M.Sc. Thesis Defense Session

in

Computer Engineering (Hardware)

Tittle:

Online EEG Signal Analysis of Collected Data from ADHD Kids

By

Farzaneh Javidmehr

1st Supervisor:

Dr. Mona Ghassemian

2nd Supervisor:

Dr. Reza Khosrowabadi

Examination Committee:

Dr. MohammadReza Daliri – Dr. HamidReza Mahdiani – Dr. Keyvan Navi

Location: Room 117, Faculty of Computer Science and Engineering.

Time: Tuesday, January 24 , 2017 (5 Bahman 1395) – 15:30.

Abstract: The brain consists of nerve cells that create the neural networks and the activity of such networks produce brain waves which are affected by various factors. Neurofeed-back is a modern approach to improve the brain’s functionality by means of self-regulation. In this approach, electrical activities of the brain are captured and demon-strated as brain waves using Brain Computer Interface (BCI) systems; in order to build a bridge between the individual and the outside environment. However, to make BCIs for therapy and neurofeedback applications, it is required to acquire, analyze and process the EEG signals. Therefore, real-time collection and processing of brain signals is the main challenge in proposing BCIs, which is chosen as the aim of this research.
In this thesis, we first review the different methods of linear and non-linear pro-cessing of brain signals, and also review classification algorithms, applied in the majority of BCI schemes; then we implement the most practiced of these algorithms in an online manner. Furthermore, we study and compare the performance of three algorithms of Linear Discriminant Analysis (LDA), Support Vector Machine (SVM) and Multilayer Perceptron (MLP) for two different scenarios with real EEG data. Extracted signals for the first scenario, which is Motor Imagery (MI) are classified into two distinguished classes of right and left; meanwhile extracted signals for the second scenario, which is Steady-State Visual Evoked Potential (SSVEP) are classified into three separate classes.
Comparing the performance accuracy of these algorithms indicate that the SVM algo-rithm with RBF non-linear kernel for the motor imagery scenario and the SSVEP scenar-io has the most accuracy with 78.68% and 78.70% respectively and has the least execu-tion time after the LDA algorithm. Finally, we study two other scenarios consisting these online approaches for controlling external devices, in which the former involves communication of the MI scenario to guide a robot and the latter is a game which can be used in the treatment of children with ADHD disorder by measuring their attention lev-el.
Performance evaluation of these scenarios demonstrates the capability of such online processing methods to be applied in brain computer interfaces and neurofeedback treat-ments; and we expect a bright horizon for such technologies in near future.
Keywords: Brain Signal Processing, Online Processing, Electroencephalogram, Brain Computer Interface, Neurofeedback, Feature Classification Algorithms