Making Some Sense Out of the Near Flat ERG With the Continuous Wavelet Transform Approach

Abstract

Purpose: : Residual electroretinograms (ERG) are challenging to analyze given their low-amplitude, low signal to noise ratio, and minimal morphological features reminiscent of the normal signal. We examined if use of the continuous wavelet transform approach could help identify remnant features of the normal ERG in these nearly flat responses.

Methods: : Analysis was performed on 58 residual photopic ERGs (background of 30 cd.m^-2 and flash intensity of 0.64 log cd.sec.m^-2) obtained from patients affected with a variety of retinopathies. Amplitudes ranged from subjectively non-recordable to 20% of normal amplitude. These pathological responses were compared to normal ERGs evoked to flashes ranging in intensities from -1.62 log cd.sec.m^-2 to 2.84 log cd.sec.m^-2 in 14 steps (photopic hill). Continuous wavelet transform was performed to obtain time-frequency representations. A chi-squared test highlighted any time-frequency region distinct from noise with 99% confidence. We then considered whether residual ERGs presented significant activity within the time-frequency region comprised between 10 and 50 msec (after flash onset) in the time domain and between 20 and 40 Hz in the frequency domain, both of which always consistently included significant levels of activity in the normal ERG signal.

Results: : Except for the brightest intensity used (e.g. 2.84 log cd.sec.m^-2), our method revealed significant signal in the 10-50 msec and the 20-40 Hz region for all ERG waveforms analyzed. Also, within the same time window of 10-50 msec post-stimulus, we considered the 90-150 Hz and 200-300 Hz frequency domains, since all normal recordings also presented significant activity. Out of the 58 residual ERGs considered, 27 presented significant activity in the low frequency region, 44 in the mid frequency and 43 in the high frequency region.

Conclusions: : The time-frequency representation obtained with continuous wavelet transform provides a useful tool to identify subtle features of residual ERGs. Such methods should allow a more accurate quantification of ERG responses, especially of the low-voltage ERGs with minimal morphological cues that can be related to the normal response. Use of this analytical approach should also facilitate follow-up estimates of disease progression in patients initially presenting with a near flat ERG or identify diagnostically significant frequency domain changes in pathological ERGs, a feature not currently used in clinical electroretinography.

Keywords: electroretinography: clinical • computational modeling • low vision