Adaptive smoothing of evoked potentials

Abstract Evoked potentials (EPs) are very low amplitude signals generated in brain or spinal nervous structures after a brief stimulation of peripheral sensory organs. Their extraction from the ongoing electroencephalogram requires sophisticated signal processing methods, because the signal-to-noise ratio is very low. The basic time-locked average of successive responses to a train of stimulations obviously improves the SNR but is unable to track the slow variations of significant waves. Therefore, adaptive techniques are used to reduce the number of stimulations and this is what the present article will attempt to describe. One can consider the successive elementary responses as the rows of a bidimensional signal, the columns of which being made up of the points having the same time delay from the stimulus. The proposed signal processing consists of a ‘vertical’ exponential filtering (equivalent to a sliding-window averaging), associated with a ‘horizontal’ forward-backward non-causal smoothing based on a new statistical model of the filtered signal, leading to a very efficient global algorithm. The proposed 2D processing allows a real-time (row after row) extraction of smoothed responses, providing an unbiased and adaptive location of the EPs waves.