity and the events themselves are unpredictable, variable and difficult to record successfully.. We studied the BOLD patterns associated with epileptic seizures in EEG-fMRI using EEG-based modelling and data-driven methods. Methods: 83 patients with focal epilepsy were scanned using EEG-fMRI and in 9, seizures were recorded. (identified by visual observation and EEG). Two analyses were performed using general linear models (GLM) derived from scalp EEG: 1) each seizure was divided into blocks corresponding to electrographic onset, clinical seizure activity and post ictal EEG change and convolving each with a haemodynamic response function (HRF); 2) seizures lasting 3 scans or more were also modelled using a Fourier basis set over their duration allowing for inter-regional variations in the BOLD signal time course. A third approach using independent component analysis (ICA) of the fMRI data was carried out to identify ictal BOLD patterns. In all 3 approaches, spatial patterns of BOLD activity were compared with electroclinical data including intracranial EEG where available. Results: In 7/9 patients the canonical GLM analysis revealed significant BOLD signal change associated with ictal activity on EEG in 7/9 patients. In 5/7 a region of significant BOLD change was found to be concordant with the seizure onset zone identified on intracranial recording. In 2/7 patients exhibiting significant BOLD signal change, the pattern was discordant with electro-clinical data. The GLM Fourier model applied in 3 cases revealed temporal changes in BOLD signal correspondingl with the original analysis in 2/3 cases. ICA revealed components which were spatially concordant with electro-clinical data in all patients. Conclusions: These results demonstrate that ictal EEG-fMRI analysed using conventional GLMs is most useful where the seizure can be accurately modelled from the scalp EEG and is often concordant with invasive recordings. . However, data driven techniques such as ICA provide useful information about haemodynamic changes in the seizure onset zone, and may also be a useful adjunct where the scalp EEG is less clear. A combination of approaches in the analysis of EEG-fMRI may provide localising information regarding seizure onset and may help in the planning of invasive recordings in patients undergoing epilepsy surgery.