A new method of cone electroretinography:
the rapid random flash response
Stephen J. Fricker and James J. Sanders
A new procedure is described for cone electroretinography using a cross-correlation method
of signal processing to give the response to flash stimuli which occur at randomly timed intervals. The output, waveform is different than the usual repetitive flicker response, and can be
presented on any desired time base. An unfamiliar aspect of this process is that most, of the
random time intervals between consecutive stimuli arc shorter in duration than the output
waveform. For the parameters described, the output waveform gives the response to approximately 1,640 stimuli in 65.5 sec, with a signal-to-noise ratio which is higher than that obtained when conventional averaging techniques arc used over similar time periods. This allows
more precise and statistically significant estimates to be made of the time and amplitude
parameters of the cone response. Normal values arc given for implicit (delay) time and
amplitude, ami examples arc provided for comparison of the random flash response and conventional average elcctrorctinograms (ERC's) in normal subjects and patients with retinal
degeneration.
Key words: electroretinography, cone response, cross-correlation, random flash,
signal averaging, synchronous detection, signal-to-noise ratio,
implicit (delay) time, amplitude, retinal degeneration.
M
made of some of their different response
characteristics. The background illumination can be reduced and blue stimuli used
so as to enhance rod function, or normal
photopic illumination can be used with red
or white stimuli so as to suppress the rod
function and enhance cone function. The
repetition frequency of the stimulus also is
significant, as rods function only at relatively low stimulation frequencies, up to
approximately 17 to IS Hz., whereas cones
continue to function at considerably higher
stimulation frequencies, up to SO to 90 Hz.
This difference in response has been used
to characterize cone function by the "flicker
response," for example, at a constant stimulation frequency of 30 Hz.1 '
An inherent problem with ERG record-
any articles have been written on
the application of specific electroretinogram
(ERG) recording techniques to research
and clinical problems.1' - In order to separate rod and cone function, use has been
From the Howe Laboratory of Ophthalmology,
Harvard University Medical School, Boston,
Mass.
This study was supported by the Research Grant
No. EY008S5 from the Eye Institute, National
Institutes of Health, United States Public Health
Service.
Submitted for publication June 13, 1974.
Reprint requests: Dr. Stephen J. Fricker, Howe
Laboratory of Ophthalmology, Harvard University Medical School, Massachusetts Eye and
Ear Infirmary, 243 Charles St., Boston, Mass.
02114.
131
Downloaded From: http://iovs.arvojournals.org/ on 07/28/2017
132
hivcstigtitioc
Fricker and Sanders
Flash Trigger
Unit
_n
(r
Ophthalmology
February 1975
Amplifier
">>
Flash
S
'
Oscillator
C,(fr)
-+
PRBS
,nn n
Generator
Oscillator
CORRELATOR
C
Variable
Low-Pass
2 (fp)
Filter
Fig. 1. Block diagram of system.
J
LJLTLJ
LTLTL
Fig. 2. Top: basic random-biriary-secjuence output.
Center: flash trigger pulses. Bottom: reference
waveform.
ings is that they always contain some
degree of "noise" which affects the measurements to varying degrees, depending upon
the signal-to-noise ratio. The use of intense
flash stimuli combined with the Ganzfeld
technique produces relatively large signals
from normal subjects, but it may be questioned whether such stimuli are entirely
appropriate for testing a system which normally operates at lower light levels. Also,
patients with abnormal retinal function
usually give reduced amplitude ERG's, and
the signal definition problem becomes correspondingly more difficult. Various techniques now are available to increase the
signal-to-noise ratio of noisy recordings,
and if measurements of amplitude and time
delay are to be made with any degree of
statistical reliability then simplified applications of signal detection theory show
that some noise reduction process usually
is necessary.'1 If this theoretical treatment
is not convincing, it is only necessary to
scan the ERG literature to see that it is
Downloaded From: http://iovs.arvojournals.org/ on 07/28/2017
replete with examples of waveforms from
which implicit times and amplitudes have
been scaled, but which quite evidently
show such gross contamination by noise
that the readings obtained must be of very
doubtful statistical value.
The use of averaging computers permits
a significant improvement to be made in
the signal-to-noise ratio, but it should be
emphasized that the average response thus
obtained still is a measure of the response
of the retina to stimuli which occur at relatively low stimulation rates, for example,
one per second. A different and very effective method of increasing the signal-to-noise
ratio using rapid (fixed) repetition-rate
stimuli is that of synchronous detection/ 1 s
with the output consisting of numerical
amplitude and relative phase information at
each stimulation frequency. These two
methods of noise reduction are in fact two
aspects of the same basic process of crosscorrelation." In order to obtain an optimum
measure of cone function, it would be
desirable to use stimuli at rapid repetition
rates with a system which provides a more
conventional type of ERG response waveform with a high signal-to-noise ratio. A
means for accomplishing this is provided
by a cross-correlation technique using white
flash stimuli under photopic conditions,
with the stimuli occurring at (rapid) randomly timed intervals. The noisy signal obtained from the subject contains responses
occurring at varying intervals, and the output is obtained by cross-correlating this
Volume 14
Number 2
Cone electroretinography
133
30-
25-
Random Binary Sequence
Basic Interval 10 Milliseconds
o —
0)
<T3
.20-
a >
o
^
(0
•;
.O
O
<=
""
.15-
.10-
05-
20
(50)
30
(33.33)
40
(25)
50
(2 0)
60
70
80
(I6 67) (I4 29) (I2 5)
90
(nil)
I00
Milliseconds
(10)
Interflash Interval, Milliseconds
(Equivalent Flash Rale Pet Second)
Fig. 3. Probability distribution of interHash intervals for fr = 100 Hz.
noisy signal with a reference waveform
which is derived from the basic random
timing sequence for the stimuli. By appropriate choice of the shape of the individual portions of the reference waveform
effective low-pass filtering of the signal is
carried out with zero added time delay,
and the signal-to-noise ratio of the output
waveform is increased significantly compared with that obtained using conventional averaging techniques for the same
time period. An unfamiliar feature of the
output is that it represents the response to
multiple stimuli with interflash intervals
which mainly are shorter than the duration
of the output waveform.
Test procedure and comparison with conventional averaging technique. The actual
operation of the system has been described
in detail in its application to VER measurements, and will only be outlined here.1"
The basic system is shown in the block
diagram of Fig. 1. The signal from the
subject (amplified x6,000) provides one
Downloaded From: http://iovs.arvojournals.org/ on 07/28/2017
input to the correlator (Hewlett-Packard
3721A), with the bandwidth controlled by
a low-pass filter, usually set at 500 Hz. The
timing sequence is provided by a noise
generator giving a random-binary-sequence
output at a basic frequency of fr Hz.; the
waveform is illustrated in the upper portion
of Fig. 2. Trigger pulses for the flash stimuli
are derived from the positive-going transitions of this waveform, as shown in the center of Fig. 2. With fr = 100 Hz., the stimuli
may occur at minimal intervals of 20 msec,
or at longer intervals of 30 to 40 msec, etc.
The reference waveform is formed by single
cycles of a sinusoidal waveform at fp = 55
Hz., occurring at the randomly timed intervals as illustrated at the bottom of Fig. 2.
The cross-correlation is carried out for 65.5
sec, with a total of approximately 1,640
stimuli during this interval. Thus, from
Fig. 3, it is evident that approximately 410
interflash periods will be 20 msec, in duration, another 410 will have periods of 30
msec, 307 will have periods of 40 msec,
134
Fricker and Sanders
Frlirtmnj 1975
Fig. 4. A, random flash output fr/fP = 100/55,
500 Hz. bandwidth. B, averaged ERC, stimulus
rate 3 Hz., 256 responses, 500 Hz. bandwidth.
C, averaged ERC, stimulus rate 3 Hz., 256 responses, 150 Hz. bandwidth. D, averaged ERC,
stimulus rate 3 Hz., 256 responses, 50 Hz. bandwidth.
Fig. 5. A and B, OD and OS, random flash outputs, 500 Hz. bandwidth, fr/fP = 100/55. C and
D, OD and OS, averaged ERC's, stimulus rate
3 Hz., 256 responses, 500 Hz. bandwidth. £ and
F, OD and OS, averaged ERC's, stimulus rate
3 Hz., 256 responses, 50 Hz. bandwidth.
and there will be a decreasing number of
longer interflash periods, with only a relatively small number having interflash
periods as long as 100 msec. It should be
emphasized that the sweep duration for
the display of the output is independent of
the interflash intervals; for example it can
be 100 msec, or 10 seconds.
The subject's pupils are dilated, connections are made to a cotton wick electrode
in the lower fornix and the ipsilateral ear,
and the test is carried out under normal
photopic illumination conditions, using a
white stimulus flash from a Grass PS-2
stimulator set at output level two and
placed two feet from the subject's face. A
typical output cross-correlation waveform
on a 100 msec, time base is shown in Fig.
4, A for a normal subject. The waveform
has a well-defined initial dip, followed by a
major peak, and then a second dip. Again
it should be emphasized that this is the
result of summing approximately 1,640 individual responses in 65.5 sec, with indi-
vidual interflash intervals which may be
20 msec, 30 msec, 40 msec, etc., according
to the probability distribution of Fig. 3.
The same subject was tested using conventional signal waveform averaging, with
256 responses summed in approximately
85 sec at a stimulation rate of 3 Hz. The
test conditions were the same as described
for the cross-correlation procedure. The
time base was 100 msec, with the response
sampled at 1 msec increments, and the
high-frequency limit of 500 Hz. provided
by a low-pass filter. Fig. 4, B shows that
the waveform is similar to that given by
the cross-correlation method, but obviously
the signal-to-noise ratio is smaller. The results obtained by decreasing the bandwidth
to 150 Hz. and 50 Hz. are shown in Figs.
4, C and 4, D. These results also show how,
with conventional waveform averaging,
elimination of the higher frequency components by low-pass filtering changes the
overall shape of the response, as well as
introducing an extra time delay.
Downloaded From: http://iovs.arvojournals.org/ on 07/28/2017
Volume 14
Number 2
As an example of how the random flash
method compares with the conventional
average waveform method in a patient with
abnormal retinal function, results are
shown in Fig. 5 from tests with a 27-yearold patient with retinitis pigmentosa. Two
years previously, her ERG had been reported as "extinguished." The random flash
method was used with the same parameters
as described above, giving the results
shown in Figs. 5, A and B for the two eyes.
The average waveforms for the summation
of 256 responses, at bandwidth settings of
500 Hz. and 50 Hz., are shown in Figs.
5, C, D, E, and F. It is evident that the
random flash waveforms art: well-defined,
with the signal from the left eye being
considerably larger than that from the righl
eye. The average waveforms at the same
bandwidth setting of 500 Hz. are not as
well-defined, particularly for the right eye.
With a decrease in bandwidth to 50 Hz.,
the effect of the noise is reduced and the
extra time delay introduced by the filtering
is demonstrated clearly.
As a third example, Fig. 6 shows the
results obtained from a 48-year-old patient
with an unusual localized type of retinal
degeneration. The posterior pole of each
eye, including the macular area, appeared
relatively normal and functioned well.
Further out each retina showed degenerative changes, with thin vessels and some
bone-spicule pigmentation. There were indications that the condition may have been
relatively stationary for the last 30 years.
The upper traces of Figs. 6, A and B show
the random Hash cone response for each
eye, while the lower traces, Figs. 6, C and
D, show the conventional average waveform signal, all at bandwidths of 500 Hz.
The elicet of the noise is very evident in
the average waveform responses, and the
delay times are poorly defined. The random
flash waveforms allow delay times to be
measured in an unambiguous manner, and
provide more precise amplitude measurements.
Normal values for random flash cross-correlation cone responses. Normal values have
been obtained with the implicit (or delay)
Downloaded From: http://iovs.arvojournals.org/ on 07/28/2017
Cone clectroretinography
135
Fig. 6. A and B, OD and OS, random Hash
outputs, 500 Hz. bandwidth, fr/f,, = 100/55.
C and D, OD and OS, averaged ERC's, stimulus
rate 3 Hz., 25o' responses, 500 Hz. bandwidth.
time measured to the major positive peak
(T h ), and the amplitude measured from
the initial dip to the succeeding peak value.
Results are shown in Table I for the implicit times, with a division at age 40 years.
The standard deviations are also given,
and as the distributions were reasonable
approximations to normal Gaussian distributions the +2 standard deviation range
from the mean has been indicated as an
approximate measure of the range for 95
per cent of normal subjects.
A question arises regarding the amplitude
calibration of the random flash outputs.
The amplitude of the output depends upon
both the subject's signal amplitude and the
amplitude and wave shape of the reference
waveform. One method of treating this is
to use the root-mean-square value of the
reference waveform (which is constant for
a given set of parameters) to normalize
the output cross-correlation waveform. The
normal amplitudes listed in Table II were
treated in this manner. The distribution
of amplitudes was not a good approximation to a normal Gaussian curve; consequently only the range of measurements
is indicated.
136
i native Ophthalmology
February 1975
Fricker and Sanders
Table I. Normal delay (implicit) times
(msec.) for RF cone ERG's
< 40 years
(N = 45)
£ 4 0 years
(N = 25)
Delay
time Th
28.1
S.D.
1.8
±2 S.D. range
24.5-31.7
29.3
2.1
25.1-33.5
Table II. Normal amplitudes for RF cone
ERG's (100/55 sine-wave reference)
Average total amplitude
Range
9.3 /*v
2-16 A»V
All nges N = 70.
Discussion
The question arises of whether any significant rod contribution is included in the
random flash output waveform described
above. The probability distribution of the
interflash intervals (Fig. 3) indicates that
only 11 per cent of the total of 1,640 flashes
have interflash periods greater than 60
msec, i.e., at effective stimulation rates less
than 16% Hz. at which the rods might be
expected to contribute if the flash and
background illumination conditions were
appropriate. Such possible contributions are
minimized by the use of white flash stimuli
under photopic background conditions,
with the patient normally light adapted.
Although lack of space does not allow
detailed description here, the random flash
system has been applied to the determination of rod function, using blue flash stimuli
in the dark with an effective average
stimulation rate of 5 Hz. and a maximum
rate of 10 Hz. The rod response lias a
longer implicit time and a different shape
than the cone response described above,
and actually is very similar to a smoothed
version of conventional average waveform
rod ERG's. There is no indication of a contribution of this type in the cone random
flash output signals. From a clinical point
of view, tests with patients with well
documented cases of achromatopsia have
shown no significant: response to the random flash cone test, even though small
Downloaded From: http://iovs.arvojournals.org/ on 07/28/2017
abnormal responses sometimes are obtained
with conventional (slow) average waveform "cone" recordings. Such patients
usually give basically normal rod responses
both by conventional techniques and the
random flash rod technique mentioned
briefly above. Hence, for all practical purposes, there is no indication of any significant rod contribution to the random
flash cone responses with the parameters
described.
The general similarity of the shape of
the random flash response to that of the
conventional average cone ERC is perhaps
unexpected when one considers the distinctly different shape of flicker responses
obtained at fixed stimulus frequencies of
25 to 30 Hz., and indicates that the cone
response to stimuli with random (or
pseudorandom) timing is basically different
than their response to a steady rapid
stimulation. A significant feature here is
the randomly timed nature of the flash
stimuli, which over a short period of time
stimulate the system with a range of input
frequencies, and produce a corresponding
range of output signals. The correlation
process in effect sums all of these varying
frequency flicker type outputs, and gives
the composite waveform as its final output.
These two different types of output waveforms are familiar in the physical sciences,
where for linear systems constant stimulation rate responses are regarded as measures of the frequency response, while the
cross-correlation response to random type
stimuli yields the impulse response of the
system, which of course also can be obtained from an impulse input. For linear
systems, these two distinctly different outputs are mathematically related to each
other, but to what extent such general concepts can be applied to ERG responses is
not clear. It is evident that some analogies
can be drawn, as shown for example by
determining the time delay of the signal
by measuring the signal's relative phase
change with respect to frequency. The synchronous detector procedure (equivalent to
flicker responses at many different fixed
Vohiitir 14
N inn her 2
stimulation frequencies) gives direct phase/
frequency output data, while similar phase
data can be obtained from the Fourier
transform of the random flash response.
The two independent measures of time
delay obtained from these tests usually
agree very closely, thus indicating that the
concepts of steady-state frequency response
and impulse response have some application to ERG measurements.
It is evident that the time characteristics
of the peaks and dips of the random Hash
waveform usually can be measured to the
nearest 0.5 msec., with no ambiguity or
uncertainty concerning these values. The
amplitudes of the signals have a wide
range, so that in describing ERG characteristics it appears most desirable to concentrate upon timing measurements, and
not to rely too much upon amplitude. This
is illustrated by the results shown in Fig.
5 for the patient with retinitis pigmentosa.
The conventional waveforms in Figs. 5, C
and D evidently are all'ected by the poor
signal-to-noise ratio, and little reliance can
be placed upon estimates of implicit times
of such signals. The extra filtering necessary
to increase the signal-to-noise ratio, as
shown in Figs. 5, E and F for a bandwidth
of 50 Hz., allows the modified implicit time
for the left eye to be estimated with somewhat greater reliability, but the right eye
response still presents obvious problems.
The average waveforms in Figs. 6, C and D
have implicit times which can be estimated
only very approximately, and appear either
to coincide with or occur slightly later than
the peak of the corresponding random flash
waveform, rather than earlier as in Fig. 5.
These results illustrate one of the basic
problems with many ERG measurements,
where the delay time or implicit time cannot be specified with any accuracy with
such low signal-to-noise ratios. The practice
of visually estimating the position of the
peak of such noisy signals and drawing an
arrow to indicate this point does not solve
the problem.
To a large extent, the random flash cross-
Downloaded From: http://iovs.arvojournals.org/ on 07/28/2017
Cone elcctroretinographij
137
correlation method for measurement of
cone function yields results which are appropriate for objective measurements of
ERG parameters. The rapid (random)
stimulus rate and flash parameters emphasize cone function over a range of
stimulation frequencies and eliminate rod
responses. In addition, the random timing
characteristics reduce the ettect of artifacts
such as blinking and external interference.
From a practical point of view the procedure is rapid and allows well-defined
cone signals to be obtained in a test procedure of approximately one minute duration.
The authors would like to thank Dr. D. C.
Cognn for his continued support during the course
of this work.
REFERENCES
1. Comas, P.: Electroretinography: some basic
principles,
INVEST.
OPHTHALMOL.
9:
557,
1970.
2. Krill, A. E.: The electroretinogram and
electro-oculogram: clinical applications, INVEST. Ol'HTHALMOL. 9: 600, 1970.
3. Henkes, U. E.: Recent advances in dicker
electroretinography, Doc. Ophthalmol. 18:
307, 1964.
4. Cavonius, C. R.: Color sensitive responses in
the human fhcker-ERC, Doc. Ophthalmol. IS:
101, 1964.
5. Piidmos, P., and Norren, D. V.: Cone spectral
sensitivity and chromatic adaptation as revealed by human Hicker-electioretinography,
Vis. Res. I I : 27, 1971.
6. Flicker, S. |., and Sanders, J. |.: ERC's and
noise: detection probability, time, and amplitude errors, Doe. Ophthalmol. In press.
7. Fricker, S. ).: Application of synchronous
detector techniques for ERG studies in patients with retinitis pigmentosa, INVEST. OIMITHALMOL. 10: 329, 1971.
8. Padmos, P., and Norren, D. V.: The vector
voltmeter as a tool to measure electroretinogram spectral sensitivity and dark adaptation, INVEST. OPHTHALMOL.
11: '783, 1972.
9. Whalen, A. D.: Detection of signals in noise,
New York, 1971, Academic Press.
10. Flicker, S. J., and Sanders, J. ].: Clinical
studies of the evoked response to rapid
random Hash, Electroenceph. Clin. Neurophysiol. 30: 525, 1974.