Can graph metrics be used for EEG-BCIs based on hand motor imagery?

摘要: Abstract The study of motor imagery (MI) has been a subject great interest within the brain-computer interface (BCI) community. Several approaches have proposed to solve problem classifying cerebral responses due MI, mostly based on power spectral density mu and beta bands; however, no optimum manner proceeding through fundamental steps MI-BCI yet established. In this work, we explored relatively novel approach regarding feature generation for by assuming that functional connectivity patterns brain are altered during hand MI. We modelled interactions among EEG electrodes graph, extracted metrics from it left right MI eight subjects classified signals using commonly employed techniques in BCI community (LDA SVM). also compared more established method signal as classifier features. With graph method, confirmed only specific provide relevant information data classification. A first provided maximum average classification rates across all 86% band 87% band. For PSD were 98% 99% bands, respectively. However, much larger number features was needed: (130 ± 44) (273 ± 89) Aiming reproduce these pairwise inputs combinations tested. They proved be sufficient obtain essentially same accuracy rates, but with considerably smaller – about 60 features, both bands. thus conclude is feasible option signals.