| Abstract |
Electroencephalography (EEG) is one of the most frequent tools used for brain activity analysis. Its high temporal resolution allows us to study the brain at rest or during task execution with details not available in traditional imaging modalities. The low cost and simple mode of application enable EEG to be used for studying patient populations to help in the diagnosis and treatment of various brain injuries and disorders such as stroke, Alzheimer’s or Parkinson’s disease.
Unfortunately, EEG signals are frequently contaminated by undesirable noise and physiological artifacts, which may distort the underlying true neural information and lead to false diagnoses or unreliable experimental data that cannot be used for valid scientific studies. Consequently, artefact removal is a key step in EEG signal processing, and due to the complexity of the problem, it is still an active, open research area. This thesis presents novel methods developed to solve problems in EEG artifact removal. The fully automatic methods allow us to remove eye movement, blink (EOG) and heart-related (ECG) artifacts without using additional reference channels. Independent Component Analysis (ICA) was applied to the measured data, and the Independent Components (ICs) were examined for the presence of both ECG and EOG. An adaptive threshold based QRS detection algorithm was applied to the ICs to identify ECG activity using a rule-based classifier. EOG artifacts were removed from ocular artifact ICs in a selective way using wavelet decomposition minimising the loss of neural information content during the artifact removal process.
The second part of the thesis focuses on functional connectivity methods that allow the construction of resting state and task-related brain activity networks. First, resting state connectivity methods were used to analyse stroke patient brain activity in order to discover potential biomarkers for stroke recovery. Data set was recorded form healthy volunteers and stroke patients during resting state and functional connectivity graphs were created for the delta, theta, alpha and beta frequency bands. A comparison was performed between patients and control subjects as well as between start and end of the stroke rehabilitation period. The results showed differences in the graph degree, clustering coefficient, global and local efficiency that correlate with brain plasticity changes during stroke recovery, and that these can be used as biomarkers to identify stroke severity and outcome of recovery.
To uncover changes in the connectivity network during task execution, Dynamic Brain Connectivity (DBC) methods must be used. Traditional techniques to reveal temporal changes are based on the Short-Time Fourier Transform or wavelet transformation, which have limits on temporal resolution due to the time-frequency localization trade-off. In this work, a high time-frequency resolution method using Ensemble Empirical Mode Decomposition was proposed that generates phase-based dynamic connectivity networks based on the instantaneous frequency of the signals. A comparison with sliding-window techniques was conducted to validate the accuracy of the method. The results showed that the new method can track fast changes in brain connectivity at a rate equal to the sampling frequency.
|
