الفهرس 
Chapter 1 IntroductionOnly 14 pages are availabe for public view
 |  | from | 132 |  |  |
| | | from | 132 | | |
Abstract
Epilepsy is a central nervous system (neurological) disorder in which
brain activity becomes abnormal, causing seizures or periods of unusual
behavior, sensations, and sometimes loss of awareness. Epilepsy affects both
males and females of all ages. Therefore, it is essential to notify the patient’s
medication-resistant epileptic seizure to the caretaker and analyze the pattern
of related signals before, during, and after the seizure onset.
In the medical diagnosis of epileptic seizures, classification is a significant
step that directly affects the results. However, the visual examination of the
Electroencephalogram (EEG) is a comparatively common analytic procedure
of epilepsy, although it is costly, time-consuming, and relies on the
experiences of the doctor. In this thesis, we provide and propose an efficient
and accurate epileptic seizure automatic diagnosis system suitable for clinical
epileptic seizure diagnosis as an urgent task in recent times.
Firstly, we propose and evaluate the performance of KNN classifier in
classifying the epilepsy of epileptic patients from EEG signals. The
preprocessing is performed to overcome the problem of imbalanced data using
different sampling techniques. Finally, an optimization is performed by
applying KNN classifier to obtain the optimal value for k on epileptic seizure
recognition dataset. The result demonstrated that the optimized KNN with
cuckoosearch and SMOTE resampling techniques give 98.5%, 99%, 99% for
accuracy, precision, and recall respectively.
Secondly, we propose a system that can select a channel by calculating
the variance parameter for each channel. The highest three channels of
variance will be selected. Then extracted the features that fed to the machine
learning for classification in an efficient manner. In this model, ensemble
technique achievied a sensitivity of 100%.
Thirdly, an efficient framework is proposed by converting EEG signals into
spectrogram images. Anew dataset was created by extracting the most
important features of the EEG signals that detect spikes and after that
converting it to spectrogram images. These images are fed to the pre-trained
model to be classified as normal or seizure. The proposed system is evaluated
through different experiment circumstances over the CHB-MIT dataset. For
Tegear method, the highest accuracy of 93.06% was achieved using adam
Abstract
vi
solver using squeezenet model. Also, this hybrid system provides a new
approach for Early Epilepsy in Medical Internet of Things applications.
Further, the results indicate the proposed system is suitable for IoT-based real-
time seizure recognition from EEG recording along with providing an
automated biomarker for normal and epileptic EEG signals in smart healthcare
systems in the context of the smart city.
Recently, data privacy is a major concern when accessing and processing
sensitive medical data. Therefore, fourthly, an enhancement of the last
proposed system by adding security using a combination of CNN and
encrypted spectrogram images. An encrypted Spectrogram and CNN-based
method are planned for the detection of Seizures from EEG signals. The
accuracy of the suggested structure has been estimated up to 86.11 % and
84.72 % using googlenet with Arnold and chaotic methods respectively. The
suggested system’s accuracy demonstrates that this system can be one effective
system for seizure detection.