الفهرس 
AbstractOnly 14 pages are availabe for public view
 |  | from | 179 |  |  |
| | | from | 179 | | |
Abstract
Our thesis dealt with a vital and important issue in medical signal processing, that is the detection and prediction of epileptic seizures in various domains, achieving a high performance, compared to the other well-established techniques that have been regularly employed in the state-of-the-art EEG-based systems. Two approaches have been demonstrated in this thesis.
In the first approach, statistical time-frequency analysis for EEG channel selection and seizure prediction has been introduced. The purpose of this approach is to distinguish between pre-ictal and inter-ictal epochs, which enables specialists and other care-givers to anticipate the onset of the seizure as earlier as possible. Different wavelet families have been considered including Haar, db4 and d8, Sym4 and Coif4 wavelets for selecting detail and approximate coefficients. Statistical analysis was performed on different attributes including local mean, local median, local variance, derivative of signal in addition to entropy as input features to be further classified using a thresholding strategy.
This approach is of multi-channel nature, and it depends on pre-defined constrains on the required prediction and false-alarm probabilities. Decision fusion and moving average post-processing steps are utilized to reduce the false-alarm effects and to make robust decisions regarding signal activities. The proposed approach has been tested for different prediction horizons. Simulation results have revealed that db4 wavelet achieves the best prediction results as the filter lengths in db4 wavelet are longer than those in Haar wavelet, and hence more details of EEG signals are incorporated into the wavelet transform process. The proposed approach demonstrated the best result in comparison with the other previous results, revealing a high sensitivity of 100% for db4 with a low average FPR of 0.0818h-1 and a high average PT of 38.1676 min.
The second approach investigates the effect of the FWHT and several types of statistics for further processing of EEG signals to distinguish between interictal and ictal epochs in a realistic signal acquisition scenario. The main benefit of this approach, in comparison with other approaches in this field, is the avoidance of the classification problems represented in training, testing, complexity and overfitting problems. The simulation results prove that the mean curve length with FWHT demonstrates the best performance in comparison with other features, achieving an average sensitivity of 98.59%, an average specificity of 96.26 and an average accuracy of 96.83%.
The third proposal suggests another method for diagnosing epileptic seizures in a more accurate, reliable and robust way. The main idea of this method is based on utilization of EMD for obtaining different IMFs of the processed EEG signals. Statistical analysis is performed on different attributes including ZCR, TEO, AVR, and LVAR of the processed EEG signals that are used for further classification using a thresholding strategy. Simulation results have revealed that the IMF1 achieves the best detection results in comparison with other intrinsic modes. The proposed method has demonstrated the best results in comparison with the other previous works achieving 100% accuracy and zero FAR from the IMF1 with zero FAR for all cases except case V that achieves 99% accuracy with zero FAR.