Detecting MMN in Infants EEG with Singular Value Decomposition

In English Mismatch negativity (MMN) is an EEG voltage fluctuation caused by the brain's automatic reaction to unexpected changes in a repetitive stimulation. In an experiment we studied 68 infants of which 2/3 were born preterm. Due to noise of large amplitude, the MMN is difficult to detect in a single infant's EEG. Therefore grand average, which is a average of many subjects EEG recordings, is sometimes used. In this paper singular value decomposition (SVD) is proposed as an alternative to grand average. Consider the SVD USigmaVT = M, where the rows of M contains noisy EEG epochs. Usually data is projected onto the leftmost column of V since this column represent the largest common component of the rows of M. When data is affected by noise of a very large amplitude we may need to choose another column of V. In this paper we propose to choose the leftmost column of V such that the elements of the corresponding column of U has approximately equal values