Loading...

Subspace Identification and Brain Connectivity Estimation of Electroencephalogram Signals Using Graph Signal Processing

Einizadeh, Aref | 2023

71 Viewed
  1. Type of Document: Ph.D. Dissertation
  2. Language: Farsi
  3. Document No: 56109 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Hajipour Sardouie, Sepideh; Shamsollahi, Mohammad Bagher
  7. Abstract:
  8. EEG brain signals have gained particular attention among researchers in the field of brain signal processing due to their easy and cheap recording, high temporal resolution, and non-invasiveness. On the other hand, defects such as high vulnerability to various types of noise and artifacts have caused the main challenge before processing them to improve the signal-to-noise ratio and the interpretability of brain connectivity obtained from them. In order to solve these challenges, two important problems of "separation of desired and undesired signal subspace" and "functional and effective connectivity analysis" have been raised, respectively. In solving both problems, EEG signals are usually processed and analyzed in matrix form (with channel and time dimensions). In these processes, the graph nature of the signals (in the domain of time or space) is not considered. This causes the loss of important information in the EEG signals recorded in this multi-channel structure. Graph concepts can be an up-to-date and efficient tool to solve this gap. Graph signal processing (GSP) is a nascent and new field; less than ten years have passed since the introduction and development of its concepts. In this domain, signals can be located on an irregular graph, while in classical methods, signals are only considered on regular time samples. However, one of the most important challenges of using graph concepts is the fact that the input structural graph is not known in many real applications, and this thesis tries to solve this challenge by proposing six algorithms based on GSP as follows: The proposed GraphJADE-GL and U-GraphJADE-GL algorithms deal with separating independent graph signals generated from auto-regressive (AR) stochastic processes and simultaneously learning related graphs. Also, their performance on brain epileptic signals and separation of speech signals is investigated, and superiority over classical methods is shown. The proposed MI-BSS-GS algorithm implements the separation of another important class of graph signals, called smooth graph signals, assuming the availability of source-related graphs and with better performance than existing graph and classical methods. The proposed MI-BSS-GL algorithm extends the proposed MI-BSS-GS algorithm to the case of unknown graphs and simultaneously learns the underlying graphs.The proposed algorithm of ProductGraphSleepNet, which is an attentive graph neural network (GNN), performs classification and learning of product graphs from sleep-related brain signals, which in addition to the competitive results of classification with current popular methods, the analysis of the learned temporal and spatial graphs is fairly supported by the neuroscience findings. The proposed CGP-LiNGAM algorithm is also presented for learning directed acyclic graphs (DAGs) to reveal the causal relationships between graph temporal signals with much fewer free parameters than the current methods
  9. Keywords:
  10. Graph Learning ; Graph Signal Processing ; Graph Neural Network ; Brain Connectivity ; Electroencphalogram Signal ; Signal Subspace Identification ; Brain Waves

 Digital Object List

 Bookmark

...see more