E-Book, Englisch, 258 Seiten
Gandhi Brain-Computer Interfacing for Assistive Robotics
1. Auflage 2014
ISBN: 978-0-12-801587-2
Verlag: Elsevier Science & Techn.
Format: EPUB
Kopierschutz: 6 - ePub Watermark
Electroencephalograms, Recurrent Quantum Neural Networks, and User-Centric Graphical Interfaces
E-Book, Englisch, 258 Seiten
ISBN: 978-0-12-801587-2
Verlag: Elsevier Science & Techn.
Format: EPUB
Kopierschutz: 6 - ePub Watermark
Vaibhav Gandhi (author) received a First Class (Dist.) degree in Instrumentation & Control engineering in 2000, a First Class (Dist.) Masters degree in Electrical engineering in 2002 and a Ph.D. degree in Computing & Engineering in 2012. He was a recipient of the UK-India Education & Research Initiative (UKIERI) scholarship for his Ph.D. research in the area of Brain-Computer Interface for assistive robotics carried out at the Intelligent Systems Research Center, University of Ulster, UK and partly at IIT Kanpur, India. His Ph.D. focused on quantum mechanics motivated EEG signal processing, and an intelligent adaptive use-centric human-computer interface design for real-time control of a mobile robot for BCI users. His post-doctoral research involved work on shadow-hand multi-fingered mobile robot control using EMG/muscle signals, with contributions in the 3D printing aspects of a robotic hand.He joined the department of Design Engineering & Mathematics, School of Science & Technology, Middlesex University London in 2013, where he is currently Lecturer in Robotics, Embedded Systems and Real-time Systems.His research interests include brain-computer interfaces, biomedical signal processing, computational intelligence and neuroscience, use-centric graphical user interfaces, and assistive robotics.
Autoren/Hrsg.
Weitere Infos & Material
1;Front Cover;1
2;Brain–Computer Interfacing for Assistive Robotics;4
3;Copyright Page;5
4;Contents;6
5;List of Figures;8
6;List of Tables;12
7;Preface;14
8;Acknowledgments;18
9;List of Acronyms;20
10;1 Introduction;24
10.1;1.1 Introduction;24
10.2;1.2 Rationale;27
10.3;1.3 Objectives;29
11;2 Interfacing Brain and Machine;30
11.1;2.1 Introduction;30
11.2;2.2 The Brain and Electrode Placement;30
11.2.1;2.2.1 EEG Wave Rhythms;32
11.3;2.3 Operational Techniques in BCI;34
11.3.1;2.3.1 Synchronous BCI;34
11.3.2;2.3.2 Asynchronous BCI;36
11.4;2.4 Data Acquisition;37
11.4.1;2.4.1 The Basics of Data Acquisition;37
11.4.1.1;2.4.1.1 SMR BCIs;42
11.5;2.5 Preprocessing: A Signal Enhancement Requirement along with Noise Reduction;44
11.5.1;2.5.1 Referencing Method;45
11.5.2;2.5.2 Principal Component Analysis [PCA];46
11.5.3;2.5.3 Independent Component Analysis [ICA];47
11.5.4;2.5.4 Common Spatial Patterns [CSP];48
11.5.5;2.5.5 Neural time series prediction preprocessing [NTSPP];49
11.5.6;2.5.6 Kalman Filter;50
11.5.7;2.5.7 Autoregressive (AR) Modeling;51
11.5.8;2.5.8 Summary;52
11.6;2.6 Feature Extraction;52
11.6.1;2.6.1 Band Power Features;52
11.6.2;2.6.2 Power Spectral Density Features;53
11.6.3;2.6.3 Time-frequency Method;54
11.6.4;2.6.4 Hjorth Features;55
11.6.5;2.6.5 Hilbert-Huang Transform;56
11.6.6;2.6.6 Summary;56
11.7;2.7 Classification;57
11.7.1;2.7.1 Linear Discriminant Analysis Classifier;57
11.7.2;2.7.2 Support Vector Machine Classifier;59
11.7.3;2.7.3 Regression Classifier;60
11.7.4;2.7.4 Summary;60
11.8;2.8 Post-processing;62
11.8.1;2.8.1 Confidence Intervals and Rejection;63
11.8.2;2.8.2 Multiple Thresholding with Windowing Concept;63
11.8.3;2.8.3 De-biasing;64
11.8.4;2.8.4 Error Potential (ErrP);64
11.9;2.9 Validation and Optimization Techniques;65
11.9.1;2.9.1 Cross-validation;65
11.9.2;2.9.2 Genetic Algorithm;66
11.9.3;2.9.3 Particle Swarm Optimization;68
11.10;2.10 Graphical User Interface [GUI];70
11.10.1;2.10.1 The Necessity for an Interface;71
11.10.2;2.10.2 Expectations from a Good GUI Design;75
11.10.3;2.10.3 Recent Developments in BCI GUI Design;76
11.10.3.1;2.10.3.1 GUI Designs for Speller Applications;76
11.10.3.1.1;2.10.3.1.1 P300-based Virtual Keyboard;76
11.10.3.1.2;2.10.3.1.2 MI-based Virtual Keyboard;77
11.10.3.1.3;2.10.3.1.3 MI-based hex-o-spell Typewriter Interface;78
11.10.3.2;2.10.3.2 GUI Designs for Robot Control;78
11.10.3.2.1;2.10.3.2.1 P300-based GUI with Predefined Fixed Locations;78
11.10.3.2.2;2.10.3.2.2 MI-based Robot Control Interface;79
11.11;2.11 Strategies in BCI Applications;80
11.11.1;2.11.1 Shared Control BCI System;80
11.12;2.12 Performance Measures of a BCI System;82
11.13;2.13 Conclusion;85
12;3 Fundamentals of Recurrent Quantum Neural Networks;88
12.1;3.1 Introduction;88
12.2;3.2 Postulates of Quantum Mechanics;88
12.3;3.3 Quantum Mechanics and the Schrodinger Wave Equation;89
12.3.1;3.3.1 A Classical vs. Quantum Register;92
12.3.2;3.3.2 Quantum Neural Network;93
12.3.2.1;3.3.2.1 Compelling Motivation Towards the Quantum Filtering Approach;94
12.4;3.4 Theoretical Concept of the RQNN Model;96
12.5;3.5 Traditional RQNN-Based Signal Enhancement;98
12.5.1;3.5.1 Pseudocode for the RQNN Model;101
12.5.2;3.5.2 RQNN Parameters;101
12.5.3;3.5.3 Filtering Simple Signals;103
12.5.3.1;3.5.3.1 Method and Performance Analysis;103
12.5.3.2;3.5.3.2 Concluding Remarks;107
12.6;3.6 Revised RQNN-Based Signal Enhancement;107
12.6.1;3.6.1 Pseudocode for the Revised RQNN Model;109
12.6.2;3.6.2 Understanding the Parameters for the Revised RQNN Model;109
12.6.3;3.6.3 Numerical Implementation;110
12.6.4;3.6.4 Filtering Simple Signals;111
12.6.4.1;3.6.4.1 Method and Performance Analysis;111
12.7;3.7 Discussion;113
12.8;3.8 Conclusion;116
13;4 The Proposed Graphical User Interface (GUI);118
13.1;4.1 Introduction;118
13.2;4.2 Overview of the Proposed GUI Within the BCI Framework;120
13.2.1;4.2.1 Interface for the Mobility Control Application;123
13.2.1.1;4.2.1.1 Supervised Mobility Control Interface (Non-Adaptive Form);123
13.2.1.2;4.2.1.2 Supervised Mobility Control Interface (Adaptive Form);125
13.2.1.2.1;4.2.1.2.1 The iAUI Architecture;125
13.2.1.2.2;4.2.1.2.2 Flowchart and State Machine Diagram;127
13.2.1.2.3;4.2.1.2.3 iAUI Operation in an Example Scenario;127
13.2.1.3;4.2.1.3 Autonomous Mobility Control Interface (MOB);132
13.2.2;4.2.2 Interface for Arm Control Applications;133
13.3;4.3 Interfacing MATLAB and Visual Basic;136
13.4;4.4 Conclusion;137
14;5 Recurrent Quantum Neural Network (RQNN)-Based EEG Enhancement;140
14.1;5.1 Introduction;140
14.2;5.2 Traditional RQNN Model for EEG Enhancement;143
14.2.1;5.2.1 EEG Filtering without Scaling;143
14.2.2;5.2.2 Scaling the EEG Prior to Filtering;144
14.3;5.3 Revised RQNN Model for EEG Signal Enhancement;145
14.3.1;5.3.1 Scaling the EEG Prior to Filtering (with a Large Number of Spatial Neurons);149
14.3.2;5.3.2 Scaling the EEG Prior to Filtering (Reduced Number of Spatial Neurons);153
14.4;5.4 Towards Subject-Specific RQNN Parameters;163
14.4.1;5.4.1 Two-Step Inner-Outer Five-Fold Cross-Validation for RQNN Parameter Selection;164
14.4.1.1;5.4.1.1 The Method;164
14.4.1.2;5.4.1.2 Results and Analysis;167
14.4.1.3;5.4.1.3 Concluding Remarks;170
14.5;5.5 Discussion;170
14.6;5.6 Conclusion;172
15;6 Graphical User Interface (GUI) and Robot Operation;174
15.1;6.1 Introduction;174
15.2;6.2 The EEG Acquisition Process;175
15.3;6.3 RQNN-Based EEG Signal Enhancement;177
15.4;6.4 Autonomous and Supervised GUI Operation;179
15.4.1;6.4.1 Maneuvering the Mobile Robot Under a 100% BCI Accuracy Assumption;180
15.4.2;6.4.2 Evaluating the Interface Designs;184
15.4.2.1;6.4.2.1 Evaluation Quantifiers;184
15.4.2.2;6.4.2.2 Comparing the Interfaces;190
15.5;6.5 Maneuvering the Simulated Mobile Robot Using Only MI EEG;192
15.5.1;6.5.1 Training Paradigm;192
15.5.2;6.5.2 Methodology;193
15.5.3;6.5.3 Results and Discussion;194
15.5.4;6.5.4 Concluding Remarks;202
15.6;6.6 Maneuvering the Physical Mobile Robot Using Only MI EEG;203
15.6.1;6.6.1 Methodology;203
15.6.2;6.6.2 Results and Discussion;205
15.6.3;6.6.3 Concluding Remarks;208
15.7;6.7 Conclusion;208
16;7 Conclusion;210
16.1;7.1 Contributions of the Book;211
16.1.1;7.1.1 Investigation of QM and SWE for Filter Development;211
16.1.2;7.1.2 Understanding the Parameters of the RQNN Models;212
16.1.3;7.1.3 Tuning/Selecting RQNN Model Parameters;212
16.1.4;7.1.4 Real-Time Implementation of the RQNN Model for EEG Signal Enhancement;212
16.1.5;7.1.5 Investigation into GUIs for Use in BCI Systems;213
16.1.6;7.1.6 Intelligent Adaptive User Interface (iAUI) for Mobility Control;214
16.1.7;7.1.7 Adaptive User Interface for Robot Arm Control;214
16.2;7.2 Future Research Directions;215
16.2.1;7.2.1 Tuning/Selecting the RQNN Model Parameters;215
16.2.2;7.2.2 Three-Class Classifier;215
16.2.3;7.2.3 Hybrid BCI Systems;216
16.2.3.1;7.2.3.1 Hybrid BCI (SSVEP+ERD/ERS);217
16.2.3.2;7.2.3.2 Hybrid BCI (EEG+Eye Tracker System);218
16.3;7.3 Conclusion;222
17;Appendix A: Understanding Evaluation Quantifiers for the Proposed Interface;224
18;Bibliography;234
19;Index;254
Introduction
Verbal or non-verbal information exchange is the basis of human communication. However, some people lose this fundamental ability of communication through accidents or inherited neuromuscular disorders. In the absence of methods for repairing or restoring function due to disease or damage, various alternatives in the form of assistive devices to enable individuals to communicate with and control their environment have been developed. The brain–computer interface (BCI), i.e., electroencephalography (EEG)-based communication, is a new way of controlling devices that does not require eye movement or muscle activity. This chapter introduces the various components of a typical BCI system and explains each component’s importance and function within the complete BCI system.
Keywords
Brain–computer interface; electroencephalography; graphical user interface; motor imagery; signal filtering
1.1 Introduction
Verbal or non-verbal information exchange is the basis of human communication. However, some people lose this fundamental ability of communication because of accidents or inherited neuromuscular disorders. The purpose of the work presented in this book is to contribute to the development of novel methods to allow people to regain freedom of movement/communication by way of controlling devices directly with their brain, bypassing the normal communication channels.
The human brain is estimated to contain about 100 billion neurons [1–4]. The spinal cord acts as an intermediate cable that carries information to and from our brain to control various body parts and their movements. People with an injury to the spinal cord are still able to generate the output signals from the brain, but these signals do not reach the specific body parts because the intermediate spinal cable is damaged. Several technologies using a joystick, head movement, eye gazing and many more may help a physically challenged person to control a robotic device or a wheelchair [5–9]. However, these techniques require the use of partial movement control through the hand, head or eyes etc., and therefore make the control issue less complicated. The issue becomes more challenging when people with complete loss of control over their voluntary muscles are involved, a condition generally known as locked-in syndrome [10,11], in which people are unable to speak and move but are conscious and can think and reason. A number of neurological diseases such as stroke1, severe cerebral palsy2, motor neuron disease (MND)3, amyotrophic lateral sclerosis (ALS), and encephalitis4 can result in such severe motor paralysis [12]. Many of these diseases can lead to restrictions in communication capacity. A brain–computer interface (BCI) can enable such physically challenged people to achieve greater independence by making technology accessible. BCI technology provides an alternative communication channel between the human brain (that does not depend on the brain’s normal output channels of the peripheral nerves and muscles) and a computer [13–21]. The three most commonly discussed diseases/injuries cited in the BCI literature as being a case of locked-in syndrome are ALS, high spinal cord injury and brain stem stroke [16,22–24].
• Patients suffering from ALS can undergo severe physical impairment due to the degeneration of nerve cells that control the voluntary muscles. In the later stages of ALS, the most basic human actions are affected, including speech, swallowing and breathing [25].
• Spinal cord injury (SCI) can result in damage to myelinated fiber tracts or the nerve roots that carry the signals to and from the brain [25]. In complete SCI, most of the motor functions and sensation below the neurological level are affected or completely lost [26]. SCI has a global annual incidence of 15–40 cases per million population [27] and less than 5% of people suffering from SCI recover locomotion [26].
• Brain stem stroke can be fatal, as the brain stem controls many of the basic and fundamental activities for life, such as breathing, heart rate, blood pressure, swallowing and eye movement [28]. People with severe brain stem stroke may also enter into a locked-in state and lose motor functions [29].
BCI (i.e., electroencephalography [EEG])-based communication produces new channels for controlling devices which would not be possible through the modes of communication that require eye movement or some muscle activity. Hans Berger performed a systematic study of the electrical activity of the human brain, and developed the EEG5. The first scientific literature referring to communication between the brain and the computer dates back to the early 1970s, and is due to Vidal [18], who suggested the feasibility of direct brain communication. To achieve this, the intent of the user must be extracted from the brain via the EEG or brain waves.
A typical BCI scheme generally consists of a data acquisition system, preprocessing of the acquired signals, the feature extraction process (FEP), classification of the features and finally the control interface and device controller, as shown in Figure 1.1. The EEG signals are acquired by mounting electrodes on the scalp of the user. These raw EEG signals have very low amplitude [30], very low signal-to-noise (SNR) ratio and considerable noise contamination. Preprocessing is carried out to obtain cleaner EEG signals by removing the unwanted components embedded in the EEG, which can considerably reduce the computational load on the rest of the BCI components.
Figure 1.1 Basic functional block diagram of a simple BCI system.
The work presented in this book has focused on the preprocessing stage for signal enhancement and extracting more motor imagery (MI)6 (mental imagination of movement) [31] related information from the acquired noisy EEG. These raw EEG signals are considered as a realization of a random or stochastic process [32]. When an accurate description of the signal is not available, a stochastic filter can be designed based on probabilistic measures. Therefore, the approach undertaken in this book is to use the concepts from quantum mechanics (QM) and the Schrodinger wave equation (SWE). A recurrent quantum neural network (RQNN) is constructed by using a layer of neurons within the neural network framework by computing a time-varying probability density function () of the noisy input signal (cf. Chapter 3). This evolves recurrently under the influence of the SWE and helps to enhance the EEG (cf. Chapter 5).
Features are extracted from the filtered EEG. The classifier interprets these extracted features to categorize the input signal into a designated output class. The control interface or the graphical user interface (GUI) further interprets the classifier output in the form of a command to be sent to the controlled device. The GUI also provides appropriate feedback information to the BCI user [7], and can quicken the issuance of the command from the BCI user to the device that is controlled. A two-class BCI system has two output classes in the form of a left hand MI or a right hand/foot MI. But the task of maneuvering a mobile robot requires commands in the form of forward, left, right, backward and start/stop, using just the two-class information; i.e., there is very limited communication bandwidth. However, given the inherent higher accuracy compared to multi-class BCIs, this book focuses on utilizing a two-class BCI. This book proposes an intelligent, adaptive and user-centric interface design that plays a major role in compensating for the low bandwidth of a two-class BCI and simultaneously capitalizes on the intrinsic higher accuracy characteristic that is typical of a two-class BCI system (cf. Chapter 4 and Chapter 6).
In summary, this book outlines the recent developments in MI-based BCI, specifically focusing on reviewing the existing signal processing and classification methodologies, as well as different interface designs for a BCI system. It proposes an alternative nature-inspired information processing approach based on the concepts from QM, which is referred to as the RQNN model and is utilized for EEG signal enhancement (cf. Chapter 3 and Chapter 5). It also proposes an intelligent user interface design (cf. Chapter 4) which is customized to provide effective control of a wheelchair/mobile robot and a robot arm for a quicker communication process (cf. Chapter 6).
1.2 Rationale
BCI technology has not yet reached a critical level of acceptance even forty years after its inception. The challenges in this domain begin right at the stage of acquiring the EEG signals from the brain. An EEG is recorded non-invasively, so it is a mixture of the signal of interest from the activity of the underlying neural networks and an unknown amount of noise. Therefore, the raw signals need to be filtered in order to obtain cleaner EEG signals. Several groups work in the field of EEG filtering [19,33–39]. Most of their approaches involve subject-specific parameters, which, if tuned properly, can enhance the performance of an individual subject in terms of the classification accuracy (CA) [35,40]. However, these frequency-selective techniques lead to an unknown amount of loss of information from the...
