Sparse Bayesian Learning for EEG Source Recovery

Student thesis: Master Thesis and HD Thesis

  • Trine Nyholm Kragh
  • Laura Nyrup Mogensen
4. semester, Mathematical Engineering, Master (Master Programme)
This thesis treats the problem of recovering original brain source signals from low-density EEG scalp measurements.
Based on state of the art methods, an algorithm is proposed to reproduce the current results.
The algorithm leverage a covariance-domain dictionary learning (Cov-DL) method and a multiple sparse Bayesian learning (M-SBL) method. The proposed application of Cov-DL did not succeed.
Thus, an alternative solution was proposed.
The final algorithm was tested on EEG and compared to solutions obtained by independent component analysis of high-density EEG.
A frequency analysis was performed comparing raw EEG and the recovered sources.
With respect to practical use, an estimation of the unknown number of active source signals was proposed.
It is concluded that the proposed algorithm is unable to reproduce the state of the art results.
However, the M-SBL method alone is successful and a potential is seen for an estimation of the number of active sources, from M-SBL.
Publication date3 Jun 2020
Number of pages119
ID: 333525204