Abstracts

Rationale: Accurate localization of the epileptic seizure onset zones (SOZ) and their propagations, which usually depends on the information obtained from EEG recordings, is crucial for successful epilepsy surgery. Analysis of high-frequency oscillations (> 80 Hz) in EEG recordings using the non-parametric Granger Causality method (GC), which uses heavy mathematical computations, relies on matrix spectral factorization, is reported as one of the additional ways to refine SOZ. To date, Wilson's Spectral Factorization Algorithm (WSFA) dominates in neuroscience applications, but recent investigations show that an alternative Janashia-Lagvilava algorithm (JLA) is more reliable for unstable matrices than the former WSFA. As the activity of pathological neuronal systems in epilepsy and their network interactions are very unstable, JLA may be effective in neuroscience but it is still unexplored. The aim of the study was to assess the possibilities of JLA as an additional confirmation of SOZ along with other traditional research methods and on the basis of real data from EEG recordings during presurgical evaluation of people with drug-resistant seizures.

Methods: Two regions (X  and Y) of interest and a time epoch were isolated by visual inspection of recorded EEG data during the epilepsy seizure of a patient. In order to apply non-parametric GC estimation for these regions, first we preprocessed these data.  In particular, data cleaning, bad segments removal and line noise suppression were performed by standard functions of EEGLAB Toolbox. Then the multitaper estimation was used to obtain cross power spectral density matrix S(f)  in frequency domain. The spectral factorization of this matrix 
                                                S (f)=H(f) ΣH*(f),
where H(f) is a transfer function and Σ  is a noise covariances matrix, was performed by Janashia-Lagvilava algorithm. The Granger-Geweke frequency dependent causality estimations I_y-x (f) and I_x-y (f) were computed for high frequency values (f >80 Hz ) by the standard formula using H(f) and Σ. These estimations were used to confirm the visually suspected seizure onset region and it propagation.

Results: The JLA was first time used on specific real EEG data as additional confirmation for identification of SOZ and its propagation by non-parametric GC method. The algorithm was compared to the corresponding algorithm of Wilson (Figures 1 and 2). _x000D_                                                               _x000D_ Figures 1 and 2 show GC estimation by JLA of the EEG the 35-year-old male with drug-resistant seizures and right mesial temporal scleroses on MRI; active electrodes: T2 (7); T4 (8); T6 (9). Based on these data, it can be assumed that T4(8) is the active site from which epileptic activity spreads directly to T6(8- >9).

Conclusions: The recently developed JLA has the potential to substitute the Wilson corresponding algorithm which is widely used in computational neuroscience at present. A thorough comparative analysis of these two methods should be the subject of the future work

Funding: The study was conducted with the financial support of an internal fundamental scientific grant from Caucasus International University, Tbilisi, Georgia.