Baseline removal from EMG recordings

A method for baseline (BL) removal in needle EMG records is presented. Different processing techniques are sequentially used. Firstly motor unit action potentials (MUAPs) are extracted from the signal by means of a wavelet transform-based procedure. Potential-free, discontinuous segments are thus obtained, whose fluctuation is assumed to be related to BL wander. These signals are then time averaged to attenuate the effect of noise and of low amplitude MUAPs originated distant from the electrode. Spline interpolation is then used to build a continuous reconstructed signal whose spectral characteristics approximate that of the real BL. The spectrum of this signal is estimated by AR modeling and an FIR filter is implemented accordingly for filtering out the BL, low frequency components from the original EMG signal. Two merit figures are devised, which measure the degree of BL fluctuation present in an EMG record. These figures are used to compare our method with the conventional approach which consider the BL to be a constant value. Experiments for BL removal from real and simulated EMG signals are carried out. The superior performance of our approach is shown regarding these merit figures and visual inspection.