Loading...

Detection of High Frequency Oscillations from Brain Electrical Signals Using Time Series and Trajectory Analysis

Gharabaghi, Ali | 2024

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 57676 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Hajipour Sardouie, Sepideh
  7. Abstract:
  8. The analysis of cerebral signals, encompassing both invasive and non-invasive electroencephalogram recordings, is extensively utilized in the exploration of neural systems and the examination of neurological disorders. Empirical research has indicated that under certain conditions, such as epileptic episodes, cerebral signals exhibit frequency components exceeding 80 Hz, which are designated as high frequency oscillations. Consequently, high frequency oscillations are recognized as a promising biomarker for epilepsy and the delineation of epileptic foci. The objective of this dissertation is to evaluate the existing methodologies for the detection of high frequency oscillations and to propose a novel and innovative approach for the diagnosis or therapeutic intervention for individuals with epilepsy. Given that a portion of the information pertaining to high frequency oscillations resides within their morphological characteristics, as well as their temporal and frequency variations, and considering that the manual or visual identification of these occurrences is exceedingly labor-intensive and financially burdensome; the application of fully automated techniques, such as time series analysis, can prove to be highly effective for the detection of these patterns. In this study, we introduce a new technique for detecting high frequency oscillations that employs time series analysis tools and the dynamic time warping algorithm as a metric to evaluate the degree of similarity between time series. For this endeavor, we utilized the electrocorticogram data provided by the Fedele group to implement our proposed methodology. Our suggested approach, which can be executed in both the time and frequency domains, consists of four principal components: preprocessing, event clustering, feature extraction, and classification. The features employed in this methodology are derived from the types of distances between time series, which are computed in the time domain using the dynamic time warping algorithm and in the frequency domain utilizing the Euclidean distance. The features extracted from both the time and frequency domains can be utilized independently, or alternatively, they can be integrated into a feature matrix. The resultant features are subsequently fed into various classification models to detect high frequency oscillations. Ultimately, following the implementation of the proposed models, the approach that amalgamates features from both time and frequency domains, in conjunction with the logistic regression classifier, was identified as the preeminent model. This model achieved an area under the receiver operating characteristic curve of 97.11% and an area under the precision-recall curve of 58.2%
  9. Keywords:
  10. Epilepsy ; Electrocorticography (ECOG) ; High Frequency Oscillations (HFO) ; Dynamic Time Warping ; Time Series Analysis ; Electroencephalography ; Signal Processing ; Brain Waves

 Digital Object List

 Bookmark

No TOC