Brain-Computer Interfaces 1

Name: Brain-Computer Interfaces 1 | Methods and Perspectives
Brand: Wiley-ISTE
Price: 139.99 EUR
Availability: OnlineOnly

Methods and Perspectives

Maureen Clerc Laurent Bougrain Fabien Lotte(Herausgeber*in)

Wiley-ISTE (Verlag)

Erschienen am 14. Juli 2016

330 Seiten

E-Book

ePUB mit Adobe-DRM

Systemvoraussetzungen

978-1-119-14499-1 (ISBN)

139,99 €inkl. 7% MwSt.

Systemvoraussetzungen

für ePUB mit Adobe-DRM

E-Book Einzellizenz

Als Download verfügbar

Beschreibung

Weitere Details

Weitere Ausgaben

Inhalt

Cover
Title Page
Copyright
Contents
Foreword
Introduction
I.1. History
I.2. Introduction to BCIs
I.2.1. Classification of BCIs
I.2.2. BCI applications
I.2.3. Other BCI systems
I.2.4. Terminology
I.3. Book presentation
I.3.1. Foundations and methods
I.3.2. Reading guide
I.4. Acknowledgments
I.5. Bibliography
PART 1: Anatomy and Physiology
1: Anatomy of the Nervous System
1.1. General description of the nervous system
1.2. The central nervous system
1.2.1. The telencephalon
1.2.2. The diencephalon
1.2.3. The brain stem
1.3. The cerebellum
1.4. The spinal cord and its roots
1.5. The peripheral nervous system
1.5.1. Nerves
1.5.2. General organization of the PNS
1.5.3. The autonomic nervous system
1.6. Some syndromes and pathologies targeted by Brain-Computer Interfaces
1.6.1. Motor syndromes
1.6.2. Some pathologies that may be treated with BCIs
1.7. Conclusions
1.8. Bibliography
2: Functional Neuroimaging
2.1. Functional MRI
2.1.1. Basic principles of MRI
2.1.2. Principles of fMRI
2.1.3. Statistical data analysis: the linear model
2.1.4. Independent component analysis
2.1.5. Connectivity measures
2.2. Electrophysiology: EEG and MEG
2.2.1. Basic principles of signal generation
2.2.2. Event-related potentials and fields
2.2.3. Source localization
2.2.4. Independent component analysis
2.2.5. Time-frequency analysis
2.2.6. Connectivity
2.2.7. Statistical analysis
2.3. Simultaneous EEG-fMRI
2.3.1. Basic principles
2.3.2. Applications and data analysis
2.3.3. Connections between EEG and fMRI
2.4. Discussion and outlook for the future
2.5. Bibliography
3: Cerebral Electrogenesis
3.1. Electrical neuronal activity detected in EEG
3.1.1. Action and postsynaptic potentials
3.1.2. Resting potential, electrochemical gradient and PSPs
3.1.3. From PSPs to EEG
3.2. Dipolar and quadrupole fields
3.2.1. Field created by an ion current due to the opening of ion channels
3.2.1.1. Field created by an inflow of ions during a synapse (PSP)
3.2.1.2. Field created by an ion inflow at the axon (AP)
3.2.1.3. Field created by other neuronal activities
3.2.2. Factors determining the value of the potential created by an ion current
3.3. The importance of geometry
3.3.1. Spatial summation, closed fields and open fields
3.3.2. Effect of synapse position on the polarity of EEG
3.3.3. Effect of active areas' position
3.4. The influence of conductive media
3.4.1. Influence of glial cells
3.4.2. Influence of skull bones
3.5. Conclusions
3.6. Bibliography
4: Physiological Markers for Controlling Active and Reactive BCIs
4.1. Introduction
4.2. Markers that enable active interface control
4.2.1. Spatiotemporal variations in potential
4.2.1.1. Slow variations of average cortical potential
4.2.1.2. BP or readiness potential
4.2.2. Spatiotemporal wave variations
4.3. Markers that make it possible to control reactive interfaces
4.3.1. Sensory evoked potentials
4.3.1.1. Visual evoked potential
4.3.1.2. Other steady-state potentials: steady-state auditory evoked potential, auditory steady-state response and steady-state somatosensory evoked potential
4.3.2. Endogenous P300 potential
4.4. Conclusions
4.5. Bibliography
5: Neurophysiological Markers for Passive Brain-Computer Interfaces
5.1. Passive BCI and mental states
5.1.1. Passive BCI: definition
5.1.2. The notion of mental states
5.1.3. General categories of neurophysiological markers
5.2. Cognitive load
5.2.1. Definition
5.2.2. Behavioral markers
5.2.3. EEG markers
5.2.4. Application example: air traffic control
5.3. Mental fatigue and vigilance
5.3.1. Definition
5.3.2. Behavioral markers
5.3.3. EEG markers
5.3.4. Application example: driving
5.4. Attention
5.4.1. Definition
5.4.2. Behavioral markers
5.4.3. EEG markers
5.4.4. Application example: teaching
5.5. Error detection
5.5.1. Definition
5.5.2. Behavioral markers
5.5.3. EEG markers
5.5.4. Application example: tactile and robotic interfaces
5.6. Emotions
5.6.1. Definition
5.6.2. Behavioral markers
5.6.3. EEG markers
5.6.4. Application example: communication and personal development
5.7. Conclusions
5.8. Bibliography
PART 2: Signal Processing and Machine Learning
6: Electroencephalography Data Preprocessing
6.1. Introduction
6.2. Principles of EEG acquisition
6.2.1. Montage
6.2.2. Sampling and quantification
6.3. Temporal representation and segmentation
6.3.1. Segmentation
6.3.2. Time domain preprocessing
6.4. Frequency representation
6.4.1. Fourier transform
6.4.2. Frequency filtering
6.5. Time-frequency representations
6.5.1. Time-frequency atom
6.5.2. Short-time Fourier transform
6.5.3. Wavelet transform
6.5.4. Time-frequency transforms of discrete signals
6.5.5. Toward other redundant representations
6.6. Spatial representations
6.6.1. Topographic representations
6.6.2. Spatial filtering
6.6.2.1. Surface Laplacian
6.6.2.2. Cortical current density
6.6.3. Source reconstruction
6.6.4. Using spatial representations in BCI
6.7. Statistical representations
6.7.1. Principal component analysis
6.7.2. Independent component analysis
6.7.3. Using statistical representations in BCI
6.8. Conclusions
6.9. Bibliography
7: EEG Feature Extraction
7.1. Introduction
7.2. Feature extraction
7.3. Feature extraction for BCIs employing oscillatory activity
7.3.1. Basic design for BCI using oscillatory activity
7.3.2. Toward more advanced, multiple electrode BCIs
7.3.2.1. Spatial filtering
7.3.3. The CSP algorithm
7.3.4. Illustration on real data
7.4. Feature extraction for the BCIs employing EPs
7.4.1. Spatial filtering for BCIs employing EPs
7.5. Alternative methods and the Riemannian geometry approach
7.6. Conclusions
7.7. Bibliography
8: Analysis of Extracellular Recordings
8.1. Introduction
8.1.1. Why is recording neuronal populations desirable?
8.1.2. How can neuronal populations be recorded?
8.1.3. The properties of extracellular data and the necessity of spike sorting
8.2. The origin of the signal and its consequences
8.2.1. Relationship between current and potential in a homogeneous medium
8.2.2. Relationship between the derivatives of the membrane potential and the transmembrane current
8.2.3. "From electrodes to tetrodes"
8.3. Spike sorting: a chronological presentation
8.3.1. Naked eye sorting
8.3.2. Window discriminator (1963)
8.3.3. Template matching (1964)
8.3.4. Dimension reduction and clustering (1965)
8.3.5. Principal component analysis (1968)
8.3.6. Resolving superposition (1972)
8.3.7. Dynamic amplitude profiles of action potentials (1973)
8.3.8. Optimal filters (1975)
8.3.9. Stereotrodes and amplitude ratios (1983)
8.3.10. Sampling jitter (1984)
8.3.11. Graphical tools
8.3.12. Automatic clustering
8.4. Recommendations
8.5. Bibliography
9: Statistical Learning for BCIs
9.1. Supervised statistical learning
9.1.1. Training data and the predictor function
9.1.2. Empirical risk and regularization
9.1.3. Classical methods of classification
9.1.3.1. Linear discriminant analysis
9.1.3.2. Support Vector Machines (SVM)
9.2. Specific training methods
9.2.1. Selection of variables and sensors
9.2.2. Multisubject learning, information transfer
9.3. Performance metrics
9.3.1. Classification performance metrics
9.3.2. Regression performance metrics
9.4. Validation and model selection
9.4.1. Estimation of the performance metric
9.4.1.1. Random sampling
9.4.1.2. K-cross-validation and leave-one-out
9.4.1.3. Bootstrapping
9.4.2. Optimization of hyperparameters
9.5. Conclusions
9.6. Bibliography
PART 3: Human Learning and Human-Machine Interaction
10: Adaptive Methods in Machine Learning
10.1. The primary sources of variability
10.1.1. Intrasubject variability
10.1.2. Intersubject variability
10.2. Adaptation framework for BCIs
10.3. Adaptive statistical decoding
10.3.1. Covariate shift
10.3.2. Classifier adaptation
10.3.2.1. Sliding window retraining
10.3.2.2. Gradient descent
10.3.3. Subject-adapted calibration
10.3.3.1. Reinforcement learning
10.3.4. Optimal tasks
10.3.5. Correspondence between task and command
10.4. Generative model and adaptation
10.4.1. Bayesian approach
10.4.2. Sequential decision
10.4.3. Online optimization of stimulations
10.4.3.1. Adaptive experiments in cognitive neuroscience
10.5. Conclusions
10.6. Bibliography
11: Human Learning for Brain-Computer Interfaces
11.1. Introduction
11.2. Illustration: two historical BCI protocols
11.3. Limitations of standard protocols used for BCIs
11.4. State-of-the-art in BCI learning protocols
11.4.1. Instructions
11.4.2. Training tasks
11.4.3. Feedback
11.4.4. Learning environment
11.4.5. In summary: guidelines for designing more effective training protocols
11.5. Perspectives: toward user-adapted and user-adaptable learning protocols
11.6. Conclusions
11.7. Bibliography
12: Brain-Computer Interfaces for Human-Computer Interaction
12.1. A brief introduction to human-computer interaction
12.1.1. Interactive systems, interface and interaction
12.1.2. Elementary tasks and interaction techniques
12.1.3. Theory of action feedback
12.1.4. Usability
12.2. Properties of BCIs from the perspective of HCI
12.3. Which pattern for which task?
12.4. Paradigms of interaction for BCIs
12.4.1. BCI interaction loop
12.4.2. Main paradigms of interaction for BCIs
12.5. Conclusions
12.6. Bibliography
13: Brain Training with Neurofeedback
13.1. Introduction
13.2. How does it work?
13.2.1. Design of an NF training program
13.2.2. Course of an NF session: where the eyes "look" at the brain
13.2.3. A learning procedure that we still do not fully understand
13.3. Fifty years of history
13.3.1. A premature infatuation
13.3.2. Diversification of approaches
13.4. Where NF meets BCI
13.5. Applications
13.6. Conclusions
13.7. Bibliography
List of Authors
Index
Contents of Volume 2
Other titles from ISTE in Cognitive Science and Knowledge Management
EULA

Introduction

A Brain-Computer interface (BCI) is a system that translates a user's brain activity into messages or commands for an interactive application. BCIs represent a relatively recent technology that is experiencing a rapid growth. The objective of this introductory chapter is to briefly present an overview of the history of BCIs, the technology behind them, the terms and classifications used to describe them and their possible applications. The book's content is presented, and a reading guide is provided so that you, the reader, can easily find and use whatever you are searching for in this book.

I.1. History

The idea of being able to control a device through mere thought is not new. In the scientific world, this idea was proposed by Jacques Vidal in 1973 in an article entitled "Toward Direct Brain-Computer Communications" [VID 73]. In this article, the Belgian scientist, who had studied in Paris and taught at the University of California, Los Angeles, describes the hardware architecture and the processing he sought to implement in order to produce a BCI through electroencephalographic signals. In 1971, Eberhard Fetz had already shown that it was possible to teach a monkey to voluntarily control motor cortex brain activity by providing visual information according to discharge rate [FET 71]. These two references show that since that time, BCIs could be implemented in the form of invasive or non-invasive brain activity measurements, that is, measurements of brain activity at the neural or scalp levels. For a more comprehensive history of BCIs, the reader may refer to the following articles: [LEB 06, VAA 09].

Although BCIs have been present in the field of research for over 40 years, they have only recently come to the media's attention, often described in catchy headlines such as "writing through thought is possible" or "a man controls a robot arm by thinking". Beyond announcements motivated by journalists' love for novelty or by scientists and developers' hopes of attracting the attention of the public and of potential funding sources, what are the real possibilities for BCIs within and outside research labs?

This book seeks to pinpoint these technologies somewhere between reality and fiction, and between super-human fantasies and real scientific challenges. It also describes the scientific tools that make it possible to infer certain aspects of a person's mental state by surveying brain activity in real time, such as a person's interest in a given element of his or her environment or the will to make a certain gesture. This book also explores patients' expectations and feedback, the actual number of people using BCIs and details the material and software elements involved in the process.

I.2. Introduction to BCIs

Designing a BCI is a complex and difficult task that requires knowledge of several disciplines such as computer science, electrical engineering, signal processing, neuroscience and psychology. BCIs, whose architecture is summarized in Figure 1.1, are closed loop systems usually composed of six main stages: brain activity recording, preprocessing, feature extraction, classification, translation into a command and feedback:

- Brain activity recording makes it possible to acquire raw signals that reflect the user's brain activity [WOL 06]. Different kinds of measuring devices can be used, but the most common one is electroencephalography (EEG) as shown in Figure I.1;
- Preprocessing consists of cleaning up and removing noise from measured signals in order to extract the relevant information they contain [BAS 07];
- Feature extraction consists of describing signals in terms of a small number of relevant variables called "features" [BAS 07]; for example, an EEG signal's strength on some sensors and on certain frequencies may count as a feature;
- Classification associates a class to a set of features drawn from the signals within a certain time window [LOT 07]. This class corresponds to a type of identified brain activity pattern (for example the imagined movement of the left or right hand). A classification algorithm is known as a "classifier";
- Translation into a command associates a command with a given brain activity pattern identified in the user's brain signals. For example, when imagined movement of the left hand is identified, it can be translated into the command: "move the cursor on the screen toward the left". This command can then be used to control a given application, such as a text editor or a robot [KÜB 06];
- Feedback is then provided to the user in order to inform him or her about the brain activity pattern that was recognized. The objective is to help the user learn to modulate brain activity and thereby improve his or her control of the BCI. Indeed, controlling a BCI is a skill that must be learned [NEU 10].

Figure I.1. Architecture of a BCI working in real time, with some examples of applications

Two stages are usually necessary in order to use a BCI: (1) an offline calibration stage, during which the system's settings are determined, and (2) an online operational stage, during which the system recognizes the user's brain activity patterns and translates them into application commands. The BCI research community is currently searching for solutions to help avoid the costly offline calibration stage (see, for example, [KIN 14, LOT 15]).

I.2.1. Classification of BCIs

BCIs can often be classified into different categories according to their properties. In particular, they can be classified as active, reactive or passive; as synchronous or asynchronous; as dependent or independent; and as invasive, non-invasive or hybrid. We will review the definition of those categories, which can be combined when describing a BCI (for example a BCI can be active, asynchronous and non-invasive at the same time):

- Active/reactive/passive [ZAN 11]: an active BCI is a BCI whose user is actively employed by carrying out voluntary mental tasks. For example, a BCI that uses imagined hand movement as mental commands is an active BCI. A reactive BCI is a BCI that employs the user's brain reactions to given stimuli. BCIs based on evoked potentials are considered reactive BCIs. Finally, a BCI that is not used to voluntarily control an application through mental commands, but that instead passively analyzes the user's mental state in real time, is considered a passive BCI. An application monitoring a user's mental load in real time to adapt a given interface is a passive BCI;
- Synchronous/asynchronous [MAS 06]: user-system interaction phases may be determined by the system. In such a case, the user can only control a BCI at specific times. That kind of system is considered a synchronous BCI. If interaction is allowed at any time, the interface is considered asynchronous;
- Dependent/independent [ALL 08]: a BCI is considered independent if it does not depend on motor control. It is considered dependent in the opposite case. For example, if the user has to move his or her eyes in order to observe stimuli in a reactive BCI, then BCI is dependent (it depends on the user's ocular montricity). If the user can control a BCI without any movement at all, even ocular, the BCI is independent;
- Invasive/non-invasive: as specified above, invasive interfaces use data measured from within the body (most commonly from the cortex), whereas non-invasive interfaces employ surface data, that is, data gathered on or around the head;
- Hybrid [PFU 10]: different neurophysiological markers may be used to pilot a BCI. When markers of varied natures are combined in the same BCI, it is considered hybrid. For example, a BCI that uses both imagined hand movement and brain responses to stimuli is considered hybrid. A system that combines BCI commands and non-cerebral commands (e.g. muscular signals) or more traditional interaction mechanisms (for example a mouse) is also considered hybrid. In sum, a hybrid BCI is a BCI that combines brain signals with other signals (that may or may not emanate from the brain).

I.2.2. BCI applications

Throughout the last decade, BCIs have proven to be extremely promising, especially for handicapped people (in particular for quadriplegic people suffering from locked-in syndrome), since several international scientific results have shown that it is possible to produce written text or to control prosthetics and wheelchairs with brain activity. More recently, BCIs have also proven to be interesting for people in good health, with, for example, applications in video games, and more generally for interaction with any automated system (robotics, home automation, etc.). Finally, researchers have shown that it is also possible to use BCIs passively in order to measure a user's mental state (for example stress, concentration or tiredness) in real time and regulate or adapt their environment in response to that state.

I.2.3. Other BCI systems

Let us now examine some systems that are generally related to BCIs. Neuroprostheses are systems that link an artificial device to the nervous system. Upper limb neuroprostheses analyze electric neuromuscular signals to identify movements that the robotic limb will carry out....

Systemvoraussetzungen

Als PDF speichern Als Link merken