Estimating brain effective connectivity from EEG signals of patients with autism disorder and healthy individuals by reducing volume conduction effect

Studying brain connectivity has shed light on understanding brain functions. Electroencephalogram signals recorded from the scalp surface comprise inter-dependent multi-channel signals each of which is a linear combination of simultaneously active brain sources as well as adjacent non-brain sources whose activity is widely volume conducted to the scalp through overlapping patterns. Evaluation of brain connectivity based on multivariate autoregressive (MVAR) model identification from neurological time series can be a proper tool for brain signal analysis. However, the MVAR model only considers the lagged influences between time series while ignoring the instantaneous effects (zero-lagged interactions) among simultaneously recorded neurological signals. Hence predicting instant interactions may result in fake connectivity, which may lead to misinterpreting in results. In this study, we aim to find instantaneous effects from coefficients of the MVAR model acquired using an ADALINE neural network and investigate the efficiency of the proposed algorithm by applying it to a simulated signal. We show that our coefficients are estimated accurately from channels of the simulated signal. Moreover, we apply the proposed method on a dataset of a group of 18 healthy children and 10 children with autism by comparing their effective connectivity estimated by direct directed transfer function method using new and old coefficients. Finally, to show the efficiency of the algorithm we exploit the support vector machine method for classifying the dataset. We show that there is a significant improvement in the results obtained from the proposed method.