Analyzing Brain Signals using Functional Geostatistics

Abstract

Electroencephalography (EEG) signals represent the brain’s electrical activity ob tained through placing electrodes on the scalp. These recordings can be analyzed as curves across space and time. In this work, EEG signals are treated as functions or curves using functional data analysis. The construction of the curves is achieved through a set of B-splines basis functions. Due to the inherent spatial and tem poral dependencies of EEG signals, various applications in spatial statistics have emerged, especially for modeling brain neuroimaging data. Here, we consider the brain as a random field where the EEG signals are analyzed as spatially correlated curves. Consequently, functional geostatistics methods are employed to predict curves at unsampled sites. The spatial prediction of the curves is performed using functional kriging, which employs the empirical functional principal components’ representation of the EEG curves. The predictive performance is assessed through leave-one-out functional cross-validation. The EEG signals used in this study were obtained from a controlled experiment involving inner speech. The analysis extracted key features from the smoothed curves, revealing that inner speech involves multiple brain regions, including the association area, Broca’s and Wernicke’s area.