Single-trial evoked potential estimation using wavelets
Computers in Biology and Medicine
Learning from M/EEG data with variable brain activation delays
IPMI'13 Proceedings of the 23rd international conference on Information Processing in Medical Imaging
| Hi-index | 0.00 |
A new method is introduced for estimating single-trial magneto- or electro-encephalography (M/EEG), based on a non-linear fit of time-frequency atoms. The method can be applied for transient activity (e.g. event-related potentials) as well as for oscillatory activity (e.g. gamma bursts), and for both evoked or induced activity. In order to benefit from all the structure present in the data, the method accounts for (i) spatial structure of the data via multivariate decomposition, (ii) time-frequency structure via atomic decomposition and (iii) reproducibility across trials via a constraint on parameter dispersion. Moreover, a novel iterative method is introduced for estimating the initial time-frequency atoms used in the non-linear fit. Numerical experiments show that the method is robust to low signal-to-noise conditions, and that the introduction of the constraint on parameter dispersion significantly improves the quality of the fit.