Improving the reliability of pattern electroretinogram recording

Electroencephalography and clinical Neurophysiology, 84 (1992) 394-399 © 1992 Elsevier Scientific Publishers Ireland, Ltd. 0168-5597/92/$05.00

394

EVOPOT 91036

Technical note

Improving the reliability of pattern electroretinogram recording Jeffrey Froehlich and David I. Kaufman Michigan State University, Neuro-Visual Unit, East Lansing, MI 48824-1316 (U.S.A.) (Accepted for publication: 24 February 1992)

Summary The pattern electroretinogram (PERG) is a small electrical response of the retina to a reversing checkerboard pattern, usually less than 6/xV in amplitude. Unfortunately, the PERG can be obscured by artifacts such as blinks, eye movements, poor fixation, and amplifier saturation. Amplitude criterion artifact rejection systems found on commercial signal averagers eliminate large amplitude artifacts but are insensitive to small amplitude artifacts associated with amplifier saturation. Such saturation often occurs for several recording sweeps after large amplitude signals such as eye blinks are rejected. The presence of post-saturation artifacts complicates clinical PERG analysis. In this paper we describe procedures to remove these small amplitude artifacts from the PERG. These include computer selection of inputs for averaging and use of tracings with small input numbers to approximate PERG amplitudes. These procedures greatly reduce the variability of PERG amplitudes in the normal population, making PERG amplitude a more reliable clinical measure.

Key words: Pattern electroretinogram; PERG artifacts; PERG methodology; VEP

Several reports have suggested that the clinical usefulness of the pattern electroretinogram (PERG) is compromised by the large variation in PERG amplitudes for normal subjects (Holopigian et al. 1988; Rimmer and Katz 1989; Juen and Kieselbach 1990). This variability is reflected in lower limits of normal B-wave amplitude reportedly as low as 0.5/xV for testing with 30 min checks (Celesia and Kaufman 1987; Kaufman et al. 1988; Tan et al. 1989; Odom et al. 1990; Song and Wray 1991). With such low normal limits, virtually any measurable amplitude would be normal and yet easily obscured by unfavorable signal-to-noise conditions. We wanted to determine whether such low limits reported for normal PERG amplitude could be partly due to the recording of fiat input artifacts. These artifacts are often generated for several recording sweeps after amplifier saturation by large amplitude signals such as blink artifacts. Because these fiat signals are not rejected by amplitude criterion artifact reject systems, they reduce PERG amplitudes by averaging zeroes into the final wave form. We show the effect of these flat input artifacts on the statistics of PERG amplitudes from 50 normal subjects. In this paper, we present a series of steps to remove these artifacts from the PERG. A computer program to eliminate flat artifacts from being averaged is used,

Correspondence to: David I. Kaufman, D.O., Michigan State University, B-309 West Fee Hall, East Lansing, MI 48824-1316

(U.S.A.).

and results obtained are compared with PERGs done with visual monitoring. The possibility of using recordings which are averages of as few as 20 inputs to estimate PERG amplitudes is also studied.

Methods and materials

The visual stimulus for all simultaneously recorded PERGs and VEPs (PERG/VEP) was a black and white checkerboard screen consisting of 30 min checks reversing luminance at a rate of 1.8/sec. Mean luminance of the monitor screen was 34 foot-lamberts with a contrast of 94%. PERGs were recorded with bipolar Burian-Allen contact lens electrodes, after refraction to best visual acuity. Further details for recording the P E R G / V E P are described elsewhere (Froehlich and Kaufman 1991). Unless explicitly stated all P E R G / V E P tracings are averages of 100 inputs recorded with a Nicolet Pathfinder II evoked potential mini-computer at a frequency bandpass of 1-100 Hz. Artifact reject was set for 47.5/zV and amplifier gain at 60,000. PERG A-wave and B-wave peaks and VEP N70 and P100 peaks were labeled on the final wave forms. B-wave amplitude was the voltage difference between the A-wave and B-wave peaks. Positive voltage is up for PERGs, down for VEPs. Amplitudes for all control recordings, including normals, were measured from the computer-generated

IMPROVING PERG RECORDING

395

(a)

itoring or by a computer program (described below) which rejected fiat input prior to averaging. (2) During 100-input recordings with visual monitoring, every occurrence of a post-saturation flat signal noted on inspection was manually counted. If this count exceeded 10, the test was repeated until two tracings with 10 or fewer such artifacts were recorded. (3) The presence of post-saturation artifacts in the monitored input was used to elicit patients' voluntary control of eye movements and muscle tension, particularly around the orbits. U p o n request, patients were usually able to suppress such tension, with marked decreases in post-saturation artifacts. (4) Input for low input recordings (fewer than 100 inputs) was analyzed by a computer program (described below), running concurrently with the test, which identified post-saturation flat signals and rejected them for averaging. Low input recordings, including single input ones, were done with continuous check reversal between recordings.

Post-Saturation (Fiat) Artifacts

Physiologic Inputs B~ AO

a~ AS

.t4 84

A3

A2 8J A1

I

I

I

I

I

i

I

I

I

I

I I I I I l l l l l

(~

BJ

,44

Computer program to monitor and select inputs I

I

I

I

I

I

I

I

I

I

I

I

I

I

I

~

l

I

1

a

Fig. 1. a: examples of physiologicinputs and post-saturation artifacts. b: sample computer analysis to distinguish physiologic inputs from flat artifacts. Coefficients of determination are reported near the regression lines for the 20 msec intervals 15-35 msec and 140-160 msec. The 5 points used for each regression line are indicated on each input. If either of the two coefficients was greater than 0.95, an input was rejected by the computer program. In a, there are two unusual inputs where amplifier de-saturation occurred within the sweep. These would be rejected. One unit on the horizontal axis is 25 msec for all tracings in this report. In a, one vertical unit is 10 ~V, in b 2.5/zV. Inputs marked with (*) were rescaled for the analysis in b.

off-line averages of duplicate tracings. Because of inter-eye correlations (Rosner 1982), One eye randomly chosen from the two eyes tested in any control subject was used for analysis. Limits of normal for all measured quantities were defined to be 2.5 S.D. above and below calculated means. Fifty subjects, age 18-48 years, with normal neuroophthalmologic examinations, were studied. All subjects gave written consent for their participation in this study.

PERG recording method." elimination of post-saturation artifacts Flat saturation artifacts do not have the usual high frequency (25 Hz or greater) components normally found in physiologic input (Fig. la). The following steps were taken to eliminate these flat signals: (1) Input was constantly monitored, either for acceptance of post-saturation artifacts during visual mon-

Artifact reject systems using amplitude criteria for input rejection do not eliminate post-saturation flat signals from being averaged. Essentially a straight line over a small 20 msec time interval, a post-saturation signal has the characteristic that the regression line determined by any number of points on that interval will closely approximate the signal itself. This will not be true for a physiologic signal, which has high frequency variation over that 20 msec interval. Thus, after an input was initially accepted, an on-line program determined voltages for that input at 5 msec intervals between 15 and 35 msec and similarly between 140 and 160 msec. This program then performed linear regressions (Hogg and Tanis 1977) for the 5 data points on each 20 msec interval. The coefficient of determination (Huntsberger 1967) was calculated for each regression line. If both coefficients were less than or equal to 0.95, the input was accepted for averaging. This continued until a predetermined number of inputs was accepted. Fig. lb illustrates the analysis. During analysis of an input, no other inputs could be accepted. The coefficient of determination always lies between 0 and 1 and measures the goodness of fit of the regression line to the points determining it. A value of 1 means a perfect fit. One hundred post-saturation flat signals used to set the rejection criterion of 0.95 all had coefficients of determination of 0.96 or greater. One hundred test physiologic inputs had a much greater range (0.00-0.70) because the 5 points used for regression would occasionally lie close to the regression line. Because the program required about 1 sec per input, it was less attractive for 100-input recordings since it tripled the recording time.

396

J. FROEHLICH, D.I. KAUFMAN

only 2 recordings, 8 required 3, and 12 needed 4 recordings before two such tracings were obtained. (Note that in Table I, the 30 subjects requiring only 2 recordings were used in both group I calculations and group III calculations described below.) The data obtained from the 8 subjects requiring 3 recordings and from the 12 subjects requiring 4 recordings to obtain two acceptable tracings are analyzed separately (Table I, group II). For these 20 individuals, the mean B-wave amplitude for the tracings having more than 10 post-saturation artifacts (the range was

Results

Laboratory normative data for PERG amplitudes PERGs from 50 normal subjects, recorded during visual input monitoring, provided our laboratory normals (Table I, calculation group I). The lower limit of normal B-wave amplitude was 1.8 /xV for these subjects. To set this limit, only duplicate 100-input recordings having 10 or fewer flat saturation artifacts by direct count were used. Of the 50 normals, 30 required (a)

Averages of Indicated Number of Single Inputs to the Left

Single Input Recordings

80

Ae"

~

B7

~ i a l s

~7

6~ A8

J'lO0 inputs each

AT-

B: 45 e, A~

2 e,

,12

I

I

I

I

I

I

I

I

I

1~

I

I

I

I

I

I

I

I

I

I

I

I

I

I

I

I

I

I

i

Ibl Single Inputs Duplicate Tracings of 1O0 PERG with Indicated Number Before Patient Instruction Inputs After Patient of Inputs After Instruction B-wave Amp = 3.2 /~V 8~

t

8B ,,leB7 aT"

b J~ e ~l

4

o tracings w )

A6-

es. A5

B4

84

.5

A4 B3 A3

a,? A2

/

aJ

Bi

VEP

AJ

.1

5 I

[

I

I

I

J

I

I

J

I

I

I

I

I

I

~L

Z ~

L

I

I

I

I

I

I

I

P

I

I

(el Duplicate Trials of 1O0 Offline Average of Inputs in Difficult Patient. 16 Singly Recorded Inputs B-wave

A.~

Amp = 4.75/tV

A3

I

I

l

I

I

I

I

I

I

I

I

I

I I

I

I

I

J

l

I

Fig. 2. Single input and low input recordings from 3 individuals. (a) Single input recordings and cumulative averages in easily recorded PERG. Duplicate 100-input recordings shown for comparison. Cumulative ,averaging was done with single inputs in the order indicated by the numbering. (b) Individual who had given a spuriously low PERG amplitude before appropriate instruction and encouragement during continuous monitoring of input, Visual acuity was 20/20. (c) Normal individual whose normal PERG amplitude was recorded only by collecting 16 single inputs free of post-saturation artifact. One unit on the vertical axis represents 2.50 p,V for all PERG tracings of more than one input. For the single input recordings in a and b, one vertical unit equals 10.0 p,V for ease of illustration. One horizontal unit equals 25 msec.

IMPROVING PERG RECORDING

397

TABLE I Effect of saturation artifacts on the statistics of normal B-wave amplitudes. Calculation

B-wave amplitude (~V)

group

Mean (S.D.)

Lower limit of normal (mean minus 2.5 × S.D.)

4.1 (0.92) Range (2.5-7.3)

1.8

3.9 (0.80) range (2.5-5.6)

1.9

1.7 (1.15) range (0.6-3.5)

0.0

3.2 (1.05)

0.6

I

II

III

50 normals (10 or fewer saturation artifacts/ recording) Tracings from subgroup of 20 normals from group I subjects having: (A) 10 or fewer saturation artifacts (used for group I calculations) (B) Greater than 10 saturation artifacts (range: 23-67) 50 normals (using PERGs from group liB having greater than 10 saturation artifacts instead of those from group IIA)

23-67 such artifacts) was less than half the mean amplitude for tracings having 10 or fewer. Table I (group III) also shows that the normal lower limit of PERG amplitude would only be 0.6 /xV if tracings having greater than 10 saturation artifacts (B) from the 20 group II subjects were used in the normal data base instead of the tracings with 10 or fewer (A). Introducing such spuriously low amplitudes into normal data bases may account for the low limits of normal B-wave amplitude commonly reported.

Post saturation flat signals - recordings with small numbers of inputs Fig. 2 shows computer-monitored PERG recordings with small numbers of inputs from 3 different subjects for comparison with 100-input recordings obtained with visual monitoring. Fig. 2a shows 16 consecutive singly recorded inputs from one normal subject. Two postsaturation signals were not rejected by the artifact reject system. Fig. 2b shows how prevalent fiat artifacts can be (8 consecutive accepted inputs, left panel). After these were detected by visual input monitoring, the patient was encouraged to control both eye movement and

muscle tension around the eyes. This greatly reduced the production of flat signals, making a normal PERG recordable (middle panel). On the right are 4 consecutive computer monitored collections of 5 inputs. Fig. 2c (left panel) shows two PERG recordings of 100 visually monitored inputs each in a normal subject for whom it was impossible to get continuous stretches of input, free of flat post-saturation artifact. By manual count, 62 and 70 of the accepted inputs were fiat artifacts for these recordings, providing an objective reason for not including this subject's PERG amplitudes in the normal data base for Table I. Indeed, in Fig. 2c, the average of 16 computer-monitored artifact-free inputs indicates a normal amplitude could be obtained for this individual. The computer program analyzed 50 inputs to find these 16 non-flat signals. All other input was post-saturation artifact. The subject in Fig. 2c was 1 of 4 controls initially recruited from whom artifact-free 100-input recordings were unobtainable. In each case a normal amplitude was obtained with recordings having small numbers of inputs.

Statistics of recordings with small numbers of inputs In Fig. 3a are shown the PERGs resulting from 5, 10, 15, 20 and 100 computer-selected inputs recorded from a normal subject. Duplicate recordings of 100 inputs using visual monitoring provided a reference PERG amplitude. The numbers in parentheses represent fractions of the reference amplitude. Fig. 3b illustrates the results of doing this procedure on one eye of 20 normal subjects. The data show that as few as 15-20 inputs can give a good estimate of reference PERG amplitude. Thus, the amplitudes reported in the right panels of Fig. 2b and c are correct to within 250% of the actual value, using normal limits (mean _+ 2.5 S.D.) calculated from the variation illustrated in Fig. 3b for 15-20 inputs.

Discussion

In this report we address a problem in the measurement of PERG amplitudes which has made the PERG seem clinically unreliable as an indicator of retinal function (Holopigian et al. 1988; Juen and Kieselbach 1990). Namely, tracings from individual patients may have spuriously low amplitudes due to contamination with flat inputs from saturated amplifiers. This appears to be particularly troublesome in low vision patients. Furthermore, spuriously small amplitudes recorded from control subjects for a normal data base may make lower limits of normal so low that a "normal" amplitude might be any amplitude greater than zero. Although there are other types of poor quality PERG input, post-saturation signals are prevalent

398

J. FROEHLICH, D.I. KAUFMAN

(a)

# of (Fraction monitored B-wave of ,nputs Amp Ref. Val. J' ~ 3.74 ,u~V(o.go)

Duplicate tracings of 100 inputs

.43

~ ,t2

B-wave Reference = 4.26/~V

.

A_~/V~t'63/~V 11.07I '

W

^

A

A-wave

~o

__1

[

I

I

I

I

I

I

I

I

I I 1 ~ 1 1 1 1 1 1

(b) 1.2

'~E 1.0

o.4 g~ "~ ~ °'6 o.2f 0.0

5

lb

1'5 20

:'1(~0

# of MonitoreEI )nputs Fig. 3. Single-trial PERG recordings and amplitudes as function of number of monitored inputs for one individual (a) and for normal population (b) of 20 individuals. B-wave amplitudes are reported in a along with the fraction of the reference B-wave amplitude in parentheses. In b, error bars represent +_1.0 S.D. Reference amplitude is the amplitude of duplicate 100-input trials, as in a. One horizontal unit equals 25 msec.

enough to affect statistical norms and can be eliminated once detected. Removing these artifacts greatly reduces the ambiguity in deciding the normalcy of PERG amplitudes. The lower limit of normal for Bwave amplitude (Table I, group I) is high enough to ensure that low amplitude PERGs are unequivocally abnormal, given that monitored input is of high quality. In this study we have used both manual and computer monitoring of input to eliminate post-saturation artifacts from averaging, with both methods resulting in the same B-wave amplitudes (Fig. 3). By including the method of binocular viewing commonly used in patients with monocular visual loss (Chiappa 1990; Froehlich and Kaufman 1990b), the chances of obtaining recordings free of such artifacts are greatly improved. The use of as few as 15-20 inputs to estimate amplitudes in difficult cases is a novel approach to

PERG recording, justified by the data in Fig. 3. The importance of accepting input free of flat contaminants should be emphasized. In our experience with difficult recording subjects (Froehlich and Kaufman 1990a), direct monitoring shows that input quality often decreases with increasing test length, thwarting a strategy requiring large input numbers. Indeed, much of the improvement in B-wave amplitudes found in a study of 15 unilateral optic neuritis patients using binocular viewing of the monitor to test the affected eye (Froehlich and Kaufman 1990b) appeared to be due to a great reduction in post-saturation artifacts compared with monocular stimulation. In any event, our method offers a necessary assessment of PERG quality at the input level. Without this assessment the only way to have been certain that the technical quality of a PERG was adequate was if the final result was normal. Consequently, reduced amplitude PERGs could not be considered reliable evidence of abnormality, because of the specter of unknown sources of technical error coupled with low normal limits for PERG amplitude. Our method, by improving the recording of individual tests and creating clinically useful limits of normal, makes a PERG of low amplitude as reliable a finding as one of normal amplitude.

References Celesia, G.G. and Kaufman, D.I. Effect of gender and aging on the pattern electroretinogram. Electroenceph. clin. Neurophysiol., 1987, 68: 161-171. Chiappa, K.H. Evoked Potentials in Clinical Medicine. Raven Press, New York, 1990: 95-110. Froehlich, J.E. and Kaufman, D.I. PERG may be falsely reduced in patients with low vision: an algorithm to ensure proper recording. Neurology, 1990a, 40: 322. Froehlich, J. and Kaufman, D.I. Use of inter-eye amplitude ratio and binocular viewing in monocular visual loss. Invest. Ophthalmol. Vis. Sci., 1990b, 31: 415. Froehlich, J.E. and Kaufman, D.I. Effect of decreased retinal illumination on simultaneously recorded pattern electroretinograms and visual evoked potentials. Invest. Ophthalmol. Vis. Sci., 1991, 32: 310-318. Hogg, R.V. and Tanis, E.A. Probability and Statistical Inference. Macmillan, New York, 1977: 232-249. Holopigian, K., Snow, J., Seiple, W. and Siegel, I. Variability of the pattern electroretinogram. Doc. Ophthalmot., 1988, 80: 104-115. Huntsberger, D.V. Elements of Statistical Inference. Allyn and Bacon, Boston, MA, 1967: 255-284. Juen, S. and Kieselbach, G.F. Electrophysiologic changes in juvenile diabetics without retinopathy. Arch. Ophthalmol., 1990, 108: 372-376. Kaufman, D.I. and Froehlich, J.E. Serial PERG N95 amplitude measurements in optic neuritis. Invest. Ophthalmol. Vis. Sci., 1991, 32: 949. Kaufman, D.I., Lorance, R.W., Woods, M. and Wray, S.H. The pattern electroretinogram: a long term study in acute optic neuropathy. Neurology, 1980, 38: 1767-1774. Odom, J.V., Feghali, J.G., Jin, J. and Weinstein, G.W. Visual deficits in glaucoma. Arch. Ophthalmol., 1990, 108: 222-227.

IMPROVING PERG RECORDING Rimmer, S. and Katz, B. The pattern electroretinogram: technical aspects and clinical significance. J. Clin. Neurophysiol., 1989, 6: 85 -99. Rosner, B. Statistical methods in ophthalmology: an adjustment for correlation between eyes. Biometrics, 1982, 38: 105-114.

399 Song, H. and Wray, S.H. Bee sting optic neuritis. J. Clin. NeuroOphthalmol., 1991, 11: 45-49. Tan, C.B., King, P.J.L. and Chiappa, K. Pattern ERG: effects of reference electric site, stimulus mode and check size. Electroenceph. clin. Neurophysiol., 1989, 74: 11-18.