Fully Automated EEG Source Imaging Using Structured Sparsity for Single and Multiple Synchronous Epileptic Activities

Abstract

Accurate localization of epileptic zones from High-Resolution ElectroEncephaloGraphy (HR-EEG) data can be challenging, especially when multiple synchronous zones are involved, and is highly dependent on the chosen EEG Source Imaging (ESI) method. Since a given scalp-level electrical pattern can result from multiple source configurations, ESI methods address this ill-posed inverse problem by imposing constraints on the structure of underlying sources. Here, we present an efficient approach that imposes sparsity on both the source-level activity and its spatial gradient. Unlike other methods that generally require a heuristic choice of a regularization parameter that balances between data fidelity and constraints, our method iteratively adjusts the parameter value based on the noise level in a fully automated way. The performance of the new method is evaluated across different scenarios of realistic synthetic HR-EEG data, including unifocal and synchronous multifocal cortical epileptic activity. Based on multiple performance indices, we demonstrate that the proposed approach outperforms traditional methods in accurately reconstructing epileptic sources. We also show that the method reduces polarity artifacts responsible for ghost sources and spatial discontinuities. Its ability to recover homogeneous, well-delineated regions of activity is further confirmed using real EEG data capturing a typical absence seizure.