A temporally constrained spatial ICA for separation of seizure bold from FMRI

Application of spatial Independent Component Analysis (ICA) to functional magnetic resonance imaging (fMRI) subject to the simultaneously recorded electroencephalography (EEG) signals as constraint, has been investigated in this work. In this novel approach, the closeness between the time course of spatial independent components of fMRI and EEG signals during epileptic seizure period is introduced as the constraint to the separation process. The performance of the algorithm has been tested on a set of simultaneous EEG and fMRI data and the results show a more accurate localization of the blood-oxygenated level-dependence (BOLD) regions, better algorithm convergence, and a higher correlation between the time course of spatial components and the seizure EEG signals than the conventional ICA method.

1. Introduction

As one of the advanced brain function monitoring modalities, fMRI provides high resolution spatial information which helps visualization of the brain activation. Blood-oxygenated level-dependence (BOLD) regions in fMRI result from event-related, movement-related, and abnormal brain activities such as seizures. The most widely used approach for fMRI data analysis was developed by Friston et al. (1995) [1], which is based on the general linear model (GLM). Based on this model, the prior knowledge or specific assumptions about the time courses contributing to the signal changes are required for model specification. Therefore, the appearance of false BOLD regions or lack of BOLD in places where the right stimulus cannot be modelled is inevitable by using this technique. This can happen in seizure detection and localization using fMRI, when the ictal duration is short and the seizure onset is difficult to be snecified.