Optimized application of double and single layer BEM for in vivo conductivity estimation

TL;DR

This paper improves the efficiency of in vivo conductivity estimation in EEG source localization by optimizing boundary element method computations, enabling faster and more accurate subject-specific head models.

Contribution

It introduces a Woodbury update algorithm for BEM that significantly accelerates conductivity estimation and derives new closed-form expressions for BEM matrix elements.

Findings

Speed gain up to 20 times compared to direct computation

Applicable to both single and double layer BEM formalism

Facilitates routine use of subject-specific head models in EEG

Abstract

Inter subject variability of the electrical conductivity of brain, skull and skin strongly limits the accuracy by which current sources underlying electro-encephalography (EEG) can be localized in the brain. This inter subject variability also constrains the possibility to use EEG amplitude parameters as a biomarker to compare the amount of neural activity between different patients. To overcome this problem, one may estimate conductivity parameters in vivo by analyzing the potentials generated by known electric currents, injected into different pairs of EEG electrodes. At present, routine application of this approach is hampered by the computational complexity of the conductivity estimation problem. Here we analyze the efficiency of this conductivity parameter estimation problem in the context of boundary element method (BEM). We assume geometries of brain, skull and skin compartments are fixed triangular meshes whereas conductivity parameters are treated as unknowns. We show that a Woodbury update algorithm can be used to obtain a fast conductivity update scheme for both the single and double layer BEM formalism. This algorithm yields a speed gain up to a factor of 20 when compared to the direct computations, apart from at most 50% of additional computation time in the initialization phase of the algorithm. We also derive novel analytically closed expressions for the efficient and accurate computation of BEM matrix elements. Finally, we discuss which further steps are needed to equip future EEG systems with software devices that enable subject tailored head models for calibrated EEG and accurate source localization, on a routine basis.