InordertoexplorethecorrelationbetweentheadjacentsegmentsofalongtermEEG,animprovedprincipalcomponentanalysis(PCA)methodbasedonmutualinformationalgorithmisproposed.Aone-dimensionEEGtimeseriesispidedequallyintomanysegments,sothateachsegmentcanberegardedasanindependentvariablesandmulti-segmentedEEGcanbeexpressedasadatamatrix.Then,wesubstitutemutualinformationmatrixforcovariancematrixinPCAandconducttherelevanceanalysisofsegmentedEEG.Theexperimentalresultsshowthatthecontributionrateoffirstprincipalcomponent(FPC)ofsegmentedEEGismorelargerthanothers,whichcaneffectivelyreflectthedifferenceofepilepticEEGandnormalEEGwiththechangeofsegmentnumber.Inaddition,theevolutionofFPCconducetoidentifythetime-segmentlocationsofabnormaldynamicprocessesofbrainactivities,theseconclusionsarehelpfulfortheclinicalanalysisofEEG.
