Loading...

A Novel Approach for Seizure Prediction using EEG Signals

Shahbazi, Mohammad | 2018

563 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 51187 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Karbalaei Aghajan, Hamid
  7. Abstract:
  8. As the fourth most common neurological disorder, epilepsy affects lots of people all around the world, some of whom have to live with unpredictable seizures uncontrollable by surgery or medication. Hence, Developing systems for detection and prediction of the epileptic seizures will help the patients to avoid the possible damages caused by sudden seizures. This study addresses the task of epileptic seizure prediction, using three different novel approaches. The first approach, which is based on anomaly detection, contains three steps: feature extraction from EEG signals, training a one-class SVM classifier, and a post-processing step. The second method exploits a recurrent neural network to model the temporal transitions of the features extracted from the EEG signals. Finally, the last method is based on the convolutional-recurrent neural network structure, which enjoys the ability to learn and extract features from the EEG signals without any human supervision. After the evaluation of the proposed methods on the CHB-MIT dataset, it was found that the method based on the anomaly detection paradigm needs more improvement to be suitable for the task on hand. In contrast, the method based on the recurrent neural network has yielded an acceptable performance for the prediction of the seizures. Finally, the best achieved performance belongs to a 2-D convolutional-recurrent structure trained on the STFT images extracted from the EEG signals, yielding a sensitivity of 98.21% and a false prediction rate of 0.13/h and a mean prediction time of about 44 minutes. The proposed structure has outperformed the structures suggested in the previous deep learning-based studies, and the achieved results are comparable to those of the state-of-the-art studies
  9. Keywords:
  10. Machine Learning ; Deep Learning ; Recurrent Neural Networks ; Anomaly Detection ; Electroencphalogram Signal ; Epilepsy ; Convolutional Neural Network ; Seizure Prediction

 Digital Object List

 Bookmark

No TOC