Loading...

EEG Denoising Using Combination of Kalman Filtetring and Blind Source Separation Approaches for Epileptic Components Extraction

Mohammadi, Marzieh | 2014

573 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 46198 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Shamsollahi, Mohammad Bagher
  7. Abstract:
  8. Epilepsy is a neurological disorder whose prevalence is estimated to be 1% of the world population. Electroencephalogram (EEG) is one of the best and convenient non-invasive tools used in diagnosis and analysis of this disease. Epileptic components extracted from EEG recordings are widely used in neuroscience in the diagnosis analysis like epilepsy source localization. However, epileptic components are often contaminated and covered with artifacts of physiological origin (baseline, EMG, ECG, EOG, etc.) or instrument noises (power supply, electrode, etc.). So, preprocessing and denoising is necessary for precise analysis of epilepsy EEG recording. Heretofore, several methods have been proposed for denoising that each of them has some advantages and disadvantages and the selection of the best one depends on the application.
    Epileptic components in EEG signals include ictal and interictal activities. One of the conventional denoising methods for inter-ictal EEG signals is blind source separation (BSS). This technique suffers from some limitation and weaknesses causing imperfect denoising. In this research we aim at suppressing noises particularly muscle artifacts from interictal EEG data by combining BSS methods and Kalman filter in order to improve denoising results based on error and source localization criteria. Two combined methods have been proposed and applied on two groups of data: artificial spikes that are contaminated by real noises and real epileptic EEG signal. The results of the first proposed approach, based on serial combination via time-varying AR (TVAR) model, show while keeping source localization results, the errors for input SNR less than -5dB is enhanced. The second proposed algorithm is utilized state space model to combine BSS and Kalman filter. In this way, error and source localization have considerable improvement. Furthermore, the outcomes of two proposed techniques on real data demonstrate the accuracy of them for denoisng and source localization
  9. Keywords:
  10. Epilepsy ; Electroencphalogram ; Noise Removing ; Kalman Filters ; Blind Sources Separation (BSS)

 Digital Object List

 Bookmark