| Abstract | Electroencephalography (EEG) is one of the most clinically and scientifically exploited signals recorded from humans. Hence, its measurement plays a prominent role in brain studies. In particular, the examination of EEG signals has been recognized as the most preponderant approach to the problem of extracting knowledge of the brain dynamics. EEG recordings are particularly important in the diagnosis of epilepsy and in brain computer interface (BCI). In BCI systems, EEG signals help to restore sensory and motor functions in patients who have severe motor disabilities. Analysing EEG signals is very important both for supporting the diagnosis of brain diseases and for contributing to a better understanding of cognitive process.
Although EEG signals provide a great deal of information about the brain, research in classification and evaluation of these signals is limited. Even today the EEG is often examined manually by experts. Therefore, there is an ever-increasing need for developing automatic classification techniques to evaluate and diagnose neurological disorders. Classification techniques can help to differentiate EEG
segments and to decide whether a person is healthy. A big challenge is for BCI systems to correctly and efficiently identify different EEG signals of different motor imagery (MI) tasks using appropriate classification algorithms to assist motor disabled patients in communication. In this dissertation, we aim to develop methods for the analysis and
classification of epileptic EEG signals and also for the identification of different categories of MI tasks based EEG signals in BCI’s development. In order to classify epileptic EEG signals, we propose two methods, simple
sampling technique based least square support vector machine (SRS-LS-SVM) and clustering technique based least square support vector machine (CT-LS-SVM). The experimental results show that both algorithms perform well in the EEG signal classification and the CT-LS-SVM method takes much less execution time compared to the SRS-LS-SVM technique. The research findings also indicate that the proposed
approaches are very efficient for classifying two categories of EEG signals. This research can help to provide clinical information about patients who have epilepsy, neurological disorders, mental or physiological problems. In BCI systems, if the MI tasks are reliably distinguished through identifying typical patterns in EEG data, motor disabled people could communicate with a
device by composing sequences of these mental states. In this dissertation, for the identification of MI tasks in BCI applications, we developed three methods:
(1) Cross-correlation based logistic regression (CC-LR).
(2) Modified CC-LR with diverse feature sets.
(3) Cross-correlation based least square support vector machine (CC-LS-SVM).
The experimental results have demonstrated the effectiveness of the methods for the identification of MI tasks. These techniques can assist clinical diagnoses and
rehabilitation tasks. Finally we investigated two issues for the MI classification:
(1) Which algorithm performed better.
(2) Which EEG data is more suitable for getting information about MI tasks.
Is it the motor area data or the all-channels data? To answer these two questions, we considered the three algorithms: the CC-LSSVM, the CC-LR and the cross-correlation based kernel logistic regression (CCKLR).
Based on the experimental results, we concluded that the CC-LS-SVM algorithm is the best algorithm for the MI tasks EEG signal classification, and the allchannels EEG data can provide better information than the motor area EEG data for
the MI tasks classification. Furthermore, the CC-LS-SVM approach can correctly identify the discriminative MI tasks, demonstrating the algorithms superiority in the
classification performance over other existing methods. |
|---|