EEG Signal Processing and Machine Learning (2^nd Ed.)

Explore cutting edge techniques at the forefront of electroencephalogram research and artificial intelligence from leading voices in the field

The newly revised Second Edition of EEG Signal Processing and Machine Learning delivers an inclusive and thorough exploration of new techniques and outcomes in electroencephalogram (EEG) research in the areas of analysis, processing, and decision making about a variety of brain states, abnormalities, and disorders using advanced signal processing and machine learning techniques. The book content is substantially increased upon that of the first edition and, while it retains what made the first edition so popular, is composed of more than 50% new material.

The distinguished authors have included new material on tensors for EEG analysis and sensor fusion, as well as new chapters on mental fatigue, sleep, seizure, neurodevelopmental diseases, BCI, and psychiatric abnormalities. In addition to including a comprehensive chapter on machine learning, machine learning applications have been added to almost all the chapters. Moreover, multimodal brain screening, such as EEG-fMRI, and brain connectivity have been included as two new chapters in this new edition.

Readers will also benefit from the inclusion of:

• A thorough introduction to EEGs, including neural activities, action potentials, EEG generation, brain rhythms, and EEG recording and measurement
• An exploration of brain waves, including their generation, recording, and instrumentation, abnormal EEG patterns and the effects of ageing and mental disorders
• A treatment of mathematical models for normal and abnormal EEGs
• Discussions of the fundamentals of EEG signal processing, including statistical properties, linear and nonlinear systems, frequency domain approaches, tensor factorization, diffusion adaptive filtering, deep neural networks, and complex-valued signal processing

Perfect for biomedical engineers, neuroscientists, neurophysiologists, psychiatrists, engineers, students and researchers in the above areas, the Second Edition of EEG Signal Processing and Machine Learning will also earn a place in the libraries of undergraduate and postgraduate students studying Biomedical Engineering, Neuroscience and Epileptology.

Preface to the Second Edition xvii

Preface to the First Edition xxi

List of Abbreviations xxiii

1 Introduction to Electroencephalography 1

1.1 Introduction 1

1.2 History 2

1.3 Neural Activities 5

1.4 Action Potentials 6

1.5 EEG Generation 8

1.6 The Brain as a Network 12

1.7 Summary 12

References 13

2 EEG Waveforms 15

2.1 Brain Rhythms 15

2.2 EEG Recording and Measurement 18

2.2.1 Conventional Electrode Positioning 21

2.2.2 Unconventional and Special Purpose EEG Recording Systems 24

2.2.3 Invasive Recording of Brain Potentials 26

2.2.4 Conditioning the Signals 27

2.3 Sleep 28

2.4 Mental Fatigue 30

2.5 Emotions 30

2.6 Neurodevelopmental Disorders 31

2.7 Abnormal EEG Patterns 32

2.8 Ageing 33

2.9 Mental Disorders 34

2.9.1 Dementia 34

2.9.2 Epileptic Seizure and Nonepileptic Attacks 35

2.9.3 Psychiatric Disorders 39

2.9.4 External Effects 40

2.10 Summary 41

References 42

3 EEG Signal Modelling 47

3.1 Introduction 47

3.2 Physiological Modelling of EEG Generation 47

3.2.1 Integrate-and-Fire Models 49

3.2.2 Phase-Coupled Models 49

3.2.3 Hodgkin–Huxley Model 51

3.2.4 Morris–Lecar Model 54

3.3 Generating EEG Signals Based on Modelling the Neuronal Activities 57

3.4 Mathematical Models Derived Directly from the EEG Signals 61

3.4.1 Linear Models 61

3.4.1.1 Prediction Method 61

3.4.1.2 Prony’s Method 62

3.4.2 Nonlinear Modelling 64

3.4.3 Gaussian Mixture Model 66

3.5 Electronic Models 67

3.5.1 Models Describing the Function of the Membrane 67

3.5.1.1 Lewis Membrane Model 68

3.5.1.2 Roy Membrane Model 68

3.5.2 Models Describing the Function of a Neuron 68

3.5.2.1 Lewis Neuron Model 68

3.5.2.2 The Harmon Neuron Model 71

3.5.3 A Model Describing the Propagation of the Action Pulse in an Axon 72

3.5.4 Integrated Circuit Realizations 72

3.6 Dynamic Modelling of Neuron Action Potential Threshold 72

3.7 Summary 73

References 73

4 Fundamentals of EEG Signal Processing 77

4.1 Introduction 77

4.2 Nonlinearity of the Medium 78

4.3 Nonstationarity 79

4.4 Signal Segmentation 80

4.5 Signal Transforms and Joint Time–Frequency Analysis 83

4.5.1 Wavelet Transform 87

4.5.1.1 Continuous Wavelet Transform 87

4.5.1.2 Examples of Continuous Wavelets 89

4.5.1.3 Discrete-Time Wavelet Transform 89

4.5.1.4 Multiresolution Analysis 90

4.5.1.5 Wavelet Transform Using Fourier Transform 93

4.5.1.6 Reconstruction 94

4.5.2 Synchro-Squeezed Wavelet Transform 95

4.5.3 Ambiguity Function and the Wigner–Ville Distribution 96

4.6 Empirical Mode Decomposition 100

4.7 Coherency, Multivariate Autoregressive Modelling, and Directed Transfer Function 101

4.8 Filtering and Denoising 104

4.9 Principal Component Analysis 107

4.9.1 Singular Value Decomposition 108

4.10 Summary 110

References 110

5 EEG Signal Decomposition 115

5.1 Introduction 115

5.2 Singular Spectrum Analysis 115

5.2.1 Decomposition 116

5.2.2 Reconstruction 117

5.3 Multichannel EEG Decomposition 118

5.3.1 Independent Component Analysis 118

5.3.2 Instantaneous BSS 122

5.3.3 Convolutive BSS 126

5.3.3.1 General Applications 127

5.3.3.2 Application of Convolutive BSS to EEG 128

5.4 Sparse Component Analysis 129

5.4.1 Standard Algorithms for Sparse Source Recovery 130

5.4.1.1 Greedy-Based Solution 130

5.4.1.2 Relaxation-Based Solution 131

5.4.2 k-Sparse Mixtures 131

5.5 Nonlinear BSS 133

5.6 Constrained BSS 134

5.7 Application of Constrained BSS; Example 135

5.8 Multiway EEG Decompositions 136

5.8.1 Tensor Factorization for BSS 139

5.8.2 Solving BSS of Nonstationary Sources Using Tensor Factorization 143

5.9 Tensor Factorization for Underdetermined Source Separation 149

5.10 Tensor Factorization for Separation of Convolutive Mixtures in the Time Domain 153

5.11 Separation of Correlated Sources via Tensor Factorization 153

5.12 Common Component Analysis 154

5.13 Canonical Correlation Analysis 154

5.14 Summary 155

References 155

6 Chaos and Dynamical Analysis 165

6.1 Introduction to Chaos and Dynamical Systems 165

6.2 Entropy 166

6.3 Kolmogorov Entropy 166

6.4 Multiscale Fluctuation-Based Dispersion Entropy 167

6.5 Lyapunov Exponents 167

6.6 Plotting the Attractor Dimensions from Time Series 169

6.7 Estimation of Lyapunov Exponents from Time Series 169

6.7.1 Optimum Time Delay 172

6.7.2 Optimum Embedding Dimension 172

6.8 Approximate Entropy 173

6.9 Using Prediction Order 174

6.10 Summary 175

References 175

7 Machine Learning for EEG Analysis 177

7.1 Introduction 177

7.2 Clustering Approaches 181

7.2.1 k-Means Clustering Algorithm 181

7.2.2 Iterative Self-Organizing Data Analysis Technique 183

7.2.3 Gap Statistics 183

7.2.4 Density-Based Clustering 184

7.2.5 Affinity-Based Clustering 184

7.2.6 Deep Clustering 184

7.2.7 Semi-Supervised Clustering 185

7.2.7.1 Basic Semi-Supervised Techniques 185

7.2.7.2 Deep Semi-Supervised Techniques 186

7.2.8 Fuzzy Clustering 186

7.3 Classification Algorithms 187

7.3.1 Decision Trees 188

7.3.2 Random Forest 189

7.3.3 Linear Discriminant Analysis 190

7.3.4 Support Vector Machines 191

7.3.5 k-Nearest Neighbour 199

7.3.6 Gaussian Mixture Model 200

7.3.7 Logistic Regression 200

7.3.8 Reinforcement Learning 201

7.3.9 Artificial Neural Networks 201

7.3.9.1 Deep Neural Networks 203

7.3.9.2 Convolutional Neural Networks 205

7.3.9.3 Autoencoders 207

7.3.9.4 Variational Autoencoder 208

7.3.9.5 Recent DNN Approaches 209

7.3.9.6 Spike Neural Networks 210

7.3.9.7 Applications of DNNs to EEG 212

7.3.10 Gaussian Processes 212

7.3.11 Neural Processes 213

7.3.12 Graph Convolutional Networks 213

7.3.13 Naïve Bayes Classifier 213

7.3.14 Hidden Markov Model 214

7.3.14.1 Forward Algorithm 216

7.3.14.2 Backward Algorithm 216

7.3.14.3 HMM Design 216

7.4 Common Spatial Patterns 218

7.5 Summary 222

References 223

8 Brain Connectivity and Its Applications 235

8.1 Introduction 235

8.2 Connectivity through Coherency 238

8.3 Phase-Slope Index 240

8.4 Multivariate Directionality Estimation 240

8.4.1 Directed Transfer Function 241

8.4.2 Direct DTF 242

8.4.3 Partial Directed Coherence 243

8.5 Modelling the Connectivity by Structural Equation Modelling 243

8.6 Stockwell Time–Frequency Transform for Connectivity Estimation 246

8.7 Inter-Subject EEG Connectivity 247

8.7.1 Objectives 247

8.7.2 Technological Relevance 247

8.8 State-Space Model for Estimation of Cortical Interactions 249

8.9 Application of Cooperative Adaptive Filters 251

8.9.1 Use of Cooperative Kalman Filter 253

8.9.2 Task-Related Adaptive Connectivity 254

8.9.3 Diffusion Adaptation 255

8.9.4 Brain Connectivity for Cooperative Adaptation 256

8.9.5 Other Applications of Cooperative Learning and Brain Connectivity Estimation 257

8.10 Graph Representation of Brain Connectivity 258

8.11 Tensor Factorization Approach 259

8.12 Summary 262

References 263

9 Event-Related Brain Responses 269

9.1 Introduction 269

9.2 ERP Generation and Types 269

9.2.1 P300 and its Subcomponents 273

9.3 Detection, Separation, and Classification of P300 Signals 274

9.3.1 Using ICA 275

9.3.2 Estimation of Single-Trial Brain Responses by Modelling the ERP Waveforms 277

9.3.3 ERP Source Tracking in Time 278

9.3.4 Time–Frequency Domain Analysis 280

9.3.5 Application of Kalman Filter 284

9.3.6 Particle Filtering and its Application to ERP Tracking 286

9.3.7 Variational Bayes Method 291

9.3.8 Prony’s Approach for Detection of P300 Signals 293

9.3.9 Adaptive Time–Frequency Methods 297

9.4 Brain Activity Assessment Using ERP 298

9.5 Application of P300 to BCI 299

9.6 Summary 300

References 301

10 Localization of Brain Sources 307

10.1 Introduction 307

10.2 General Approaches to Source Localization 308

10.2.1 Dipole Assumption 309

10.3 Head Model 311

10.4 Most Popular Brain Source Localization Approaches 313

10.4.1 EEG Source Localization Using Independent Component Analysis 313

10.4.2 MUSIC Algorithm 313

10.4.3 LORETA Algorithm 317

10.4.4 FOCUSS Algorithm 318

10.4.5 Standardized LORETA 319

10.4.6 Other Weighted Minimum Norm Solutions 320

10.4.7 Evaluation Indices 323

10.4.8 Joint ICA–LORETA Approach 323

10.5 Forward Solutions to the Localization Problem 325

10.5.1 Partially Constrained BSS Method 325

10.5.2 Constrained Least-Squares Method for Localization of P3a and P3b 326

10.5.3 Spatial Notch Filtering Approach 328

10.6 The Methods Based on Source Tracking 333

10.6.1 Deflation Beamforming Approach for EEG/MEG Multiple Source Localization 333

10.6.2 Hybrid Beamforming – Particle Filtering 336

10.7 Determination of the Number of Sources from the EEG/MEG Signals 337

10.8 Other Hybrid Methods 340

10.9 Application of Machine Learning for EEG/MEG Source Localization 340

10.10 Summary 342

References 343

11 Epileptic Seizure Prediction, Detection, and Localization 351

11.1 Introduction 351

11.2 Seizure Detection 357

11.2.1 Adult Seizure Detection from EEGs 357

11.2.2 Detection of Neonatal Seizure 363

11.3 Chaotic Behaviour of Seizure EEG 366

11.4 Seizure Detection from Brain Connectivity 369

11.5 Prediction of Seizure Onset from EEG 369

11.6 Intracranial and Joint Scalp–Intracranial Recordings for IED Detection 384

11.6.1 Introduction to IED 384

11.6.2 iEED-Times IED Detection from Scalp EEG 386

11.6.3 A Multiview Approach to IED Detection 391

11.6.4 Coupled Dictionary Learning for IED Detection 391

11.6.5 A Deep Learning Approach to IED Detection 392

11.7 Fusion of EEG–fMRI Data for Seizure Prediction 396

11.8 Summary 398

References 399

12 Sleep Recognition, Scoring, and Abnormalities 407

12.1 Introduction 407

12.1.1 Definition of Sleep 407

12.1.2 Sleep Disorder 408

12.2 Stages of Sleep 409

12.2.1 NREM Sleep 409

12.2.2 REM Sleep 411

12.3 The Influence of Circadian Rhythms 414

12.4 Sleep Deprivation 415

12.5 Psychological Effects 416

12.6 EEG Sleep Analysis and Scoring 416

12.6.1 Detection of the Rhythmic Waveforms and Spindles Employing Blind Source Separation 416

12.6.2 Time–Frequency Analysis of Sleep EEG Using Matching Pursuit 417

12.6.3 Detection of Normal Rhythms and Spindles Using Higher-Order Statistics 421

12.6.4 Sleep Scoring Using Tensor Factorization 423

12.6.5 Application of Neural Networks 425

12.6.6 Model-Based Analysis 426

12.7 Detection and Monitoring of Brain Abnormalities during Sleep by EEG and Multimodal PSG Analysis 428

12.7.1 Analysis of Sleep Apnoea 428

12.7.2 EEG and Fibromyalgia Syndrome 431

12.7.3 Sleep Disorders of Neonates 431

12.8 Dreams and Nightmares 432

12.9 EEG and Consciousness 433

12.10 Functional Brain Connectivity for Sleep Analysis 433

12.11 Summary 434

References 435

13 EEG-Based Mental Fatigue Monitoring 441

13.1 Introduction 441

13.2 Feature-Based Machine Learning Approaches 443

13.2.1 Hidden Markov Model Application 443

13.2.2 Kernel Principal Component Analysis and Hidden Markov Model 444

13.2.3 Regression-Based Fatigue Estimation 444

13.2.4 Regularized Regression 445

13.2.5 Other Feature-Based Approaches 445

13.3 Measurement of Brain Synchronization and Coherency 446

13.3.1 Linear Measure of Synchronization 446

13.3.2 Nonlinear Measure of Synchronization 448

13.4 Evaluation of ERP for Mental Fatigue 451

13.5 Separation of P3a and P3b 457

13.6 A Hybrid EEG–ERP-Based Method for Fatigue Analysis Using an Auditory Paradigm 463

13.7 Assessing Mental Fatigue by Measuring Functional Connectivity 465

13.8 Deep Learning Approaches for Fatigue Evaluation 472

13.9 Summary 474

References 474

14 EEG-Based Emotion Recognition and Classification 479

14.1 Introduction 479

14.1.1 Theories and Emotion Classification 480

14.1.2 The Physiological Effects of Emotions 482

14.1.3 Psychology and Psychophysiology of Emotion 485

14.1.4 Emotion Regulation 487

14.1.4.1 Agency and Intentionality 490

14.1.4.2 Norm Violation 490

14.1.4.3 Guilt 491

14.1.4.4 Shame 491

14.1.4.5 Embarrassment 491

14.1.4.6 Pride 491

14.1.4.7 Indignation and Anger 491

14.1.4.8 Contempt 491

14.1.4.9 Pity and Compassion 492

14.1.4.10 Awe and Elevation 492

14.1.4.11 Gratitude 492

14.1.5 Emotion-Provoking Stimuli 492

14.2 Effect of Emotion on the Brain 494

14.2.1 ERP Change Due to Emotion 494

14.2.2 Changes of Normal Brain Rhythms with Emotion 497

14.2.3 Emotion and Lateral Brain Engagement 498

14.2.4 Perception of Odours and Emotion: Why Are They Related? 498

14.3 Emotion-Related Brain Signal Processing and Machine Learning 499

14.3.1 Evaluation of Emotion Based on the Changes in Brain Rhythms 500

14.3.2 Brain Asymmetricity and Connectivity for Emotion Evaluation 501

14.3.3 Changes in ERPs for Emotion Recognition 504

14.3.4 Combined Features for Emotion Analysis 504

14.4 Other Physiological Measurement Modalities Used for Emotion Study 507

14.5 Applications 510

14.6 Pain Assessment Using EEG 510

14.7 Emotion Elicitation and Induction through Virtual Reality 512

14.8 Summary 513

References 514

15 EEG Analysis of Neurodegenerative Diseases 525

15.1 Introduction 525

15.2 Alzheimer’s Disease 527

15.2.1 Application of Brain Connectivity Estimation to AD and MCI 528

15.2.2 ERP-Based AD Monitoring 532

15.2.3 Other Approaches to EEG-Based AD Monitoring 532

15.3 Motor Neuron Disease 537

15.4 Parkinson’s Disease 537

15.5 Huntington’s Disease 541

15.6 Prion Disease 542

15.7 Behaviour Variant Frontotemporal Dementia 544

15.8 Lewy Body Dementia 545

15.9 Summary 545

References 546

16 EEG As A Biomarker for Psychiatric and Neurodevelopmental Disorders 551

16.1 Introduction 551

16.1.1 History 551

16.1.1.1 Different Psychiatric and Neurodevelopmental Disorders 553

16.1.1.2 NDD Diagnosis 554

16.2 EEG Analysis for Different NDDs 554

16.2.1 ADHD 554

16.2.1.1 ADHD Symptoms and Possible Treatment 554

16.2.1.2 EEG-Based Diagnosis of ADHD 555

16.2.2 ASD 559

16.2.2.1 ASD Symptoms and Possible Treatment 559

16.2.2.2 EEG-Based Diagnosis of ASD 560

16.2.3 Mood Disorder 561

16.2.3.1 EEG for Monitoring Depression 562

16.2.3.2 EEG for Monitoring Bipolar Disorder 564

16.2.4 Schizophrenia 565

16.2.4.1 Schizophrenia Symptoms and Management 565

16.2.4.2 EEG as the Biomarker for Schizophrenia 566

16.2.5 Anxiety (and Panic) Disorder 568

16.2.5.1 Definition and Symptoms 568

16.2.5.2 EEG for Assessing Anxiety 569

16.2.6 Insomnia 571

16.2.6.1 Symptoms of Insomnia 571

16.2.6.2 EEG for Insomnia Analysis 572

16.2.7 Schizotypal Personality Disorder 572

16.2.7.1 What Is Schizotypal Disorder? 572

16.2.7.2 EEG Manifestation of Schizotypal 573

16.3 Summary 573

References 574

17 Brain–Computer Interfacing Using EEG 581

17.1 Introduction 581

17.1.1 State of the Art in BCI 584

17.1.2 BCI Terms and Definitions 585

17.1.3 Popular BCI Directions 585

17.1.4 Virtual Environment for BCI 586

17.1.5 Evolution of BCI Design 587

17.2 BCI-Related EEG Components 588

17.2.1 Readiness Potential and Its Detection 588

17.2.2 ERD and ERS 588

17.2.3 Transient Beta Activity after the Movement 593

17.2.4 Gamma Band Oscillations 593

17.2.5 Long Delta Activity 593

17.2.6 ERPs 594

17.3 Major Problems in BCI 594

17.3.1 Preprocessing of the EEGs 595

17.4 Multidimensional EEG Decomposition 597

17.4.1 Space–Time–Frequency Method 599

17.4.2 Parallel Factor Analysis 599

17.5 Detection and Separation of ERP Signals 601

17.6 Estimation of Cortical Connectivity 603

17.7 Application of Common Spatial Patterns 606

17.8 Multiclass Brain–Computer Interfacing 609

17.9 Cell-Cultured BCI 610

17.10 Recent BCI Applications 610

17.11 Neurotechnology for BCI 614

17.12 Joint EEG and Other Brain-Scanning Modalities for BCI 617

17.12.1 Joint EEG–fNIRS for BCI 617

17.12.2 Joint EEG–MEG for BCI 618

17.13 Performance Measures for BCI Systems 618

17.14 Summary 619

References 620

18 Joint Analysis of EEG and Other Simultaneously Recorded Brain Functional Neuroimaging Modalities 631

18.1 Introduction 631

18.2 Fundamental Concepts 631

18.2.1 Functional Magnetic Resonance Imaging 631

18.2.1.1 Blood Oxygenation Level Dependence 633

18.2.1.2 Popular fMRI Data Formats 635

18.2.1.3 Preprocessing of fMRI Data 635

18.2.2 Functional Near-Infrared Spectroscopy 636

18.2.3 Magnetoencephalography 640

18.3 Joint EEG–fMRI 640

18.3.1 Relation Between EEG and fMRI 640

18.3.2 Model-Based Method for BOLD Detection 642

18.3.3 Simultaneous EEG–fMRI Recording: Artefact Removal from EEG 644

18.3.3.1 Gradient Artefact Removal from EEG 644

18.3.3.2 Ballistocardiogram Artefact Removal from EEG 645

18.3.4 BOLD Detection in fMRI 652

18.3.4.1 Implementation of Different NMF Algorithms for BOLD Detection 653

18.3.4.2 BOLD Detection Experiments 654

18.3.5 Fusion of EEG and fMRI 659

18.3.5.1 Extraction of fMRI Time Course from EEG 659

18.3.5.2 Fusion of EEG and fMRI; Blind Approach 659

18.3.5.3 Fusion of EEG and fMRI; Model-Based Approach 664

18.3.6 Application to Seizure Detection 664

18.3.7 Investigation of Decision Making in the Brain 666

18.3.8 Application to Schizophrenia 666

18.3.9 Other Applications 667

18.4 EEG–NIRS Joint Recording and Fusion 668

18.5 MEG–EEG Fusion 672

18.6 Summary 672

References 673

Index 681

Saeid Sanei, PhD, DIC, FBCS, is Professor of Signal Processing and Machine Learning at Nottingham Trent University, UK, and a Visiting Professor at Imperial College London, UK. He received his doctorate in Biomedical Signal and Image Processing from Imperial College London in 1991. He is an internationally renowned expert in signal processing, biomedical signal processing, and pattern recognition.

Jonathon A. Chambers, FREng, FIEEE, DSc (Imperial), is Emeritus Professor of Signal and Information Processing within the College of Science and Engineering at the University of Leicester, UK. His research interests are focused upon adaptive signal processing and machine learning and their application in biomedicine, communications, defense, and navigation systems.