Simplicial network analysis on EEG signals

It is well known that the epileptic signal analysis helps with automating the diagnosis and detection of a seizure, instead of relying on the traditional process of expert - visual inspection, which is both time-consuming and tedious. Recently, application of network analysis has grown as a surmounting approach for the interpretation of signals. In general, network-based signal analysis approach involves the conversion of time series procured at different physiological conditions into networks like hyper graphs, visibility graphs etc. followed by the derivation of network topological properties. In this paper, we make use of newly developed simplicial approach, where the cliques of visibility graphs are considered as simplices and are analyzed to obtain maximal cliques. Employing simplices would help in securing not only global and significant details but also localized and subtle details of dynamical behavior, for a given time series. The maximal cliques are mathematically evaluated to calculate three independent simplicial characterizers and maximum dimensionality, that define the structural anatomy and connectivity of the entire network at different topological levels. The maximum values of all the measures acquired are processed to differentiate normal signals against pathological EEG signals, using support vector machine through 10-fold cross validation. The classification analysis is performed on all the possible normal versus epileptic (ictal and inter-ictal) combinations, using two different EEG databases. In the case of University of Bonn database, the results are compared to that of conventional network parameters namely average degree, Global efficiency, Average path length and Assortativity. The classification results of both databases achieved very good accuracies, indicating that the proposed algorithm is an efficient and reliable approach for detecting epileptic EEG signals.