Brain activity at rest occurs mainly as spontaneous oscillations. These oscillations are unique and must be analyzed in single trial with a low signal to noise ratio. In the first part, this work describes an algorithm and its implementation to increase this signal to noise ratio by a wavelet denoising method, applied here to magnetoencephalography (MEG) recordings. Actual wavelet denoising methods do not consider the spatial covariance at each wavelet decomposition level. This work proposes and evaluates a new data estimation by using the discrete wavelet representation and a noise model acquired in a MEG empty room data recording.
These denoised signals allow a localization of the spontaneous activity generators. In the second part, this work focuses on the localization of oscillators for simulated and real data sets. Results show that the localization of sources is more accurate for the data denoised with this new method. Furthermore, this work refines the definition of the spatial covariance matrix of the Maximum Entropy on the Mean (MEM) optimization problem, resolved in the inverse problem for the localization of sources of brain activity.
| Date | 10 Nov 2014 |
|---|
| Original language | French |
|---|
| Awarding Institution | - École de technologie supérieure
|
|---|
| Supervisor | Jean-Marc Lina (Supervisor) |
|---|
Lacourse, K. (Author),
Lina (Supervisor),
10 Nov 2014Student thesis: Master's thesis › Master in Engineering: Engineering