Studying Spontaneous Brain Activity Using Eeg-fmri and Event-related Ica

Introduction: Simultaneous EEG and fMRI recordings (EEG-fMRI) can detect haemodynamic changes associated with spontaneous events observed on the EEG, such as interictal epileptiform discharges (“spikes”). Event-related analyses of these studies typically assume a canonical HRF model and define the start of the event as the onset of EEG changes. There have been two recent reports, however, of BOLD signal changes starting several seconds before the appearance of an epileptiform spike on the EEG [1,2], which suggests there may be related brain activity, undetected by the EEG, that occurs prior to the spike. The analysis strategy adopted in both those reports was to retain the assumption of a canonical HRF model and simply shift the event-onset backward in time relative to the spike – an approach that may poorly represent the unobserved underlying activity that leads to the BOLD signal changes preceding the spike. For example, the unobserved activity may be sustained for several seconds or could be a slow network oscillation, which would result in evoked BOLD signal changes that are quite different from a standard HRF model shifted in time relative to the spike. In this abstract we describe a new event-related independent components analysis (eICA) method for detecting BOLD signal changes associated with spontaneous EEG events. This data-driven analysis method does not rely upon a pre-defined HRF to model the expected response, which makes it ideal for detecting BOLD signal changes associated with unobserved activity that may precede the event observed on the EEG. We demonstrate this new method on an EEG-fMRI study of patients with benign epilepsy with centrotemporal spikes (BECTS). Methods: eICA – Model: The fMRI signal, Y, acquired during an EEG-fMRI experiment is represented as a n t × matrix: the rows contain the spatial maps of BOLD signal intensities sampled at n voxel locations and the columns contain the voxel time-courses sampled at t time-points. The event-related signal, B, models the BOLD signal changes associated with a single isolated event and is represented as a n p× matrix of whole-brain BOLD signal