EEG-and MRI-based epilepsy source localization using multivariate empirical mode decomposition and inverse solution method

The only treatment for patients with medical refractory epilepsy is to resect the part of the cortex that is origin of epilepsy by surgery. An extensive pre-surgical evaluation is required to define the Epileptogenic Zone (EZ) accurately. There is a large variation of neuroimaging approaches that are utilized for pre-surgical evaluation depending on the protocol of each epilepsy surgery center. Although Electroencephalography (EEG)-based source localization (ESL) estimates the EZ more precisely than other techniques but, it is used rarely in surgery centers. The reason behind the low usage of this trustworthy technique is its requirement for technical expertise together with experts’ supervision and lack of recommended guidelines for this technique. The accuracy of ESL depends on all the stages of data processing including: head model reconstruction, signal pre-processing and inverse solution. Therefore, a standardized algorithm with less supervision is desired to utilize ESL for pre-surgical evaluation. One of the factors that needs to be considered for the purpose of establishing an automated and standardized algorithm is eye blink artifact removal due to its morphological resemblance to epileptic discharges. Few studies considered eye blink artifact removal for purpose of epilepsy source localization studies and most of them used either visual inspection or computer-based approaches which still need of experts’ supervision. Besides, non-stationary, non-linear and multivariate characteristics of EEG needs to be considered for choosing a proper processing method for extracting epileptic spikes’ features. Nevertheless, patient’s realistic head model is essential to obtain accurate source localization results. Although many inverse solutions exist but, the ones which do not require specialists’ involvement with minimal error is desired. Standardized Low Resolution Tomography (sLORETA) andWeighted Minimum Norm (WMN) are linear distributed inverse solutions which lead up to zero localization error using noise-free EEG, state-of-the-art feature extractor and realistic head model. Therefore, in this study a coupled Multivariate Empirical Mode Decomposition (MEMD) with embedded automated artifact remover algorithm and inverse solution method is proposed. To remove eye blink artifacts, the mother wavelet of Bior 3.3 was used due to its high morphological resemblance to eye blink and yet differentiable characteristic to epileptic spikes. Since MEMD method is a data-driven method which meets the criteria to be applied for EEG processing, therefore this method was employed to extract EEG epileptic spike features. In the current study, clinical dataset of 20 subjects were used to examine sLORETA andWMN fed by raw EEG signals and MEMD features on each patient’s realistic head model. sLORETA in combination with MEMD feature after eye blink removal proved to be a reliable ESL algorithm with 100% accuracy. The results show significantly improved EZ localization results in comparison with similar works and capability of this algorithm to not only determine the epilepsy origin lobe, but also the exact focus on the lobe. The outcomes were validated using MRI references which are verified via post-surgical results. Therefore proposed algorithm has the advantages to localize EZ using ESL inexpensively and accurately which promotes usage of this valuable technique for epilepsy pre-surgical evaluation.

Preview Text FK 2018 102 - IR.pdf Download (1MB) | Preview

Actions (login required)

View Item