Extraction and modelling of oscillatory potentials

This paper considers the recommendation that oscillatory potentials (OP) be extracted by filtering in the frequency domain. This recommendation presumes that filtering isolates OPs from other ERG waveforms. However, we show that the leading edge of the a-wave has substantial frequency overlap with the OP spectrum at high intensities and that it contaminates these wavelets in the frequency domain. We propose a method of signal conditioning that removes a-waves prior to filtering. When this is done, the OPs show a bimodal distribution in the frequency domain that is well approximated by two Gaussians having means (±std. dev.) of 91.0±14.6 Hz and 153.1±17.1 Hz. This implies that two functions can be used to model the OPs in the time domain. However, we show that as most of the power of the Fourier spectrum (74%) is contained in a single Gaussian, a reasonable OP model can be derived by using a single function in the time domain. We test such a model on humans (n=5) and pigmented (n=14) and albino (n=14) guinea-pigs and show that it provides excellent fits to data across a range of flash exposures. Furthermore, changes in OP amplitude and timing between strains of guinea-pigs are easily detected with this model. We show that there is no statistical justification for making the model more complex by including multiple functions. Such paramatisation of the OP envelope provides a valuable and intuitive description of the OP waveforms in the time domain. The model provides an excellent description of OPs obtained with the current paradigm, however the single gaussian model may be deficient under stimulus conditions which produce highly asymmetric OP envelopes.