Statistical approaches to nonstationary EEGs for the detection of slow vertex responses

A slow vertex response (SVR) is an electric auditory evoked response used for an objective hearing power test. One of the aims of an objective hearing power test is to find infants whose hearing is less than that of normal infants. Early medical treatment is important for infants with a loss of hearing so that they do not have retarded growth. To measure SVRs, we generally use the averaged summation method of an electroencephalogram (EEG), because the signal-to-noise ratio (SVR to EEG and etc.) is very poor. To increase the reliability and stability of measured SVRs, and at the same time, to make the burden of testing light, it is necessary to device an effective measurement method of SVR. Two factors must be considered: (1) SVR waveforms change following the changes of EEGs caused by sleeping and (2) EEGs are considered as nonstationary data in prolonged measurement. In this paper, five statistical methods are used on two different models; a stationary model and a nonstationary model. Through the comparison of waves obtained by each method, we will clarify the statistical characteristics of the original data (EEGs including SVRs), and consider the conditions that effect the measurement method of an SVR.