Loading...
| Friend's email | |
| Your name | |
| Your email | |
| enter code | |
This page was sent successfuly
- Type of Document: M.Sc. Thesis
- Language: Farsi
- Document No: 46572 (05)
- University: Sharif University of Technology
- Department: Electrical Engineering
- Advisor(s): Shamsollahi, Mohammad Bagher
- Abstract:
- EEG Signal Processing has really important role in study of nerves system and nervous diseases. In most applications, signals occur in the frequency band ranging below 100 Hz are considered. In this study, the higher frequency components of the signal, especially in patients with epilepsy, reported which they are called high frequency oscillations (HFOs). Some sources considered that HFOs are up to 1000 Hz and some other considered up to 2500 Hz. In study of HFOs, the high frequencies between 100 and 200 Hz that are called ripple and seen in normal mode are separated from the faster frequencies from 200 to 500 Hz and these components can accurately help to detect epileptic foci generator (to remove the focus with surgery in patients with intractable epilepsy). The goal of this project is to investigate the detection methods of fast ripples and separate them from epileptic spikes and background activities. Various methods were used to classify and detect these fluctuations. Some of these methods have good results, such as Hidden Markov Model and Switching Kalman Filter. In application of Hidden Markov Model for a three-class classification problem, we assume that have three Hidden Markov Model for our three dynamics, Ripple, spike and background. These models are trained using the EM algorithm. Once the training phase is completed, in the test phase, the trained Hidden Markov Models are performed to classify and applied to a set of observations and the model with the highest log-likelihood is selected as its label. The average of sensitivity in fast ripples and non-fast ripples is achieved 0.99. In using Switching Kalman Filter for a three-class classification problem, we used three linear models and switched between them. In this method we have a discrete switching variable that specifies which matrices of model should be used at any time t, and then the maximum conditional probability of switch variable given observation is the label of this sample. The average of sensitivity for Switching Kalman Filter method in the case of classification of fast ripple and non fast ripple and in low fast ripple to background ratio is 0.98. In high fast ripple to background ratio Hidden Markov Model has a better result than Switching Kalman Filter
- Keywords:
- Epilepsy ; Electroencphalogram ; Detection ; Hidden Markov Model ; High Frequency Oscillations (HFO) ; Fast Ripple ; Switching Kalman Filter
Digital Object List-
محتواي کتاب - view
Bookmark