Independent component analysis of simultaneously acquired electroencephalography and functional magnetic resonance imaging in focal epilepsy patients.

摘要

Simultaneous electroencephalography and functional magnetic resonance imaging (EEG-fMRI) is a potentially useful diagnostic modality for the presurgical evaluation of patients with refractory focal epilepsy. The goal of EEG-fMRI is to localize hemodynamic correlates of epileptic discharges. The data is usually analyzed in the general linear model (GLM) framework, which assumes that the electrical and hemodynamic signals are coupled by a fixed canonical hemodynamic response function (HRF). However, the HRF is known to show some variability that may affect the sensitivity of the method. Investigating this variability may reveal additional useful information about the data.;First, the performance of the proposed method is evaluated on simulated activations under a wide variety of realistic conditions. It is also shown that the GLM framework may fail to detect activations if the HRF varies only slightly from the canonical shape. The method is then applied to recordings of epileptic seizures. The GLM analysis typically yields very widespread areas of activation, but the ICA method can decompose these areas into multiple clusters with various HRF peak delays. Clusters with early HRF delays correspond well with presumed seizure onset regions, while other clusters may be related to seizure propagation. Finally, the method is used to investigate the variability of the HRF amplitude in response to interictal epileptiform discharges (IED). It is shown that clusters of activation in the presumed epileptogenic focus show a significant correlation between HRF amplitudes and IED amplitudes, unlike clusters in distant regions. Therefore, the method can improve the specificity of the EEG-fMRI analysis.;This thesis presents a method to analyze EEG-fMRI data independently of a prior HRF model, using independent component analysis (ICA). With minimal prior assumptions, ICA can decompose the fMRI into components representing the major fluctuations present in the data. A deconvolution method then identifies components showing significant signal changes related to the epileptic discharges, independently of the HRF shape.