The electroretinogram (ERG) identifies the electrical signal that is generated by the retina in response to a light stimulus. It is the first biopotential ever recorded from a human subject, namely, by Dewar in 1877 [
It is clear from the above that the ERG waveform results from the amalgamation of several frequency components. Is it possible to monitor the frequency composition of the ERG signal without altering the signal as it is done with the bandwidth restriction approach? Would the use of such an approach significantly improve analysis of the ERG beyond what is accomplished when using time and amplitude measures of the ERG only? Although advanced analytical approaches are now frequently used when studying biopotentials, such as the electroencephalogram [
Normal photopic ERGs were obtained from 40 healthy subjects (26 females and 14 males, average age
According to a previously published method of ours, the ERGs were recorded with both eyes dilated (tropicamide 1%) using an active electrode (DTL fiber electrode) placed in the inferior conjunctival bag, with reference and ground electrodes pasted at the external canthi and forehead, respectively [
The amplitude of the a-wave was measured from the prestimulus baseline to the most negative trough of the ERG, while the amplitude of the b-wave was measured from the trough of the a-wave to the most positive peak of the ERG that followed the a-wave [
Frequency domain analysis (or Fourier analysis (FA)) of the ERG was carried out using the fast Fourier transform (FFT) algorithm implemented in MATLAB R2013b (Mathworks, Natick, MA, USA) as follows:
In order to localize the energy content of the ERG in both time and frequency we computed, using MATLAB, the continuous wavelet transform (CWT) of selected ERGs as follows:
Use of the CWT approach allowed us to analyse the ERG, continuously, at every possible scale
Mean value, standard deviation (SD), and coefficient of variation (CV) were computed for all ERG parameters that were identified using the different analytical approach.
As reported in Table
Normative data (mean ± standard deviation (SD) and coefficient of variation (CV, in bold)) obtained for each parameter (time, frequency, amplitude, power, and energy) assessed using the different analytical approaches compared in this study (time domain, frequency domain, and continuous and discrete time-frequency domain). The time domain allows timing and amplitude quantification of two major components (i.e., the a- and b-waves). The frequency domain identifies the frequency and power of three major components (probably associated with the a- and b-waves and OPs). The continuous time-frequency domain allows timing, frequency, and energy measurements of three main components (probably associated with the a- and b-waves and OPs). Finally, with the discrete time-frequency domain, the components are identified in predetermined temporal windows (i.e., intervals) and frequency bands (i.e., instead of precise timing and frequency) and allow more components to be identified and the a- and b-wave can be quantified independently (i.e., in contrast to the frequency domain or continuous time-frequency domain in which the a- and b-waves formed a single low-frequency component).
Time domain (peak time and amplitude) | ||||
---|---|---|---|---|
Time (ms) | Frequency (Hz) | Amplitude ( |
Origin | |
Main component 1 | 13.53 ± 1.55 |
n/a | 32.21 ± 5.11 |
a-wave |
Main component 2 | 30.98 ± 1.33 |
n/a | 104.81 ± 18.66 |
b-wave |
|
||||
Frequency domain (Fourier analysis) | ||||
Time (ms) | Frequency (Hz) | Power ( |
Probable origin | |
|
||||
Main component 1 | n/a | 28.81 ± 7.17 |
10.46 ± 2.15 |
a- and b-waves |
Main component 2 | n/a | 75.38 ± 7.69 |
3.43 ± 0.89 |
OPs |
Main component 3 | n/a | 146.00 ± 13.31 |
1.88 ± 0.43 |
OPs |
|
||||
Time-frequency domain (continuous wavelet transform) | ||||
Time (ms) | Frequency (Hz) | Energy ( |
Probable origin | |
|
||||
Main component 1 | 31.04 ± 0.89 |
29.70 ± 5.71 |
202.8 ± 21.08 |
a- and b-waves |
Main component 2 | 32.04 ± 2.22 |
73.81 ± 6.28 |
111.57 ± 23.17 |
OPs |
Main component 3 | 29.81 ± 1.96 |
150.28 ± 11.57 |
69.29 ± 16.79 |
OPs |
|
||||
Time-frequency domain (discrete wavelet transform) | ||||
Time interval (ms) | Frequency band (Hz) | Energy ( |
Probable origin | |
|
||||
Main component 1 | −20 to 17.5 | 20 ± 6.6 | 78.08 ± 10.66 |
a-wave |
Main component 2 | 17.5 to 55 | 20 ± 6.6 | 214.46 ± 19.02 |
b-wave |
Main component 3 | 0 to 17.5 | 40 ± 13.3 | 100.79 ± 9.39 |
a-wave |
Main component 4 | 17.5 to 55 | 40 ± 13.3 | 228.41 ± 28.05 |
b-wave |
Main component 5 | 8.125 to 55 | 80 ± 26.6 | 145.83 ± 21.75 |
OPs |
Main component 6 | 17.5 to 40.35 | 160 ± 53.3 | 80.81 ± 12.48 |
OPs |
Fourier analysis (FA) of 4 normal ERGs. The frequency spectrums are shown as normalized power spectrum, in percentage, where the spectrums are normalized to their maximal value. The associated ERGs are shown at the bottom of each spectrum. The a-wave, b-wave, and OPs are indicated as “a,” “b,” and “OPs,” respectively. (a) FA of a composite ERG, averaged from 40 subjects, showing the 3 typical frequency components that contribute to the ERG (~30 Hz: a- and b-waves contribution, black arrow; ~75 Hz and ~150 Hz: oscillatory potentials (OPs) contribution, gray arrows). (b) FA of an ERG showing enhanced OPs (increased ~75 Hz and ~150 Hz component contribution; thick gray arrows). (c) FA of a typical contaminated ERG showing the 3 standard ERG components (see arrows, same color-coding as previous panels) and a sharp, noise-related, maximal component at 60 Hz (60-cycle line interference contribution, red arrow). This sharp noise component seems to disturb the identification of the OPs component located at ~75 Hz. (d) FA of another contaminated ERG showing the 3 characteristic frequency components (see arrows, same color-coding as previous panels) of the ERG and 4 interference-related components at 60, 120, 180, and 240 Hz, respectively (60-cycle harmonics contribution, red arrows). These noise components seem to complicate the identification of the two typical OPs components located at ~75 Hz and ~150 Hz.
The frequency components contributing to the genesis of the ERG can be identified using the FA, such as that obtained with the FFT. This is best exemplified in Figure
As shown in Figure
Continuous wavelet transform (CWT) analysis of the ERGs that were shown at Figure
In each scalogram of Figure
As illustrated in Figure
Analysis method and classification improvement of normal ERGs using the discrete wavelet transform (DWT). (a) We computed the DWT of 40 normal ERGs with various signal-to-noise ratios (
As seen with the DWT scalograms of Figure
As indicated above, the a- and b-wave components were seen on both the 20 Hz (20a and 20b descriptors) and 40 Hz bands (40a and 40b descriptors). These descriptors can be used to segregate ERGs of different morphologies. This is better illustrated in Figure
In Figure
(a) ERG traces (averaged from up to 100 responses) obtained at seven time points in the right (OD) eye and left (OS) eye of a male patient affected with retinitis pigmentosa (both eyes presented with nonrecordable scotopic ERGs, constricted visual fields, pigmentary deposits, and decreased visual acuity) in a time span of 3 decades. The horizontal (time) and vertical (voltage) scale bars apply to both eyes and some traces have been magnified (×2, ×5, or ×10 times) for visualization purposes. The flash onset is indicated by the black vertical arrow. ERG progression is shown in years since the first visit on the left-hand side. (b) Scalograms computed for each pathological ERG waveform (presented in the same order than in panel (a)) in which we quantified the 20b and 40b descriptors. Note that, in some scalograms, the position of the 40b descriptors was delayed (i.e., delayed latency of the b-wave) compared to normals (see Figure
20 consecutive single-flash ERG responses (gray traces) obtained from a patient affected with retinitis pigmentosa. The average of these 20 raw ERG responses canceled the uncorrelated noise to yield the blue tracing (overlaid on top of each single-flash response). DWT denoising of the individual noisy single-flash responses reveals denoised biological responses, which are shown as red traces. All red traces are nearly identical (shape and amplitude) to the trace obtained from the average (i.e., blue traces) of the 20 noisy responses, thus validating this denoising approach. The horizontal (time) and vertical (voltage) scale bars apply to each trace. The flash onset is indicated by the black vertical arrow.
To date, analysis of the ERG relies mostly on time domain (TD) measurements (peak time and amplitude) of its two major components, namely, the a- and b-waves. However, as shown with the examples illustrated in Figures
FA methods, such as the one presented in Figure
Use of the CWT (Figure
Interestingly, it seems, from FA and CWT analyses, that the components of the ERG cannot be associated with single frequency values but rather to a range of values. For example, in the FA power spectrums of Figure
In the DWT scalograms the a- and b-waves were characterized by distinct components located in the 20 Hz (20a and 20b descriptors) and 40 Hz (40a and 40b descriptors) bands. It was difficult to accurately determine these distinct frequency components using the CWT, although in the FA some ERGs did show both the 20 and 40 Hz components (i.e., identified as double-peaks in Figures
Finally, at the end-stage of severe degenerative retinopathies (such as RP), nearly extinguished ERGs (e.g., low-SNR, such as the one shown in Figures
In this study we limited the TD approach to its most widespread descriptors (i.e., amplitude and peak time of the a- and b-waves), but other unusual descriptors (e.g., area-under-the-curve of the a- or b-wave, time to reach a certain percentage of the a- or b-wave amplitudes, steepness of the rising or descending flank of the b-wave, filtered OPs measurements, etc. [
In this paper, we have presented a brief overview of the different analytical approaches that can be used to quantify the ERG waveform. As long as the response remains measurable, the traditional measurements of the a- and b-waves can be used to monitor the peak time and the amplitude of the ERG signal. However, these measurements only look at the ERG signal as a whole, instead of looking at the different frequency components (possibly of distinct cellular origin) separately. The discrete wavelet transform offers the possibility to extract more components of the ERG signal, even in very poor SNR responses. Standardized time domain analysis of retinal function should thus be complemented with advanced DWT descriptors of the ERG. The latter should allow more sensitive/specific quantifications of ERG responses, facilitate follow-up of disease progression, and identify diagnostically significant changes of ERG waveforms that are not resolved when the analysis is only limited to time domain measurements, thus bringing the analysis and interpretation of the ERG signal in the 21st century, as it is already the case with other biopotentials such as the electroencephalogram and electrocardiogram.
The authors of the paper (Mathieu Gauvin, Jean-Marc Lina, and Pierre Lachapelle) do not have any financial disclosure or conflict of interests to report.
This study was funded by Grants-in-aid from the Canadian Institutes for Health Research (MOP-126082) and the Vision Health Research Network of the Fonds de Recherche du Québec-Santé.