Quantitative electroencephalographic (EEG) analysis\udis very useful for diagnosing dysfunctional neural states\udand for evaluating drug effects on the brain, among others.\udHowever, the bidirectional contamination between electrooculographic\ud(EOG) and cerebral activities can mislead and\udinduce wrong conclusions from EEG recordings. Different\udmethods for ocular reduction have been developed but only\udfew studies have shown an objective evaluation of their\udperformance. For this purpose, the following approaches\udwere evaluated with simulated data: regression analysis,\udadaptive filtering, and blind source separation (BSS). In the\udfirst two, filtered versions were also taken into account by\udfiltering EOG references in order to reduce the cancellation\udof cerebral high frequency components in EEG data.\udPerformance of these methods was quantitatively evaluated\udby level of similarity, agreement and errors in spectral\udvariables both between sources and corrected EEG recordings.\udTopographic distributions showed that errors were\udlocated at anterior sites and especially in frontopolar and\udlateral–frontal regions. In addition, these errors were higher\udin theta and especially delta band. In general, filtered versions\udof time-domain regression and of adaptive filtering with RLS\udalgorithm provided a very effective ocular reduction. However,\udBSS based on second order statistics showed the highest\udsimilarity indexes and the lowest errors in spectral variables.
展开▼
