bci2003数据说明

Dataset:BCI-experimentDatasetprovidedbyDepartmentofMedicalInformatics,InstituteforBiomedicalEngineering,UniversityofTechnologyGraz.(GertPfurtscheller)CorrespondencetoAloisSchlö[e-mail]fromanormalsubject(female,25y)duringafeedbacksession.Thesubjectsatinarelaxingchairwitharmrests.Thetaskwastocontrolafeedbackbarbymeansofimageryleftorrighthandmovements.Theorderofleftandrightcueswasrandom.Theexperimentconsistsof7runswith40trialseach.Allrunswereconductedonthesamedaywithseveralminutesbreakinbetween.Givenare280trialsof9slength.Thefirst2swasquite,att=2sanacousticstimulusindicatesthebeginningofthetrial,thetriggerchannel(#4)wentfromlowtohigh,andacross“+”wasdisplayedfor1s;thenatt=3s,anarrow(leftorright)wasdisplayedascue.Atthesametimethesubjectwasaskedtomoveabarintothedirectionofathecue.ThefeedbackwasbasedonAARparametersofchannel#1(C3)and#3(C4),theAARparameterswerecombinedwithadiscriminantanalysisintooneoutputparameter.(similarto[1,2]).TherecordingwasmadeusingaG.tecamplifierandaAg/AgClelectrodes.ThreebipolarEEGchannels(anterior‘+’,posterior‘-‘)weremeasuredoverC3,CzandC4.TheEEGwassampledwith128Hz,itwasfilteredbetween0.5and30Hz.Thedataisnotpublishedyet,similarexperimentsaredescribedin[1-4].Thetrialsfortrainingandtestingwererandomlyselected.Thisshouldpreventanysystematiceffectduetothefeedback.32321C3CzC415cm0123456789secTriggerBeepFeedbackperiodwithCueFigure1:Electrodepositions(left)andtimingscheme(right).FormatofthedataThedataissavedinaMatlab-fileformat.Thevariablex_traincontains3EEGchannels,140trialswith9secondseach.Thevariabley_traincontainstheclasslabels‘1’,‘2’forleftandright,respectively.x_testcontainsanothersetof140trials.Thecuewaspresentedfromt=3sto9s.Atthesametime,thefeedbackwaspresentedtothesubject.Withinthisperiod,itshouldbepossibletodistinguishthetwotypesoftrials.RequirementsandevaluationThetaskistoprovideananalysissystem,thatcanbeusedtocontrolacontinuousfeedback.Forthisreason,youshouldprovideacontinuousvalue(0class“1”,0class“2”,0non-decisive)foreachtimepoint.Themagnitudeofthevalueshouldreflecttheconfidenceoftheclassification,thesignindicatestheclass.Includeadescriptionofyouranalysissystem.EvaluationThereisacloserelationshipbetweentheerrorrateandthemutualinformation[4].Weproposethemutualinformationbecauseittakealsointoaccountthemagnitudeoftheoutputs.Thecriterionwillbetheratiobetweenthemaximumofthemutualinformationandthetimedelaysincethecue(t=3s).Onlytheperiodbetweent=4andt=9swillbeconsidered.Reference(s):[1]A.Schlögl,K.LuggerandG.Pfurtscheller(1997)UsingAdaptiveAutoregressiveParametersforaBrain-Computer-InterfaceExperiment,Proceedingsofthe19thAnnualInternationalConferenceiftheIEEEEngineeringinMedicineandBiologySociety,vol19,pp.1533-1535.[2]C.Neuper,A.Schlögl,G.Pfurtscheller(1999)Enhancementofleft-rightsensorimotorEEGdifferencesduringfeedback-regulatedmotorimagery.JClinNeurophysiol.16(4):373-82.[3]PfurtschellerG,NeuperC,SchlöglA,LuggerK.(1998)SeparabilityofEEGsignalsrecordedduringrightandleftmotorimageryusingadaptiveautoregressiveparameters.IEEETransRehabilEng.6(3):316-25.[4]SchlöglA.,NeuperC.PfurtschellerG.(2002)EstimatingthemutualinformationofanEEG-basedBrain-Computer-Interface,BiomedizinischeTechnik47(1-2):3-8.