الفهرس | Only 14 pages are availabe for public view |
Abstract This thesis is concerned with a very vital branch of biomedical signal processing. This branch is the processing of Electroencephalography (EEG) signals for either seizure detection or prediction. Seizure detection is an off-line task that can be performed on EEG recordings of multi-channel nature. On the other hand, seizure prediction is an online task that need to be performed prior to seizure occurrence with as long time as possible. The concept of sub-band decomposition is exploited in this thesis for both seizure detection and prediction to check the most appropriate sub-band for each task. The sub-band decomposition is performed with different types of digital Infinite Impulse Response (IIR) filters including Butterworth , Chebyshev type I , and Chebyshev type II filters. A comparison between these filters in terms of the accuracy of classification is presented. The seizure detection is performed with scale-space analysis based on Scale Invariant Feature Transform (SIFT) giving acceptable seizure detection results. On the other hand, a statistical approach based on Probability Density Functions (PDFs) for signal attributes is developed on the signal sub-bands. This statistical framework depends on the analysis of five signal attributes: amplitude, local mean, local variance, derivative, and local median. The strategy of classification depends on taking several decisions for every multi-channel signal segment based on the PDFs of different signal attributes. A majority voting methodology is adopted for decision fusion. For long records of data segmented into one-second segments, a moving average filter is used to smooth the signal representing the seizure prediction outcome. The proposed framework for seizure prediction gives high seizure prediction rates from gamma band with a long enough prediction horizon. |