@inproceedings{6934868,
  abstract     = {{EEG source imaging (ESI) techniques estimate 3D brain activity based on electrical activity measured on the scalp. In a clinical context, these techniques are typically used for the analysis of epileptiform activity. They play a central role in the pre-surgical planning prior to removal of the epileptic seizure focus, needed in about 30% of people with epilepsy [1]. ESI techniques make use of a parametric model of the geometry and electromagnetic properties of the subject’s head. While the geometry can be modelled precisely using an anatomical MR image of the head, there remains high uncertainty in the electrical conductivity of several types of tissue in the head (skull, white and gray matter, scalp etc.). Commonly, these conductivity values are set to a conventional value, based on previous studies. Because individual conductivity values can deviate radically from the conventional values (exceeding an order of magnitude) this can lead to errors that need to be avoided for accurate estimation of the epileptic focus location [2]. 
In this work, a first Bayesian model is proposed that is able to simultaneously estimate the source location and the subject specific skull conductivity from the measured EEG signals. The expectation-maximization algorithm was used to iteratively update the parameter estimation. As a first proof of concept, we used a three-layered spherical head model and a single dipole source to simulate electrical activity on the scalp, measured at 36 electrode positions, for a range of human skull conductivity values found in literature. We compared the source localization performance with our adaptive conductivity estimation to the performance with several conventional conductivity values used in previous studies. We found that, due to the high variation in individual skull conductivity values, the true source can be located more than 15mm away from the estimated source location using the conventional conductivity. Adaptive estimation of the conductivity with the Bayesian model lowers the maximum location error to only 3mm (see Figure 1).
The first proof of concept looks promising and will be further deployed, including better probabilistic models for the variation in measured EEG, variation in dipole location and prior distribution of conductivity values. The final goal of this work is to estimate all tissue conductivity parameters, making the head model truly adaptive to the individual subject.
[1]	Strobbe G., Carrette E., Lopez J.D., Van Roost D., Meurs E., Vonck K., Boon P., Vandenberghe S., van Mierlo P. (2015) EEG source imaging of interictal spikes using multiple sparse volumetric priors for presurgical focus localization, NeuroImage, in preparation for submission.
[2]	Kassem A.,  Jackson D., Baumann S., Williams J., Wilton D., Fink P. and Prasky B. (1998) Effect of Conductivity Uncertainties and Modeling Errors on EEG Source Localization Using a 2-D Model, IEEE Transaction on Biomedical Engineering, vol. 45, no. 9, pp. 1135-1145}},
  author       = {{Verhoeven, Thibault and Strobbe, Gregor and van Mierlo, Pieter and Buteneers, Pieter and Vandenberghe, Stefaan and Dambre, Joni}},
  booktitle    = {{International Workshop on Seizure Predictions 7, Abstracts}},
  keywords     = {{epilepsy,skull conductivity,EEG,source imaging}},
  language     = {{eng}},
  location     = {{Melbourne, Australia}},
  title        = {{A Bayesian model to estimate individual skull conductivity for EEG source imaging}},
  year         = {{2015}},
}