ADVANCED METHODS
OFBIOMEDICAL
SIGNAL PROCESSING
Edited by
SERGIO CERUTTI
CARLO MARCHESI
gjjjjjjjg IEEE Engineering in Medicine
-<• and Biology Society, Sponsor
IEEE Press Series in Biomedical Engineering
Metin Akay, Series Editor
IEEEIEEE Press
®WILEYA JOHN WILEY & SONS, INC., PUBLICATION
CONTENTS
Preface xvii
Contributors xxiii
Part I. Fundamentals of Biomedical Signal Processingand Introduction to Advanced Methods
1 Methods of Biomedical Signal Processing 3
Multiparametric and Multidisciplinary Integration toward a
Better Comprehension of Pathophysiological Mechanisms
Sergio Cerutti
1.1 Introduction 3
1.2 Fundamental Characteristics of Biomedical Signals 5
and Traditional Processing Approaches
1.2.1 Deterministic and Stochastic Systems and Signals 6
1.2.2 Stationary and Nonstationarity of Processes 7
and Signals
1.2.3 Gaussian and Non-Gaussian Processes 9
1.2.4 LTI Systems (Linear and Time-Invariant) 9
1.3 Link Between Physiological Modeling and Biomedical 15
Signal Processing
1.4 The Paradigm of Maximum Signal-System Integration 20
1.4.1 Integration among More Signals in the Same 21
System1.4.2 Integration among Signals Relative to Different 22
Biological Systems
1.4.3 Integration (or Data Fusion) from Signals and 24
Images
vi CONTENTS
1.4.4 Integration among Different Observation Scales 26
1.5 Conclusions 28
References 29
2 Data, Signals, and Information 33
Medical Applications of Digital Signal ProcessingCarlo Marched, Matteo Paaletti, and Loriano Galeotti
2.1 Introduction 33
2.2 Characteristic Aspects of Biomedical Signal 34
Processing
2.2.1 General Considerations Based on Actual 34
Applications2.2.2 Some Results ofthe Review 35
2.3 Utility and Quality of Applications 37
2.3.1 Input Information 37
2.3.2 Data Heterogeneity 38
2.3.3 Analysis of the Generalized Principal 44
Components2.3.4 Binary Variables 45
2.3.5 Wilson Metrics 46
2.4 Graphic Methods for Interactively Determining the 48
Most Discriminant Original Variables
2.4.1 Analysis of Homogeneity 49
2.5 Alarm Generation 53
Appendix 57
References 59
Part II. Points of View of the Physiologist and Clinician
3 Methods and Neurons 63
Gahriele E. M. Biella
3.1 What is an Object? 63
3.1.1 Different Perspectives 63
3.2 Which Object Property is Definitely Interesting? 66
3.2.1 A Short Introduction to Logic 67
3.2.2 Fragments 68
3.2.3 Emergence 69
3.2.4 Complexity 71
3.2.5 Closure 72
3.3 Are There Best Techniques? 73
3.3.1 Does a Specific Technique Influence the Data 73
Structure?
CONTENTS vii
3.3.2 Coding 73
3.3.3 Do Information Estimates Generated by a Single 77
Neuron Rely on Frequency Code?
3.4 Adaptedness of Techniques 79
References 80
4 Evaluation of the Autonomic Nervous System 83
From Algorithms to Clinical Practice
Maria Teresa La Rovere
4.1 Introduction 83
4.2 Relationship Between Heart Rate Variability and 84
Myocardial Infarction
4.3 Relationship Between Heart Rate Variability and Heart 87
Failure
4.4 Relationship Between Heart Rate and Blood Pressure 89
Variability4.5 Sudden Death Risk Stratification, Prophylactic Treatment, 91
and Unresolved Issues
4.6 The Role of Autonomic Markers in Noninvasive Risk 92
Stratification
References 94
Part III. Models and Biomedical Signals
5 Parametric Models for the Analysis of Interactions in 101
Biomedical SignalsGiuseppe Baselli, Alberto Porta, and Paolo Bolzern
5.1 Introduction 101
5.2 Brief Review of Open-Loop Identification 104
5.3 Closed-Loop Identification 107
5.3.1 Joint Process, Noise Nonco[relation, and 108
Canonical Forms
5.3.2 Direct Approach to Identification: Opening 109
the Loop5.3.3 Indirect Approach, Brief Remarks 110
5.4 Applications to Cardiovascular Control 111
5.4.1 Partial Spectra 111
5.4.3 Estimation ofthe Transfer Function (TF): 115
Limitation of the Traditional Approach5.4.4 Coherence and Causal Coherence 116
5.4.5 Closed-Loop Estimation ofthe Baroreflex Gain 117
5.5 Nonlinear Interactions and Synchronization 119
Vlii CONTENTS
5.6 Conclusion 122
References 122
6 Use of Interpretative Models in Biological Signal 127
ProcessingMauro Ursino
6.1 Introduction 127
6.2 Mathematical Instruments for Signal Processing 128
6.2.1 Descriptive Methods 128
6.2.2 The Black-Box Models 130
6.2.3 Interpretative Models 132
6.3 Examples 137
6.3.1 Mathematical Models and Signals in Intensive 137
Care Units
6.3.2 Mathematical Models and Cardiovascular 140
Variability Signals
6.3.3 Mathematical Models and EEG Signals during 142
Epilepsy6.3.4 Mathematical Models, Electrophysiology, and 145
Functional Neuroimaging6.4 Conclusions 148
References 150
7 Multimodal Integration of EEG, MEG, and 153
Functional MRI in the Study of Human
Brain ActivityFabio Babiloni, Fabrizio De Vico Fallani, and Febo Cincotti
7.1 Introduction 153
7.2 Cortical Activity Estimation from Noninvasive EEG 155
and MEG Measurements
7.2.1 Head and Source Models 155
7.2.2 The Linear Inverse Problem 157
7.2.3 Multimodal Integration of EEG and MEG Data 159
7.3 Integration of EEG/MEG and (MRI data 161
7.3.1 The Common Head Model 161
7.3.2 Percentage Change Hemodynamic Responses 162
Appendix I. Electrical Forward Solution for a Realistic Head 165
Model
Appendix II. Magnetic Forward Solution 166
References 166
CONTENTS IX
8 Deconvolution for Physiological Signal Analysis 169
Giovanni Sparacino, Gianluigi Pillonetto,
Giuseppe De Nicolao, and Claudio Cobelli
8.1. Introduction 169
8.2 Difficulties ofthe Deconvolution Problem 173
8.2.1 Ill-Posedness and Ill-Conditioning 173
8.2.2 Deconvolution of Physiological Signals 177
8.3 The Regularization Method 178
8.3.1 Deterministic Viewpoint 178
8.3.2 Stochastic Viewpoint 183
8.3.3 Numerical Aspects 186
8.3.4 Nonnegativity Constraints 187
8.4 Other Deconvolution Methods 188
8.5 A Stochastic Nonlinear Method for Constrained 190
Problems
8.6 Conclusions and Developments 194
References 195
Part IV. Time-Frequency, Time-Scale, and Wavelet Analysis
9 Linear Time-Frequency Representation 201
Maurizio Varanini
9.1 Introduction 201
9.2 The Short-Time Fourier Transform 203
9.3 Time-Frequency Resolution 207
9.4 Multiresolution Analysis 209
9.5 Wavelet Transform 210
9.6 A Generalization of the Short-Time Fourier Transform 215
9.7 Wavelet Transform and Discrete Filter Banks 220
9.8 Matching Pursuit 226
9.9 Applications to Biomedical Signals 228
9.9.1 Analysis of Spectral Variability of Heart Rate 228
9.9.2 Analysis of a Signal from a Laser Doppler 229
Flowmeter
9.10 Conclusions 230
References 231
10 Quadratic Time-Frequency Representation 233
Luca Mainardi
10.1 Introduction 233
10.2 A Route to Time-Frequency Representations 234
10.3 Wigner-Ville Time-Frequency Representation 235
X CONTENTS
10.4 Interference Terms 238
10.5 Cohen's Class 240
10.5.1 Exponential Distribution (ED) 245
10.5.2 Reduced Interference Distribution (RID) 246
10.5.3 Smoothed Pseudo Wigner-Ville (SPWV) 247
10.6 Parameter quantification 247
10.7 Applications 247
10.7.1 EEG Signal Analysis 248
10.7.2 ECG Signal Analysis 249
10.7.3 Heart Rate Variability Signal 251
10.7.4 Other Applications 253
10.8 Conclusions 254
References 254
11 Time-Variant Spectral Estimation 259
Anna M. Bianchi
11.1 Introduction 259
11.2 LMS Methods 261
11.3 RLS Algorithm 262
11.4 Comparison Between LMS and RLS Methods 264
11.5 Different Formulations of the Forgetting Factor 265
11.5.1 Varying Forgetting Factor 266
11.5.2 Whale Forgetting Factor 267
11.6 Examples and Applications 268
11.6.1 Myocardial Ischemia 269
11.6.2 Monitoring EEG Signal during Surgery 271
11.6.3 Study of Desynchronization and 272
Synchronization of the EEG Rhythms duringMotor Tasks
11.7 Extension to Multivariate Models 273
11.8 Conclusion 279
Appendix 1. Linear Parametric Models 281
Appendix 2. Least Squares Identification 281
Appendix 3. Comparison of Different Forgetting Factors 282
References 284
Part V. Complexity Analysis and Nonlinear Methods
12 Dynamical Systems and Their Bifurcations 291
Fabio Dercole and Sergio Rinaldi
12.1 Dynamical Systems and State Portraits 291
12.2 Structural Stability 300
12.3 Bifurcations as Collisions 301
CONTENTS Xi
12.4 Local Bifurcations 303
12.4.1 Transcritical, Saddle-Node, and Pitchfork 304
Bifurcations
12.4.2 Hopf Bifurcation 306
12.4.3 Tangent Bifurcation of Limit Cycles 308
12.4.4 Flip (Period-Doubling) Bifurcation 309
12.4.5 Neimark-Sacker (Torus) Bifurcation 310
12.5 Global Bifurcations 312
12.5.1 Heteroclinic Bifurcation 312
12.5.2 Homoclinic Bifurcation 312
12.6 Catastrophes, Hysteresis, and Cusp 314
12.7 Routes to Chaos 319
12.8 Numerical Methods and Software Packages 320
References 322
13 Fractal Dimension 327
From Geometry to PhysiologyRita Balocchi
13.1 Geometry 329
13.1.1 Topology 329
13.1.2 Euclidean, Topologic, and 330
Hausdorff-Besicovitch Dimension
13.2 Fractal Objects 331
13.2.1 Koch Curve, Cantor Set, and Sierpinski 331
Triangle13.2.2 Properties of Fractals 333
13.3 Fractals in Physiology 335
13.3.1 Self-Similarity of Dynamic Processes 336
13.3.2 Properties of Self-Similar Processes 337
13.4 Hurst Exponent 338
13.4.1 Rescaled Range Analysis 340
13.4.2 Methods for Computing H 341
13.5 Concluding Remarks 343
References 344
Further Reading 346
14 Nonlinear Analysis of Experimental Time Series 347
Maria Gabriella Signorini and Manuela Ferrario
14.1 Introduction 347
14.2 Reconstruction in the Embedding Space 350
14.2.1 Choosing the Time Delay t 353
14.2.2 Choosing the Embedding Dimension dE: The 354
False Neighbors Method
Xii CONTENTS
14.3 Testing for Nonlinearity with Surrogate Data 357
14.3.1 Surrogate Time Series 357
14.3.2 Artifacts 360
14.3.3 A Particular Case: The Spike Train 361
14.3.4 Test Statistics 363
14.4 Estimation of Invariants: Fractal Dimension and 364
Lyapunov Exponents
14.4.1 Lyapunov Exponents 365
14.4.2 Fractal Dimension 366
14.5 Dimension of Kaplan and Yorke 367
14.6 Entropy 368
14.7 Nonlinear Noise Reduction 372
14.8 Conclusion 373
Appendix 374
14.A1 Chaotic Dynamics 374
14.A2 Attractors 374
14.A3 Strange Attractors 375
References 375
15 Blind Source Separation 379
Application to Biomedical Signals
Luca Mesin, Ales Holobar, and Roberto Merletti
15.1 Introduction 379
15.2 Mathematical Models of Mixtures 380
15.3 Processing Techniques 382
15.3.1 PCA and ICA: Possible Choices of Distance 383
between Source Signals15.3.2 Algebraic PCA Method: Application to an 389
Instantaneous Mixing Model
15.3.3 Neural ICA Method: Application to 391
Instantaneous Mixing Model
15.4 Applications 395
15.4.1 Physiology of Human Muscles 395
15.4.2 Separation of Surface EMG Signals Generated 396
by Muscles Close to Each Other (Muscle
Crosstalk)15.4.3 Separation of Single Motor Unit Action 399
Potentials from Multichannel Surface EMG
Appendix 404
Eigenvalue Decomposition 404
Singular Value Decomposition 405
Acknowledgments'
406
References 407
CONTENTS Xiii
16 Higher Order Spectra 411
Giovanni Calcagnini and Federica Censi
16.1. Introduction 411
16.2. Higher Order Statistics: Definition and Main Properties 412
16.2.1. Observations 414
16.3. Bispectrum and Bicoherence: Definitions, Properties, 415
and Estimation Methods
16.3.1. Definitions and Properties 415
16.3.2. Bispectrum Estimation: Nonparametric and 416
Parametric Approaches
16.4. Analysis of Nonlinear Signals: Quadratic Phase 418
Coupling16.5. Identification ofLinear Systems 419
16.6. Interaction Among Cardiorespiratory Signals 420
16.7. Clinical Applications of HOS: Bispectral Index for 421
Assessment of Anaesthesia Depth
References 425
Part VI. Information Processing of Molecular Biology Data
17 Molecular Bioengineering and Nanobioscience 429
Data Analysis and Processing Methods
Carmelina Ruggiero
17.1 Introduction 429
17.2 Data Analysis and Processing Methods for Genomics 431
in the Postgenomic Era
17.2.1 Genome Sequence Alignment 432
17.2.2 Genome Sequence Analysis 432
17.2.3 DNA Microarray Data Analysis 433
17.3 From Genomics to Proteomics 435
17.4 Protein Structure Determination 435
17.5 Conclusions 437
References 437
18 Microarray Data Analysis 443
General Concepts, Gene Selection, and Classification
Riccardo Bellazzi, Silvio Bicciato, Claudio Cobelli,
Barbara Di Camillo, Fulvia Ferrazzi, Paolo Magni,
Lucia Sacchi, and Gianna Toffolo
18.1. Introduction 443
18.2. From Microarray to Gene Expression Data 446
18.2.1. Image Acquisition and Analysis 446
XiV CONTENTS
18.2.2. Preprocessing 447
18.2.3. Normalization and Data Warehousing 447
18.2.4. Technical and Biological Variability in Gene 448
Expression Data
18.2.5. Microarray Data Annotation.
449
18.3. Identification of Differentially Expressed Genes 450
18.3.1. The Fold-Change Approach 450
18.3.2. Approaches Based on Statistical Tests 451
18.3.3 Analysis of Time-Course Microarray Experiments 453
18.4. Classification: Unsupervised Methods 456
18.4.1 Distance-Based Methods 457
18.4.2 Model-Based Clustering 458
18.4.3 Template-Based Clustering 461
18.5. Classification: Supervised Methods 463
18.6. Conclusions 464
References 466
Internet Resources 471
19 Microarray Data Analysis 473
Gene Regulatory Networks
Riccardo Bellazzi, Silvio Bicciato, Claudio Cobelli,
Barbara Di Camilla, Fulvia Ferrazzi, Paolo Magni,
Lucia Sacchi, and Gianna Toffolo19.1 Introduction 473
19.2 Boolean Models 474
19.3 Differential Equation Models 476
19.4 Bayesian Models 478
19.4.1 Learning Conditional Probability Distributions 479
19.4.2 Learning the Structure of Bayesian Networks 480
19.4.3 Module Networks 482
19.4.4 Integrating Prior Knowledge 483
19.5 Conclusions 484
References 485
20 Biomolecular Sequence Analysis 489
Linda Pattini and Sergio Cerulti
20.1 Introduction 489
20.2 Correlation in DNA Sequences 489
20.2.1 Coding and Noncoding Sequences 489
20.2.2 DNA Sequence-Structure Relationship 491
20.3 Spectral Methods in Genomics 494
20.4 Information Theory 496
20.4.1 Analysis ofGenomic Sequences through Chaos 496
Game Representation
CONTENTS XV
20.5 Processing of Protein Sequences 498
20.5.1 Codification of Amino Acid Sequences 498
20.5.2 Characterization and Comparison of Proteins 499
20.5.3 Detection of Repeating Motifs in Proteins 500
20.5.4 Prediction of Transmembrane Alpha Helices 503
20.5.5 Prediction of Amphiphilic Alpha Helices 504
References 506
Part VII. Classification and Feature Extraction
21 Soft Computing in Signal and Data Analysis 511
Neural Networks, Neuro-Fuzzy Networks, and
Genetic AlgorithmsGiovanni Magenes, Francesco Lunghi, and Stefano Ramat
21.1 Introduction 511
21.2 Adaptive Networks 512
21.3 Neural Networks 514
21.3.1 Association, Clustering, and Classification 515
21.3.2 Pattern Completion 516
21.3.3 Regression and Generalization 516
21.3.4 Optimization 516
21.4 Learning 516
21.4.1 Nonsupervised Learning 518
21.4.2 Supervised Learning 523
21.5 Structural Adaptation 530
21.5.1 Statistical Learning Theory 531
21.5.2 SVM Support Vector Machines 533
21.6 Neuro-Fuzzy Networks 537
21.6.1 ANFIS Learning 539
21.6.2 Fuzzy Modeling 540
21.7 Genetic Algorithms 541
References 545
22 Interpretation and Classification of Patient Status 551
Patterns
Matteo Paoletti and Carlo Marchesi
22.1 The Classification Process 552
22.1.1 Classification Principles 552
22.1.2 Error and Risk During Classification 553
22.2 The Bayes Classifier 554
22.3 A Different Approach to Interpret (and Classify) Data: 556
Cluster Analysis22.4 Applications to Biomedical Data 557
XVi CONTENTS
22.4.1 Homogeneous Dataset 558
22.4.2 Heterogeneous Data 562
22.4.3 Dissimilarity Matrix 563
22.4.4 PAM (Partitioning Around Medoids) Algorithm 565
22.5 Visual Exploration of Biomedical Data 566
References 570
Index 571
IEEE Press Series in Biomedical Engineering