ICA Component Selection Based on Sparse Activelet Reconstruction for fMRI Analysis in Refractory Focal Epilepsy

EEG-fMRI is a recently emerging tool that can be used in the presurgical evaluation of focal epilepsy patients. Standard analysis techniques rely on the principle that fMRI can provide accurate localization of hemodynamic changes corresponding to events observed on EEG. However, its applicability is limited as EEG does not always provide sufficient and reliable information on the timing of the epileptic activity. Therefore, there is an increasing demand for techniques capable of localizing the epileptic activity based solely on the fMRI time series. Independent component analysis (ICA) has been shown to separate epileptic activity in the fMRI from other neural sources, artifacts and noise. We propose here to automatically detect the epileptic component based on sparse reconstruction in the activelet basis. The algorithm was evaluated on a dataset of 10 patients. It is shown that the largest activation cluster of the identified component overlapped with the ictal onset zone (IOZ) in all 3 patients with sparse interictal spike timing. In the 7 other patients, the selected component either overlapped with the IOZ and/or the ictal hyperperfusion, or correlated with the EEG-derived time course of the interictal activity. We conclude that the proposed technique might be able to identify epileptic components without using EEG.