Dynamic Bayesian Networks For Eeg Motor Imagery Feature Extraction / Basem El Asioty

Supervisor: Seif Eldawlatly

Thesis (M.A.)—Nile University, Egypt, 2018 .

"Includes bibliographical references"

Contents:
Chapter 1 - INTRODUCTION ............................................................................1
1.1 Overview ......................................................................................................... 1
1.2 Objectives ........................................................................................................ 2
1.3 Contributions ................................................................................................... 2
1.4 Thesis Organization .......................................................................................... 3
Chapter 2 - BRAIN-COMPUTER INTERFACE ......................................................5
2.1 Introduction .................................................................................................... 5
2.2 BCI Measurement Techniques .......................................................................... 6
2.3 Electroencephalography (EEG) .......................................................................... 8
2.4 EEG Rhythms ................................................................................................. 10
2.4.1 Delta Rhythms .................................................................................................. 10
2.4.2 Theta Rhythms ................................................................................................. 11
2.4.3 Alpha Rhythms ................................................................................................. 11
2.4.4 Mu Rhythms ..................................................................................................... 11
2.4.5 Beta Rhythms ................................................................................................... 11
2.4.6 Gamma Rhythms ............................................................................................. 11
2.5 EEG Control Signals used for BCI ..................................................................... 12
2.5.1 P300 Signal....................................................................................................... 12
2.5.2 Sensorimotor Rhythms ..................................................................................... 13
2.5.3 Visual Evoked Potential (VEP) .......................................................................... 14
2.6 EEG Processing Techniques for BCI ................................................................. 15
2.6.1 Preprocessing ................................................................................................... 15
2.6.2 Feature Extraction ............................................................................................ 19
2.6.3 Classification .................................................................................................... 23
2.7 BCI Applications ............................................................................................. 24
2.8 Summary ....................................................................................................... 28
Chapter 3 - BRAIN CONNECTIVITY ANALYSIS ................................................. 29
3.1 Introduction .................................................................................................. 29
3.2 The Concept of Brain Connectivity .................................................................. 29
3.3 Connectivity Analysis Methods ....................................................................... 32
3.3.1 Functional Connectivity Methods .................................................................... 32
3.3.2 Effective Connectivity Methods ....................................................................... 35
3.4 Connectivity Features for BCI Applications ...................................................... 38
3.5 Summary ....................................................................................................... 40
Chapter 4 - DYNAMIC BAYESIAN NETWORKS FOR BRAIN CONNECTIVITY ....... 41
4.1 Introduction .................................................................................................. 41
viii
4.2 Bayesian Network Principles .......................................................................... 41
4.3 Learning ........................................................................................................ 44
4.3.1 Network Structure Learning ............................................................................. 44
4.3.2 Parameter Learning ......................................................................................... 46
4.4 Inference with Bayesian Network ................................................................... 47
4.5 Dynamic Bayesian Network (DBN) .................................................................. 47
4.6 DBN for Brain Connectivity ............................................................................. 49
4.7 Summary ....................................................................................................... 51
Chapter 5 - EXPERIMENTAL METHODS .......................................................... 52
5.1 Introduction .................................................................................................. 52
5.2 Dataset: BCI Competition 2008 – Graz Dataset A (BCIIV2a) .............................. 53
5.3 Experimental Setup ........................................................................................ 53
5.4 Data Recording .............................................................................................. 54
5.5 Building the Training Data .............................................................................. 55
5.6 Preprocessing ................................................................................................ 56
5.6.1 Spatial Filtering ................................................................................................ 56
5.6.2 Frequency Band Filtering ................................................................................. 56
5.6.3 Discretization ................................................................................................... 57
5.7 Building Dynamic Bayesian Network............................................................... 58
5.8 Banjo Toolbox ................................................................................................ 60
5.9 Input Data ..................................................................................................... 60
5.10 Banjo Experimental Setup ............................................................................ 61
5.11 Output Data (matrix, graph) ......................................................................... 62
5.12 Dimensionality Reduction ............................................................................ 62
5.13 Combining with Band Power Features .......................................................... 63
5.14 SVM Classifier .............................................................................................. 63
5.15 Granger Causality Connectivity and DBN Connectivity Comparison ................ 64
5.16 Summary ..................................................................................................... 65
Chapter 6 - RESULTS ...................................................................................... 66
6.1 Introduction .................................................................................................. 66
6.2 Frequency Band for Connectivity Features ...................................................... 66
6.3 Connectivity Networks ................................................................................... 68
6.4 Combining Band Power and Connectivity Features ......................................... 69
6.5 Granger Causality Connectivity and DBN Connectivity Comparison .................. 73
6.6 DBN Connectivity Analysis Discussion ............................................................. 75
6.7 Summary ....................................................................................................... 76
Chapter 7 - CONCLUSION AND FUTURE WORK .............................................. 77
7.1 Summary and Contributions ........................................................................... 77
7.2 Future Work .................................................................................................. 78
PUBLICATION ................................................................................................. 80
REFERENCES .......................

Abstract:
Brain-Computer Interface (BCI) is a system that establishes a communication channel to enable the use of brain activity to interface and control computer systems. One of the most challenging applications of BCI is to identify movement direction from electroencephalography (EEG) data by developing models that discriminate between different motor imagery tasks such right hand movement versus left hand movement. Herein, we utilize graph-based tools to enhance the performance of motor imagery BCIs. Specifically, we introduce the use of Dynamic Bayesian Networks (DBNs) as efficient graphical tool that could be used to detect causal relationships among EEG electrodes during motor imagery tasks. We inferred the causal relationships between EEG electrodes during each of right and left hands imagery movements from 9 different subjects. We demonstrate how using the inferred connectivity as a feature enhances the discrimination among right and left hands imagery movements compared to using traditional band power features. Our analysis reveals a distinctive connectivity pattern manifested by an increase in the number of incoming connections to the right hemisphere motor area compared to the left hemisphere during right hand imagery movements. This pattern is reversed during left hand imagery movements. We demonstrate also using other connectivity features such as Granger Causality connectivity features for the same classification problem. Our results demonstrate how DBN connectivity features give better classification accuracy compared to Granger Causality.

Text in English, abstracts in English .