Prior knowledge on cortex organization in the reconstruction of source current densities from EEG

The reconstruction of the generators of electroencephalographic (EEG) signals is important for understanding brain processes. Since the inverse problem has no unique solution, additional knowledge or assumptions are needed. Often, results from other anatomical or functional measurement modalities are difficult to interpret directly in terms of EEG source strengths, but they provide valuable information about the functional similarity between brain regions, for example, in form of parcellations. We propose a novel approach to the incorporation of such parcellations as priors into the reconstruction of distributed source current densities from EEG. Two algorithms are described, based on a surface-constrained LORETA (Low Resolution Electromagnetic TomogrAphy) approach. The first, patchLORETA1, uses both topological neighborhood and prior information to define smoothness, while the second, patchLORETA2, neglects topological neighborhood. Computer simulations, using a smooth reconstruction surface on the brain envelope, reveal important aspects of the algorithms' performance, in particular the influences of noise and incongruence between measurements and prior information. It turns out that patchLORETA1 makes efficient use of the provided prior information and at the same time is quite robust towards faulty priors as well as noise. The algorithms are also tested on the localization of the sources of event-related potentials. Here, both the smooth brain and folded cortical surfaces serve as reconstruction spaces. We find that patchLORETA1 becomes ineffective on the folded cortex, while patchLORETA2 yields plausible results. We also discuss the extension of the proposed algorithms to other types of priors and propose ways to overcome shortcomings of the current implementation.