A simple wavelet-based test for evoked responses

Abstract

Statistically valid detection of evoked responses from magnetoencephalographic (MEG) sensors is complicated by temporal autocorrelation. By decorrelating time series and transforming them toward normality, the discrete wavelet transform (DWT) allows the analyst to test for an association between stimulus and sensor time series with appropriate degrees of freedom. Eswaran et al. (Neurosci. Lett. 2002a;331:128–32) used a 151-channel fetal MEG system to obtain serial recordings from 10 pregnant subjects. There were 3–8 recordings per subject. In each recording session, the fetus was stimulated by 500 Hz and 1 KHz tones with a relative frequency of 80–20%, respectively. In this new analysis of the same data, the fetal MEG signals were compared to two different stimulus waveforms: the frequent tone and the Novel stimulus, defined as a change in pitch. WaveDetect was developed to determine whether there was a significant association between the stimuli and the MEG traces. This test is performed by taking the DWT of each series and then computing the Spearman correlation between the wavelet coefficients for an appropriate scale. A significant response (i.e., correlated stimulus–sensor pair) was detected from each patient. This result suggests that the combination of serial recordings and WaveDetect may ensure reliable detection of auditory evoked responses.