Saeid Sanei - EEG Signal Processing and Machine Learning

Both authors wish to thank most sincerely our Research Associates and PhD students who have contributed so much to the materials in this work.

Saeid Sanei and Jonathon A. Chambers

There is ever‐increasing global demand for more affordable and effective clinical and healthcare services. New techniques and equipment must therefore be developed to aid in the diagnosis, monitoring, and treatment of abnormalities and diseases of the human body. Biomedical signals (biosignals) in their manifold forms are rich information sources, which when appropriately processed have the potential to facilitate such advancements. In today's technology, such processing is very likely to be digital, as confirmed by the inclusion of digital signal processing concepts as core training in biomedical engineering degrees. Recent advancements in digital signal processing are expected to underpin key aspects of the future progress in biomedical research and technology, and it is the purpose of this research monograph to highlight this trend for the processing of measurements of brain activity, primarily electroencephalograms (EEGs).

Most of the concepts in multichannel EEG digital signal processing have their origin in distinct application areas such as communications engineering, seismics, speech and music signal processing, together with the processing of other physiological signals, such as electrocardiograms (ECGs) The particular topics in digital signal processing first explained in this research monograph include: definitions; illustrations; time domain, frequency domain, and time–frequency domain processing; signal conditioning; signal transforms; linear and nonlinear filtering; chaos definition, evaluation, and measurement; certain classification algorithms; adaptive systems; independent component analysis; and multivariate autoregressive modelling. In addition, motivated by research in the field over the last two decades, techniques specifically related to EEG processing such as brain source localization, detection and classification of event‐related potentials, sleep signal analysis, seizure detection and prediction, together with brain–computer interfacing are comprehensively explained and, with the help of suitable graphs and (topographic) images, simulation results are provided to assess the efficacy of the methods.

Chapter 1of this research monograph is a comprehensive biography of the history and generation of EEG signals, together with a discussion of their significance and diagnostic capability. Chapter 2provides an in‐depth introduction to the mathematical algorithms and tools commonly used in the processing of EEG signals. Most of these algorithms have only been recently developed by experts in the signal processing community and then applied to the analysis of EEG signals for various purposes. In Chapter 3, event‐related potentials are explained and the schemes for their detection and classification are explored. Many neurological and psychiatric brain disorders are diagnosed and monitored using these techniques. Chapter 4complements the previous chapter by specifically looking at the behaviour of EEG signals in patients suffering from epilepsy. Some very recent methods in seizure prediction are demonstrated. This chapter concludes by opening up a new methodology in joint, or bimodal, EEG–fMRI analysis of epileptic seizure signals. Localization of brain source signals is next covered in Chapter 5. Traditional dipole methods are described and some very recent processing techniques such as blind source separation are briefly reviewed. In Chapter 6, the concepts developed for the analysis and description of EEG sleep recordings are summarized and the important parameters and terminologies are explained. Finally, in Chapter 7, one of the most important applications of the developed mathematical tools for processing of EEG signals, namely brain–computer interfacing, is explored and recent advancements are briefly explained. Results of the application of these algorithms are described.

In the treatment of various topics covered within this research monograph it is assumed that the reader has a background in the fundamentals of digital signal processing and wishes to focus on processing of EEGs. It is hoped that the concepts covered in each chapter provide a foundation for future research and development in the field.

In conclusion, we do wish to stress that in this book there is no attempt to challenge previous clinical or diagnostic knowledge. Instead, the tools and algorithms described in this book can, we believe, potentially enhance the significant clinically related information within EEG signals and thereby aid physicians and ultimately provide more cost effective and efficient diagnostic tools.

Both authors wish to thank most sincerely our previous and current PhD students who have contributed so much to the material in this work and our understanding of the field. Special thanks to Min Jing, Tracey Lee, Kianoush Nazarpour, Leor Shoker, Loukianous Spyrou, and Wenwu Wang, who contributed to providing some of the illustrations. Finally, this book became truly possible due to spiritual support and encouragement of Maryam Zahabsaniei, Erfan Sanei, and Ideen Sanei.

Saeid Sanei

Jonathon A. Chambers

January 2007

3DThree‐dimensionalAASMAmerican Academic of Sleep MedicineACCAnterior cingulate cortexACEAddenbrooke’s cognitive examinationACRAccuracy of responsesACTAdaptive chirplet transformADAlzheimer’s diseaseADCAnalogue‐to‐digital converterADDAD patients with mild dementiaADDAttention‐deficit disorderADHDAttention‐deficit hyperactivity disorderAEApproximate entropyAEAutoencoderAEPAudio evoked potentialsAfCAffective computingAg–AgClSilver–silver chlorideAIArtificial intelligenceAICAkaike information criterionALEAdaptive line enhancerALFAdaptive standardized LORETA/FOCUSSALMAugmented Lagrange multipliers methodALSAlternating least squaresALSAmyotrophic lateral sclerosisAMDFAverage magnitude difference functionAMIAverage mutual informationAMMAugmented mixing matrixANNArtificial neural networkAODAuditory oddballAPAction potentialApEnApproximate entropyAPGARCHAsymmetric power GARCHARAutoregressive modellingARMAAutoregressive moving averageASCOTAdaptive slope of wavelet coefficient counts over various thresholdsASDAutism spectrum disorderASDAAmerican Sleep Disorders AssociationAsIAsymmetry indexASRAutomatic speaker recognitionASSAverage artefact subtractionAUCArea under the curveBASBehavioural activation systemBBCIBerlin BCIBBIBrain‐to‐brain interfaceBCGBallistocardiogramBCIBrain–computer interfacingBDSBrock, Dechert, and ScheinkmanBEMBoundary‐element methodBFBeamformerBGDBootstrapped geometric differenceBICBayesian information criterionBISBehavioural inhibition systemBISBispectral indexBMIBrain–machine interfacingBOLDBlood oxygenation level dependentBPBereitschaftspotentialBPBipolar disorderBrain/MINDSBrain Mapping by Integrated Neurotechnologies for Disease StudiesBSEBlind source extractionBSRBurst‐suppression ratioBSSBlind source separationbvFTDBehaviour variant frontotemporal dementiaCaCalciumCAEContractive autoencoderCANDECOMPCanonical decompositionCBDCorticobasal degenerationCBFCerebral blood flowCCACanonical correlation analysisCEEMDANComplete ensemble EMD with adaptive noiseCFCharacteristic functionCFCognitive fluctuationCFSChronic fatigue syndromeClChlorideCDLSACoupled dictionary learning with sparse approximationCDRCurrent distributed‐source reconstructionCICovariance intersectioncICAConstrained ICACITConcealed information testCJDCreutzfeldt–Jakob diseaseCMACircumplex model of affectsCMTFCoupled matrix and tensor factorizationsCMOSComplementary metal oxide semiconductorCNNConvolutional neural networkCNSCentral nervous systemCORCONDIACore consistency diagnosticCoSAMPCompressive sampling matching pursuitCPSCyber‐physical systemsCRBPFConstrained Rao‐Blackwellised particle filterCSACentral sleep apnoeaCSDCurrent source densityCSFCerebrospinal fluidCSPCommon spatial patternsCTComputerized tomographyDAEDenoising autoencoderDARPADefence Advanced Research Projects AgencyDASMDifferential asymmetryDBSDeep brain stimulationDCDirect currentDCAUDifferential CausalityDCMDynamic causal modellingDCTDiscrete cosine transformdDTFDirect directed transfer functionDEDifferential entropyDeconvNetDeconvolutional ANNDFTDiscrete Fourier transformDFVDominant frequency variabilityDHTDiscrete Hermite transformDLDiagonal loadingDLEDigitally linked earsDMDefault modeDMNDefault mode networkDNNDeep neural networkDPFDifferential pathlength factorDSMDiagnostic and Statistical ManualDSTCLNDeep spatio‐temporal convolutional bidirectional long short‐term memory networkDTDecision treeDTFDirected transfer functionDTIDiffusion tensor imagingDUETDegenerate unmixing estimation techniqueDWTDiscrete wavelet transformECDElectric current dipoleECDEquivalent current dipoleECGElectrocardiogramECGElectrocardiographyECoGElectrocorticogramECTElectroconvulsive therapyEDError distanceEEGElectroencephalogramEEGElectroencephalographyEEMDEnsemble empirical mode decompositionEGARCHExponential GARCHEGGElectrogastrographyEKGElectrocardiogramEKGElectrocardiographyEMExpectation maximizationEMDEmpirical mode decompositionEMGElectromyogramEMGElectromyographyENetEfficient neural networkEOGElectro‐oculogramEPEvoked potentialEPNEarly posterior negativityEPSPExcitatory post‐synaptic potentialERBMEntropy rate bound minimizationERDEvent‐related desynchronizationERNError‐related negativityERPEvent‐related potentialERSEvent‐related synchronizationFAFactor analysisFCFunctional connectivityFCMFuzzy c‐meansFDFractal dimensionFDAFood and Drug AdministrationFDispEnFluctuation‐based dispersion entropyFDRFalse detection rateFEMFinite element modelFFNNFeed forward neural networkFETField‐effect transistorfICAFast independent component analysisFIRFinite impulse responsefMRIFunctional magnetic resonance imagingFMSFibromyalgia syndromeFNFalse negativefNIRSFunctional near‐infrared spectroscopyFOForamen ovaleFOBSSFirst order blind source separationFOCUSSFocal underdetermined system solverFOOBIFourth order cumulant based blind identificationFPFalse positiveFRDAFrontal rhythmic delta activityFRNFeedback related negativityFSORFeature selection with orthogonal regressionFSPFalsely detected source number (position)FTDFrontotemporal dementiaFuzEnFuzzy entropyGAGenetic algorithmGADGeneral anxiety disorderGANGenerative adversarial networkGARCHGeneralized autoregressive conditional heteroskedasticityGARCH‐MGARCH‐in‐meanGCGranger causalityGCNGraph convolutional networkGFNNGlobal false nearest neighboursGJR‐GARCHGlosten, Jagannathan, & Runkle GARCHGLMGeneral linear modelGMMGaussian mixture modelGPGaussian processGP‐LRGaussian process logistic regressionGSCCAGroup sparse canonical correlation analysisGSRGalvanic skin responseGWNGaussian white noiseHBO/HbOOxyhaemoglobinHBR/HbRDe‐oxyhaemoglobinHBTTotal haemoglobinHCIHuman computer interactionHDHuntington’s diseaseHEOGHorizontal electro‐oculographHFDHiguchi's fractal dimensionHHTHilbert–Huang transformHMDHead‐mounted displayHMMHidden Markov modelHOPLSHigher‐order partial least squaresHOSHigher‐order statisticsHRHemodynamic responseHRFHaemodynamic response functionHTHilbert transformIAPSInternational affective picture systemIBEInternational Bureau for EpilepsyICIndependent componentICAIndependent component analysisiCOHImaginary part of coherencyIEDInterictal epileptiform dischargeiEEGIntracranial electroencephalogramICAIndependent component analysisIIRInfinite impulse responseILAEInternational League Against EpilepsyIMFIntrinsic mode functionImSCohImaginary part of S‐coherencyINDSCALINDividual Differences SCALingIoBInternet‐of‐brainsIPLInferior parietal lobuleIPSPInhibitory post‐synaptic potentialIRImpulse responseIRLSIterative recursive least squaresISODATAIterative self‐organizing data analysis technique algorithmIsomapIsometric mappingISSWTInverse synchro‐squeezing wavelet transformITRInformation transfer rateIVEImmersive virtual environmentsJADJoint approximate diagonalizationJADEJoint approximate diagonalization of eigenmatricesjICAJoint ICAKPotassiumKcKolmogorov complexityKDTKarolinska drowsiness testKLKullback–LeiblerKLTKarhunen–Loéve transformKMIKinaesthetic motor imageryKNNk‐nearest neighbourKPCAKernel principal component analysisKSSKarolinska sleepiness scaleKTKuhn–TuckerLBDLewy body dementiaLCMVLinearly constrained minimum varianceLDLinear discriminantsLDLinearly distributedLDALinear discriminant analysisLDALong delta activityLELyapunov exponentLEMLocal EEG modelLLELargest Lyapunov exponentLMSLeast mean squareLORETALow‐resolution electromagnetic tomography algorithmLPLowpassLPMLetters per minuteLPPLate positive potentialLRCNLong‐term recurrent convolutional networkLRTLow‐resolution tomographyLSLeast squaresLSELeast‐squares errorLSTMLong short‐term memory networkLVQLearning vector quantizationLWRLevinson–Wiggins–RobinsonLZCLempel–Ziv complexityM2MMachine‐to‐machineMAMental arithmeticMAMoving averageMAFMultivariate ambiguity functionMAPMaximum a posterioriMCIMild cognitive impairmentMCMCMarkov chain Monte CarloMDIMultidimensional directed informationMDPMoving dipoleMEGMagnetoencephalogramMFDEMultiscale fluctuation‐based dispersion entropymHTTMutant huntingtinMIMutual informationMILMatrix inversion lemmaMLMaximum likelihoodMLEMaximum likelihood estimationMLEMaximum Lyapunov exponentMLPMultilayer perceptronMMNMismatch negativityMMSEMinimum mean squared errorMNIMontreal Neurological Institute and HospitalMNLSMinimum norm least squaresMPMatching pursuitsMRIMagnetic resonance imagingMRPMovement‐related potentialMSEMultiple system atrophyMSEMean squared errorMSEMultiscale entropyMSMultiple sclerosisMTLEMesial temporal lobe epilepsyMUSICMultichannel signal classificationMVARMultivariate autoregressiveNaSodiumNCNormal controlNCDFNormal cumulative distribution functionNCSPNonparametric common spatial patternsNDDNeurodevelopmental disorderNESNonepileptic seizureNIHNational Institute of HealthNIRNear infraredNIRSNear‐infrared spectroscopyNLMSNormalized least mean squareNMFNonnegative matrix factorizationNMCSPNonparametric multiclass common spatial patternsNMRNuclear magnetic resonanceNNNeural networkNNQPNonnegative quadratic programNPNeural processNREMNon‐rapid eye movementNSINonstationary indexOAOcular artefactOBSOptimal basis setOBSOrganic brain syndromeOFCOrbital frontal cortexOMPOrthogonal matching pursuitOPOddball paradigmOSAObstructive sleep apnoeaOSAHSObstructive sleep apnoea hypopnea syndromePARAFACParallel factor analysisPCAPrincipal component analysisPCANetPrincipal component analysis networkPCCPearson product correlation coefficientPCCPosterior cingulate corticesPDParkinson’s diseasePDCPartial directed coherencepdfProbability density functionPeError positivityPerEnPermutation entropyPETPositron emission tomographyPFParticle filterPFCPrefrontal cortexPICPower iteration clusteringPIFPhase interaction functionPLEDPeriodic literalized epileptiform dischargesPLIPhase lag indexPLMDPeriodic limb movement disorderPLSPartial least squaresPMBRPost‐movement beta reboundPMBSPost‐movement beta synchronizationPNRDNonrhythmic delta activityPOSTPositive occipital sharp transientsPPCPhase–phase couplingPPGPhotoplethysmographyPPMPiecewise Prony methodPSDPower spectrum densityPSDMPhase‐space dissimilarity measuresPSGPolysomnographyPSIPhase‐slope indexPSPPost‐synaptic potentialPSPProgressive supranuclear palsyPSWCPeriodic sharp wave complexesPTSDPost‐traumatic stress disorderPWVDPseudo‐Wigner–Ville distributionQEEGQuantitative EEGQGARCHQuadratic GARCHQNNQuantum neural networksQPQuadratic programmingR&KRechtschtschaffen and KalesRAPRecursively applied and projectedRASMRational asymmetryRBDREM sleep behaviour disorderRBFRadial basis functionRBPFRao‐Blackwellised particle filterRBRRelative beta ratioRCERecursive channel eliminationRERegional entropyReLURectified linear unitREMRapid eye movementRFRadio frequencyRFNNRecurrent fuzzy neural networkRKHSReproducing kernel Hilbert spacesRLSRecursive least squaresRMBFRobust minimum variance beamformerRMSRoot mean squareRNNRecurrent neural networkROCReceiver operating characteristicRPReadiness potentialRRRespiratory rateRTReaction timerTMSrepetitive transcranial magnetic stimulationRVResidual varianceSAEStacked autoencoderSampEnSample entropySASSleep apnoea syndromeSCASparse component analysisSCDSickle cell diseaseSCPSlow cortical potentialSCPSSlow cortical potential shiftSCRSkin conductance responseSCVSpectral coherence valueSCWTStroop colour and word testSDAEStacked denoising autoencoderSDTFShort‐time DTFSEMStructural equation modellingSFSSyncFastSlowSGSensory gatingSISynchronization indexSICASpatial ICASLSynchronization likelihoodsLORETAStandardized LORETASLTPShort‐ and long‐term predictionSMISample‐matrix inversionSMlSensorimotor leftSMOTESynthetic minority oversampling techniquesMRIStructural MRISNSalient networkSNNSpike neural networkSNNAPSimulator for Neural Networks and Action PotentialsSNRSignal‐to‐noise ratioSOBISecond‐order blind identificationSOBIUMSecond‐order blind identification of underdetermined mixturesSPETSingle photon emission tomographySPMStatistical parametric mappingSPQSchizotypal personality questionnaireSREDASubclinical rhythmic EEG discharges of adultsSRNNSleep EEG recognition neural networkSSASingular spectrum analysisSSLOFOShrinking standard LORETA‐FOCUSSSSPESubacute sclerosing panencephalitisSSVEPSteady‐state visual evoked potentialSSVERSteady‐state visual evoked responseSSWTSynchro‐squeezing wavelet transformSTFSpace–time–frequencySTFDSpatial time–frequency distributionSTFTShort time–frequency transformSTLShort‐term largest Lyapunov exponentSTSSuperior temporal sulcusSVSupport vectorsSVDSingular‐value decompositionSVMSupport vector machinesSVRSupport vector regressionSWASlow‐wave activitySWDAStep‐wise discriminant analysisSWSlow waveSWPSlow‐wave powerSWSSlow‐wave sleepTBITraumatic brain injuryTDNNTime delay neural networkTDOATime difference of arrivalTDP‐43Transactive response DNA‐binding protein 43 kDaTENSTranscutaneous electrical nerve stimulationTFTime–frequencyTGARCHThreshold GARCH modelTICATemporal ICATISTMS induction simulatorTLETemporal lobe epilepsyTMSTranscranial magnetic stimulationTNTrue negativeTNMTraditional nonlinear methodTOATime of arrivalTotHbTotal haemoglobinTPTrue positiveTRRepeat timeTSSATensor‐based singular spectrum analysisTTDThought translation deviceUBIUnderdetermined blind identificationUOMUnderdetermined orthogonal modelUSPUndetected source numberUSRUnderdetermined source recoveryVAEVariational autoencoderVEOGVertical electro‐oculographVEPVisual evoked potentialVLSIVery large‐scale integratedvMPFCVentral medial prefrontal corticesVPPVertex positive peakVRVirtual realityWAWald tests on amplitudesWCWord chainWCOWeakly coupled oscillatorWDCWeighted degree centralityWLWald test on locationsWMNWeighted minimum normWNWavelet networkWPEWavelet packet energywPLIWeighted phase lag indexwSMIWeighted symbolic mutual informationWTWavelet transformWUWeighted undersamplingWVWigner–Ville