Biosignal Processing and Classification Using Computational Learning and Intelligence: Principles, Algorithms, and Applications

Biosignal Processing and Classification Using Computational Learning and Intelligence: Principles, Algorithms and Applications posits an approach for biosignal processing and classification using computational learning and intelligence, highlighting that the term biosignal refers to all kinds of signals that can be continuously measured and monitored in living beings. The book is composed of five relevant parts. Part One is an introduction to biosignals and Part Two describes the relevant techniques for biosignal processing, feature extraction and feature selection/dimensionality reduction. Part Three presents the fundamentals of computational learning (machine learning). Then, the main techniques of computational intelligence are described in Part Four. The authors focus primarily on the explanation of the most used methods in the last part of this book, which is the most extensive portion of the book. This part consists of a recapitulation of the newest applications and reviews in which these techniques have been successfully applied to the biosignals’ domain, including EEG-based Brain-Computer Interfaces (BCI) focused on P300 and Imagined Speech, emotion recognition from voice and video, leukemia recognition, infant cry recognition, EEGbased ADHD identification among others.

• Provides coverage of the fundamentals of signal processing, including sensing the heart, sending the brain, sensing human acoustic, and sensing other organs
• Includes coverage biosignal pre-processing techniques such as filtering, artifiact removal, and feature extraction techniques such as Fourier transform, wavelet transform, and MFCC
• Covers the latest techniques in machine learning and computational intelligence, including Supervised Learning, common classifiers, feature selection, dimensionality reduction, fuzzy logic, neural networks, Deep Learning, bio-inspired algorithms, and Hybrid Systems
• Written by engineers to help engineers, computer scientists, researchers, and clinicians understand the technology and applications of computational learning to biosignal processing

• Language: English
• Publisher: Academic Press
• Release date: Sep 18, 2021
• ISBN: 9780128204283

Book preview

Part 1: Introduction

Outline

Chapter 1. Introduction to this book

Chapter 2. Biosignals analysis (heart, phonatory system, and muscles)

Chapter 3. Neuroimaging techniques

Chapter 1: Introduction to this book

Alejandro A. Torres-García⁎; Omar Mendoza-Montoya†; Carlos A. Reyes-García⁎; Luis Villaseñor-Pineda⁎    ⁎Instituto Nacional de Astrofísica, Óptica y Electrónica, Puebla, Mexico

†Tecnologico de Monterrey, Monterrey, Mexico

Abstract

Nowadays, there exists a growing amount of devices designed for the measurement of the functional state of human organs, which obtain biosignals as their outputs. Additionally, novel hardware advances are producing more precise and fast devices, and in some cases, both wearable and more affordable. This fact allows the generation of a huge quantity of data. The interest in capturing these signals is due to their large information portability capabilities which are suitable to be used for several purposes like medical diagnostics, biomedical engineering, education, sports medicine, personal safety among others.

Here, both computational learning and computational intelligence offer a broad group of techniques for developing biosignals-based pattern recognition systems, which can be built from known patterns or, even, they could be useful to discover non-obvious patterns in the signals. These techniques along with the biosignal processing ones are the main core of this book whose main purview is to show the bridge between biosignal processing and pattern recognition, allowing their understanding and the developing of practical and novel applications that potentially can be created by the readers.

Finally, in this chapter, we discuss the main scope of this book along with a short description of all its chapters. Especially, we show a set of interesting applications targeted to the development of tools for computer-assisted diagnostic, rehabilitation systems, biometric authentication systems, among others.

Keywords

Biosignals; Biosignal processing; Computational intelligence; Softcomputing; Computational learning; Machine learning; Pattern recognition systems

Currently, we have a broad group of devices and techniques for sensing the functional state of organs of the human body. In addition, the trend points to the development of low-cost devices and sensors with better time-resolution for measuring these biosignals. This has encouraged the rising of a broad spectrum of applications that goes from the evident ones, in medicine, to the control of interfaces, emotion recognition, and, even, for characterizing the average behavior of a person.

There are some scenarios in which to have an automatic tool for aiding the diagnostic would be relevant. For example, in some countries, it is difficult to have a doctor or an expert in a specific organ near to the poorer communities. Also, in some cases, the precise diagnostic of a disease from a type of image or study could be hard or not evident, even, for a trained doctor. Meanwhile, in practical applications will be also needed faster and accurate methods to process and classify the signals.

Despite the promising outcomes with current technologies in some domains of the biomedical area, we still have several open issues, even, because the availability of the signals for training/testing the models also depends on the cost of acquiring them and the time required for recording them. For instance, for some biomedical applications, such as in BCI, the experimenters have to do the best effort in trying to discover something relevant and useful for daily life from small training data sets. This fact takes relevance if we consider that novel machine learning methods, such as deep learning models, are sensitive to the amount of data, having outstanding outcomes when large datasets are available, which is not the case when we work with limited availability of biomedical signals.

During the writing process of this book, we have been witnesses to the collaboration between biomedicine, bioinformatics, and artificial intelligence, especially machine learning, for approaching a big problem, such as the COVID-19 pandemic. In the beginning, the first efforts were focused on the development of tools for helping or improving the diagnostic process using non-invasive measurements, such as X-Ray, CT, and audio cough units. Later on, the efforts were focused on aiding the development of a vaccine. As it is well known, after Chinese scientists decoded the sequence of the coronavirus SARS-CoV-2 the race to get a vaccine quickly started. Before the AI era, the decoding process took many years, even decades. Vaccines for lethal virus, like the one causing Ebola, started to appear after 5 years, whereas the one against the SARS-CoV-2 started very soon after the code sequence was released. This accelerated development was due to the technological advances and particularly not only in computational hardware but rather in the advances in the fields of machine learning along with computational intelligence, which have helped to make fast drugs and vaccines design as well as massive simulations in a matter of days. Previously the latter were not possible to do or lasted months to be completed.

In some moments, the number of patients was too high for the number of available expert clinicians, which had not been seen in modern times. That one along with the difficulty to find a pattern within the available measurements made evident the necessity of having machine-learning-based tools for getting confident outcomes for time optimization in some steps during monitoring of some disease.

1.1 Brief description of the contents of this book

This book presents the fundamentals necessary to understand the automatic processing of the signals emitted by a living being—mainly the human being—when performing a certain activity. These signals, or biosignals, measure chemical, electrical or mechanical activity during a given period. Once these biosignals are acquired, they are processed to identify patterns to discriminate between different activities or states. The identification of these patterns can be automated by taking advantage of the capabilities of today's computers. In general, this process starts with a transformation of the biosignals, so that a computer can process them. These data are then used to feed a machine learning algorithm. As a result of this last inductive procedure, a discriminative model is built to identify the desired activities or states in new scenarios. This book describes all the steps involved in this process, describing the techniques and approaches used depending on the particularities of each biosignal, as well as the various computational learning methods currently in use.

The first chapters describe the different biosignals emitted by humans, their characteristics, and the methods used to capture these biosignals. Chapter 2 describes how the biosignals associated with heartbeat, vocal emission, and muscle movement are acquired. Chapter 3 describes the basics of neuroimaging, which allows us to create an image of the brain activity.

Subsequently, the second part of the book presents, through Chapters 4 and 5, the different techniques to extract the features depending on the biosignals to be processed, as well as the techniques to select those features with greater relevance to achieve an appropriate discrimination.

The third part describes the main computational learning approaches that from the characterization of the biosignals can be applied to induce classification models. In this third part, Chapter 6 presents a brief introduction of these computational learning approaches, and Chapter 7 the fundamentals for a proper evaluation of the resulting models.

The fourth part of the book presents the main approaches to computational intelligence: Chapter 8 describes fuzzy logic and fuzzy systems; Chapter 9 presents the fundamentals of neural networks and describes deep learning; Chapter 10 introduces two new neural network approaches: spiking neural networks and dendrite morphological neural networks; finally, Chapter 11 describes the area of bio-inspired computation and presents different algorithms used in the field of biosignals.

Once all these elements have been described, the book continues with a series of chapters, which describe solutions, or present literature reviews, to specific problems, all based on biosignal processing.

1.2 What are the applications and reviews presented in this book?

In this book, a set of applications and reviews will be presented aiming to provide a better idea of how computational learning and intelligence are applied in the biosignal processing and analysis domain.

From a point of view of the biosignals, we will present in Part 5 how the following biosignals are analyzed with different purposes: EEG (Chapter 12, Chapter 13, Chapter 14, and Chapter 21), fNIRS (Chapter 18), FMRI (Chapter 22), human acoustics (voice in Chapter 15 and infant cry in Chapter 17), ECG (Chapter 16), and EMG (Chapter 19). Also, the images of bone marrow cells Chapter 20 and facial expressions (Chapter 15) will be analyzed in this book.

Before the unfortunate COVID-19 arising, most of the efforts of the biomedical-engineering-related scientists had been focused on the analysis of brain imaging sources; this could be explained due to brain-related initiatives having been a relevant trend during the last decade.

Some examples of these are, the BRAIN Initiative¹ (Brain Research through Advancing Innovative Neurotechnologies), the Human Brain Project,² and the Human Connectome Project.³

It is important to highlight that this trend is also observed in the number of brain-related Chapters in Part 5 Applications and reviews of the book, such as Chapters 12, 13, 14, 18, 21, and 22.

As to the final purposes of these chapters, these go from the diagnostic and classification of leukemia Chapter 20, cardiac arrhythmias (Chapter 16) and baby's diseases (Chapter 17), psychological disorders (such as attention Deficit/Hyperactivity Disorder (ADHD) Chapter 21 and schizophrenia patients dataset Chapter 22), rehabilitation (Chapter 19), practical applications, such as interfaces (Chapter 12 and Chapter 13), to emotion recognition (Chapter 15) and biometric identifier (Chapter 14).

1.3 Who should read this book?

This book is intended for students, professors, and scientists motivated to learn and put into practice signal processing principles and computational intelligence for biomedical data analysis. We expect anyone interested in recent trends in applied artificial intelligence for biomedical engineering will find this book a valuable reference source. At the same time, students and researchers who are working or would like to work with biosignals may discover in this book new ideas for their studies and experiments or adopt the techniques described in the applications included in the last chapters.

The content of this book is appropriate for advanced undergraduate students, graduate students, and researchers in engineering and computer science. Biomedical engineers who are not familiar with artificial intelligence can learn from this book the basics of computational intelligence and apply these concepts to their academic or professional projects. Likewise, readers who already know about artificial intelligence and want to work with biosignals can learn from this book the theory behind biosignal processing and feature extraction techniques. Finally, this book may also help students from other disciplines who would like to learn more about how artificial intelligence and biosignal processing work.

To read this book, it is not necessary to have prior knowledge of biosignals or computational intelligence. The prerequisites for this text are elementary probability theory, basic statistics, linear algebra, logic, and computational algorithms. Although the most challenging topics may require some knowledge of numerical optimization and advanced signal processing, the techniques and algorithms explained in this book are presented intuitively without developing the mathematical details. The references included at the end of each chapter may help those readers who want to learn the implementation details and the mathematical aspects of the described methods.

This book has been written to encourage the use of modern computational techniques for the analysis and processing of biomedical signals. To achieve this goal, we consider that it is essential to address both the biological perspective of biosignals and the computational aspects of signal processing and computation intelligence. This text is an invitation to the community of researchers and students to continue developing new applications based on biosignals and biomedical information.

---

¹  

"https://braininitiative.nih.gov/."

²  

"https://www.humanbrainproject.eu/en/."

³  

"http://www.humanconnectomeproject.org/."

Chapter 2: Biosignals analysis (heart, phonatory system, and muscles)

Rita Q. Fuentes-Aguilar⁎; Humberto Pérez-Espinosa‡; María A. Filigrana-de-la-Cruz†,§    ⁎Tecnologico de Monterrey, Escuela de Ingeniería y Ciencias, Guadalajara, Mexico

†Instituto Nacional de Astrofísica, Óptica y Electrónica, Puebla, Mexico

‡CICESE-UT³, Tepic, Mexico

§Centro de Investigación y Estudios Avanzados del Instituto Politécnico Nacional (CINVESTAV), Mexico City, Mexico

Abstract

The analysis of a system starts by identifying the inputs and outputs of the phenomena under study. If the system is visualized as a complex object collection, whose parts or compounds interacting together to achieve an objective, then the more information about its parts we have, the better we can understand it. In the field of biomedical engineer, the physical phenomena that give information about the systems of interest are naturally biological. The main goal of a bio-engineer is to study these signals to solve a medical problem, to increase the understanding about it or, to improve the way diagnosis is done. Nowadays, the field of medical data processing is experiencing rapid expansion. This is due to the advances in integrated circuits, technology, software engineer and the increasing computational tools based in numerical methods and artificial intelligence. This chapter presents the physiological principles of biosignals and their relationship with the way they are measured. Also, an explanation about the electrophysiological basis and the transducers, as well as the instrumentation system used to record them can be found in every section.

Keywords

Biosignals; Electrophysiology; Biological measurement; Electrocardiogram; ECG leads; Heart conduction system; Echocardiogram; Imaging ultrasound modes; Electromyography; Infant cry; Speech analysis; Paralinguistic analysis

2.1 Introduction

The data of a biological phenomenon that allows describing a system can be of electrical, mechanical, acoustic, thermal, optical nature, and so forth. All of these are referred to as biosignals. In brief, to determine how a system behaves, biomedical data is acquired by measuring these signals. A biosignal is, then, any signal that can be measured and recorded from biological beings. A biosignal can be electrical or non-electrical. The most common of the biosignals are bioelectrical, usually taken to be electric currents produced by the sum of electrical potential differences across a specialized tissue, organ or cell system, such as the nervous system. These signals are measured with a differential amplifier, which registers the difference between two electrodes attached to the skin. Electrical currents and changes in electrical resistances across tissues can also be measured from plants. Examples of electrical biosignals are electroencephalogram (EEG), electrocorticogram (ECoG), electrocardiogram (ECG), electromyogram (EMG), electrogastrogram (EGG), electroretinogram (ERG), electrooculogram (EOG), phonocardiogram (PCG), galvanic skin response (EDA), among others. Metabolic signals (such as infrared measurements), gases interchange, photopletismography, emotion recognition, sound identification, or odor identification are some examples of nonelectrical biosignals. A heart signal is a record of electrical activity of the heart. The contraction activity of the heart is driven by the electric natural pacemaker (cardiac cells that have the property of automatic generation of action potentials). Myocardial cells have the unique property of transmitting active potentials from one cell to an adjacent cell by means of direct current spread. Another biosignal that comes from the heart is the ballistocardiogram (BCG). A ballistocardiography is the noninvasive record of repetitive motions from the sudden ejection of blood into the great vessels with each heartbeat. It is caused by the mechanical movement of the heart. Complementary to the BCG, the phonocardiogram (PCG) is the record of sounds and murmurs made during a cardiac cycle by various cardiac structures pulsing and moving blood. Both biosignals, BCG and PCG are measured using piezoelectric transducers as well as blood pressure. While ECG is one of the most studied of the biosignals because of its deterministic nature, periodic response and wide studied characteristics, the other given examples are still under study and search for characterization to use them in diagnosis and other applications. A brain signal is a record of the electrical signal generated by the cooperative action of brain cells, or more precisely, the time course of neuronal extracellular field potentials generated by their synchronous action. If the signal can be measured by means of electrodes placed on the scalp or directly on the cortex, the obtained signals are called EEG; in the case of the use of invasive electrodes, it is called ECoG or iEEG (intracranial EEG). Brain electric field generated as a response to external or internal stimulus is called an event-related potential (ERP). Electric fields measured intracortical with electrodes implanted in the brain structures were named local fields potentials (LFP). An emerging technology for the direct monitoring of brain activity is the functional near-infrared spectroscopy (fNIRS). It is a non-invasive method to record the hemodynamic response that use, instead of electrodes, optical sensors in the range of red and infrared spectrum. Near-infrared spectrum light (700–900 nm) is the optical window, in which haemoglobin (Hb) and deoxygenated-haemoglobin (deox-Hb) are stronger absorbers of light. The differences between the absorption of Hb and deox-Hab allow the measurement of changes in oxygen concentration in blood. This property makes it possible to detect when a cerebral region is presenting activity. The electromyogram (EMG) is a record of electrical muscle activity. In the book, only EMG of striated muscles will be discussed, the electrical activity of specific smooth muscles, such as stomach (electrogastrogram (EGG)), will not.

On the other hand, there is a great variety of acoustic biosignals, both originated by internal organs, such as the heart, lungs, and intestines, and originated by the speech apparatus, such as speech, crying, and coughing. This type of biosignal includes essential information about a person's current state, indicating pathologies, and emotional or physical state. In this chapter, two types of acoustic biosignals generated by the human phonatory system are addressed, infant crying and speech. Both biosignals transmit information about the state of health, emotional state and physical characteristics of the individual. Capturing the acoustic signal of crying and speech is very simple and does not require sophisticated equipment, as it is usually done with a conventional microphone. The analogue signal is converted to digital using a sample rate of 16,000 hertz, which is enough to capture the frequency range of the human phonatory system's