Denoising EMG and EEG for monitoring small animal models during NMR experiments

The present growing field of molecular imaging, including multimodality microimaging techniques and spectroscopic approaches, is mainly based on small animal studies. Monitoring such models requires an efficient treatment and use of electrophysiological signals which may be spoiled by environmental effects especially when working with nuclear magnetic resonance (NMR) since radiofrequency (RF) pulses and magnetic field gradient commutations may create spurious supplementary signals. In this work, a method is given for EEG and EMG denoising of signals acquired during phosphorous magnetic resonance (MR) brain spectroscopy data acquisition on a rat model developed for sleep/awake studies. The proposed approach is based on wavelet decomposition and the key method is to turn into profit the shape variations of EMG during the time course of sleep/awake cycles. Statistical properties of the noise are studied using EMG recorded during paradoxical sleep as noise model. A specific estimation of noise level using EMG recorded during slow sleep leads to an optimal wavelet coefficients thresholding. This approach is well suited to improve signal to noise ratio of EEG and EMG and to preserve small amplitude electro physiological signals.