Abstracts

Rationale: Typical childhood absence seizures (CAS) are brief 5-10 second episodes, accompanied by short-term impairment of consciousness, and by 3Hz spike-and-wave discharges which begin and end abruptly on EEG. Recently, functional magnetic resonance imaging (fMRI) with simultaneous EEG recording has been instrumental for our current understanding of the anatomical and functional basis of CAS. These EEG-fMRI studies typically have used a general linear model (GLM) with a pre-defined hemodynamic response function (HRF) to assess dynamic changes of blood-oxygenation-level-dependent (BOLD) signal in whole brain, which reflects neuronal activation, albeit indirectly. However, there is some important evidence that the actual hemodynamic response for diverse brain regions may differ from the standard HRF. Methods: We performed a model-free clustering approach to investigate BOLD changes during 51 CAS in 8 pediatric patients with typical childhood absence epilepsy (CAE). This approach has three steps, 1) the time courses of single voxels during the time period from -20s to 40 s relative to seizure onset were averaged across patients. 2) 116 gray matter anatomic volumes of interest (AVOI) were pre-defined from the SPM2 MRI template (MARSBAR), and the temporal correlations between pairs of mean time courses of AVOIs were computed. The number of expected clusters was determined by analyzing these correlations using a hierarchical clustering method. 3) Correlations between each pair of time courses of voxels were computed. Then we performed a k-mean method to create partitions of voxels exhibiting similar time courses, using the number of expected clusters obtained from step 2. Results: We found that 116 AVOIs can be divided into four clusters by using the hierarchical clustering method. By applying the k-mean method, the partition of areas into clusters emerged as follows: 1) thalamus and occipital cortex; 2) lateral and part of medial parietal cortex, medial temporal, basal ganglia, and cerebellum; 3) part of medial frontal, lateral frontal, orbital frontal, and rolandic cortices; 4) part of medial frontal, medial parietal, medial temporal cortices and insula. Furthermore, we observed that fMRI change of the cluster involving thalamus and occipital cortex was closer to the conventional HRF, and changes of the other three clusters appeared to differ greatly from the conventional HRF. Conclusions: Our results demonstrate a complex sequence of fMRI changes in absence seizures, which are not detectible using conventional HRF modeling. These results also revealed that current clustering methods can effectively identify regions of similar activation. Finally, our present findings suggest that the clustering method might be a very useful tool for analysis of activation patterns in fMRI for other types of generalized seizures.