Research Seminars

Three new methods for MEG/EEG signal analysis and denoising

Alain de Cheveigne
" Audition" team, Department of Cognitive Sciences, Ecole Normale Superieure, Paris, France

16 November 2007

Abstract

The level of noise in recordings from MEG and EEG (magneto- and encephalography) sets a limit to what can be learned with these techniques. Noise also reduces the feasablity of applications such as brain-machine interfaces (BMI), in which brain signals are used to directly control the environment via a prosthesis. Progress in noise reduction translates into a cleaner picture of brain processes. I will present three new methods for noise reduction that target the three main sources of noise observed in electrophysiology: environmental noise from power lines and machinery, sensor noise, and physiological noise (e.g. heartbeat or ongoing brain processes). The methods are based on a variety of signal processing techniques (linear filtering, PCA, subspace projection, and ICA-related methods such as DSS) that are combined in new ways to greatly enhance the level of brain signals relative to noise.