- Email: [email protected]
Automated perimetry-Part I
special feature Automated perimetry-Part I Richard P. Mills, M. D. Seattle, Washington
A revolution in perimetry is occurring. This revolution is no less dramatic than the one which has occurred in cataract surgery, and it is developing along parallel lines. The use of intraocular lenses forced a reanalysis of extracapsular extraction, which made lens placement technically easier and seemed to produce fewer complications than intracapsular extraction. Similarly, the automated perimeter is forcing a reawakening of interest in the static method of exploring the visual field. It is technically easier to computerize than the kinetic method and is more sensitive to field defects. KINETIC PERIMETRY The tangent screen and the Goldmann perimeter are the time-honored devices for testing visual fields. Stimuli of various intensities are moved from the non seeing periphery toward the center, and the patient responds when he or she detects the stimulus. The border line between seeing and non seeing can be drawn by connecting the points derived from a single stimulus strength, the familiar isopter of kinetic perimetry. If the chosen stimulus is quite weak, the isopter may form a circle around fixation inside the physiologic blind spot. If the stimulus is strong, the maximum field area is outlined by the isopter at the Goldmann perimeter, whereas the limits of testing at the tangent screen are exceeded. An array of isopters can be plotted on a visual field chart and all ophthalmologists feel comfortable interpreting such charts. Unfortunately, however, kinetic perimetry suffers from four principal disadvantages that limit its usefulness. To explain these disadvantages requires reference to the old Traquair analogy of the island of vision in a sea of blindness (Figure 1). The highest points on the island represent good sensitivity, at which weak stimuli can be detected. Down near the waterline, only the strongest stimuli can be seen, and sensitivity is poor. The hill is of course enshrouded in fog (Figure 2) and our job is to map it.
- -<::::::: :::;::
:::: ::: ::::::::::::
:::::
-~:;;;::::::::::::::
Fig. 1.
(Mills) Traquair's island of vision in a sea of blindness. The highest point, representing the fovea, has the greatest sensitivity to test stimuli. At the water level the sensitivity is zero and no stimulus can be seen. The slope of the normal island of vision varies, being steepest close to the fovea and at the edge of the island, and more gentle in the intermediate sections.
Fig. 2.
(Mills) The problem in clinical perimetry is that the island of vision is enshrouded in a fogbank and we must use indirect methods of mapping it. The peak is exposed, since we know something about it from visual acuity determinations, but the rest of the island is uncharted.
Reprint requests to Richard P. Mills, M. D., Department of Ophthalmology,
RJ -10, University ofWashington,
AM INTRA-OCULAR IMPLANT SOC J-VOL 10, SUMMER 1984
Seattle, Washington 98195. 347
In kinetic perimetry, a stimulus of known strength is moved centrally, like an airplane flying toward the island at known altitude. The airplane "stimulus" will crash when it hits the island. After a number of airplane stimuli have crashed at one altitude, we can draw an isopter (Figure 3). After drawing several isopters, we will have a topographic map of the island of vision.
~llil ~
Fig. 3.
flll'll'"
",7, "" "'" \ \,~
(Mills) One difficulty with kinetic perimetry is that depressions of the interior of the island of vision are easy to miss. The airplanes will all crash on the periphery of the island or fly over the depression to crash centrally. Only if we are lucky enough to start an airplane in the center of the depression will we discover its presence and be able to map it.
Fig. 5.
(Mills) Another difficulty with kinetic perimetry is that flat sections of the island are difficult to map. Airplanes flying toward the island at one altitude may crash centrally, while those flying at a slightly different altitude will crash peripherally. Moreover, undulations of the flat section will be undiscovered.
(Mills) One method of mapping the island is to fly many airplanes toward it at the same altitude. We could record where they crashed, and knowing the altitude, we could draw an isopter. After drawing several isopters we would have an idea of the shape of the hill of vision. The same principle is used in kinetic perimetry as practiced on the Goldmann instrument or tangent screen. Stimuli are moved centrally until seen, and all points discovered by the same stimulus are connected by an isopter line.
The first of the four disadvantages of kinetic perimetry is that hollowed-out valleys in the interior of the island of vision are difficult to detect (Figure 4). Stimuli (airplanes) coming from the outside will miss them. We could start the stimulus within the valley, of course, if we knew where to start. Since we do not, we would be forced to select places to start the stimulus randomly, hoping we would start inside a valley (scotoma). The chance of missing isolated valleys (scotomas) is high, especially with small or shallow ones. The second disadvantage is that kinetic perimetry does a poor job of mapping flat sections of the island of vision (Figure 5). An airplane stimulus coming from outside at one altitude may crash quite a distance away from one coming at a slightly different altitude. Undulations of the flat section, which may be quite important, will be unmapped. Most patients have at least some relatively flat areas in their visual island, like a plateau, but in some patients the entire island resembles a mesa. In such patients, kinetic perimetry tends not to be a very productive exercise. Third, kinetic perimetry by definition uses moving stimuli. It has long been known that moving stimuli are easier to detect than stimuli that are standing still, 348
Fig. 4.
especially within defective areas of the visual field. This is known as statokinetic dissociation, and contrary to popular belief, the phenomenon occurs with lesions at all levels of the visual pathways, though it may be most prominent in occipital lobe disease. The fact that the stimulus is moving may cause a visual field defect to be missed. Finally, kinetic perimetry has proven very difficult to automate. For example, the Coherent Perimetron had difficulty mapping scotomas, particularly those which "broke through" to the periphery. Inordinate amounts of time were consumed as the computer seemed to get lost inside the scotoma. Production of the instrument was subsequently discontinued. Thus, no competitively priced automated kinetic perimeter is currently available commercially.
AM INTRA-OCULAR IMPLANT SOC J-VOL 10, SUMMER 1984
Before the advent of automated perimeters, Armaly developed and Drance modified a method of screening for glaucomatous field defects which lessened the effect of the first three disadvantages of kinetic perimetry. They did so by including in the test strategy a series of stimuli presented statically within the central visual field. So the Armaly-Drance method of glaucoma screening is actually a mixture of kinetic and static perimetry. STATIC PERIMETRY Let us return to the island-of-vision analogy to explain how static perimetry works. We could drop parachutes with altimeters onto the island and record the height at which each one landed (Figure 6). Thus, at a number of discrete locations in space, we would know the height (sensitivity). This method is known as threshold static perimetry since it determines the threshold between seeing and nonseeing at a number of preselected locations .
45' o-~----f§~-+------22 '
(
~----~~~;;;--'.:--35· Ifl lilit / I I i i I \ \ \ \ , \ \ \' \, \
Fig. 6. (Mills) Another me thod of mapping the island of vision is to drop parachutes at known coordinates toward the island, recording the height at which they land. The principle is the same for static threshold determinations in clinical perimetry. While time consuming, we know all there is to know about each point and can map the island if we test enough points.
Observe that threshold static perimetry overcomes the disadvantages of kinetic perimetry. Isolated valleys (scotomas) in the island of vision are less likely to be missed if one uses a large enough number of parachutes (stimuli). Flat areas of the island are accurately defined, and undulations unmasked. The stimulus is not moving, and the process is easy to computerize. Unfortunately, threshold determinations are very time consuming, and long testing sessions are required to develop adequate point density. The reluctance of ophthalmologists and patients to commit to long testing sessions is a problem, but this becomes even worse if screening of significant patient numbers is required. In an effort to save testing time but retain some of the advantages of static perimetry, suprathreshold static testing methods were developed. Helicopters are employed to fly to certain locations over the island of
Fig. 7.
(M ills) Anothe r m ethod of mapping the island of vision is to employ helicopte rs to fly to known coordinates and altitude and report simply whether they are in the air or have lande d . The same principle applies for suprathreshold testing in clinical perimetry. A large number of points can be tested in a short time, but we know only whether each point is seen or not seen. With a sufficient variety of stimulus strengths, a map of the island of vision can be generated.
vision and hover at a specified altitude (Figure 7). They radio whether they are still in the air or whether they have landed. A large number of points in space at different altitudes can be tested in relatively short order. If we select our points well, we can detect valleys (scotomas) in the interior of the island. We will still have trouble defining flat sections of the island (field) and determining undulations. But we are not moving the stimulus, and automation of the process is easy. One real problem with suprathreshold static perimetry is that it is easy to waste time testing with stimulus strengths that are either way above threshold or way below it. Consider the example of a single stimulus strength tested at about 150 locations, a simple strategy employed by many automated perimeters. Toward the periphery of the field, the stimulus is far too weak to be seen, and toward the center it is too strong. If our objective is to detect subtle scotomas, we will find them only in areas where the stimulus is presented close to the threshold between seeing and nonseeing. We can get around this problem somewhat by testing these 150 locations at multiple stimulus strengths. In other words, use several "runs" with the automated perimeter, at the cost of added test time (and more visual field charts to shuflle). If we only knew what the expected threshold between seeing and nonseeing was at any point on the island, we could select our stimulus strength close to the expected threshold. This would allow us to detect scotomas across the whole field because we would be using different stimulus strengths at different places in the field: weaker toward the center and stronger toward the outside. To accomplish this, we could use a threshold static method at a few points in the field, determining the actual threshold at those points. We could then select
AM INTRA-OCULAR IMPLANT SOC J-VOL 10, SUMMER 1984
349
stimuli for 'our suprathreshold test related to (interpolated from) those few threshold determinations. Such a strategy is called threshold-related suprathreshold static perimetry. Other threshold-related strategies for suprathreshold static perimetry select stimuli based on a "normal" visual field (derived from large populations and corrected for patient age) or the "normal" gradual decrease in sensitivity from the foveal threshold at more peripheral visual field locations. Obviously, with the many types of threshold-related strategies, it is important to inquire which type was used before interpretation is attempted. In summary, three types of static perimetry are commonly used. Threshold static perimetry determines the threshold stimulus strength at specified locations in the field. Suprathreshold static perimetry presents a preselected stimulus at specified locations and determines whether it is seen or not. Thresholdrelated supra threshold static perimetry presents stimuli at specified locations, with stimulus strength chosen on the basis of a few known threshold points. THE UNCERTAINTY PRINCIPLE IN PERIMETRY Up to this point, we have been talking about perimetry as though the threshold of a point on the island of vision is an absolute number. Stimuli stronger than threshold would always be seen and stimuli weaker than threshold would never be seen. U nfortunately, this is not the case. Threshold is actually a relative number, referring to the level at which a patient will see 50% of the stimuli presented. A stimulus stronger than threshold increases the chance of seeing the stimulus and a stimulus weaker than threshold decreases the chance of seeing it (Figure 8). This uncertainty is common in biologic systems. For example, recall the LD50 concept from toxicology, a dose level at which half of the animals receiving a drug die from its toxic effects. The uncertainty principle plays a role no matter which type of perimetry is being performed, but it is most measurable by threshold static testing. We can measure threshold at a single point over and over again, and obtain different numbers within a range. Presumably the "true" threshold lies somewhere near the middle of that range. The uncertainty is magnified by lapses in patient attentiveness caused by distractions, fatigue, or disease. It is also magnified in areas of defective visual field. Some evidence is accumulating that the earliest sign of glaucomatous field loss is a local widening of the zone of uncertainty. Stated differently, this early sign is represented by an increased local fluctuation in multiple threshold determinations. This may happen while the average measured threshold is still normal. 350
Q)
U
c
0
..r:.
-
U c
Q)
u
.... Q) a..
VI :;) :;)
E
100
(/)
0>
c
'(ij Q)
(/)
--0
0 50
55
59
Brightness of Stimulus Fig. 8.
(Mills) The uncertainty principle in perimetry. As a stimulus becomes brighter, the chance that it will be seen by the patient increases. The 50% level is called the threshold. In this example, a stimulus of brightness 55 is seen half the time and not seen half the time. Brightness 55 thus represents the threshold for that point in this patient. However, in attempting to measure threshold, a stimulus of 50 will be seen 5% of the time and a stimulus of 58 will be missed 5% of the time. Multiple trials could establish a mean threshold, but time expenditure would be excessive if we tested many points.
For the clinician, the uncertainty of the visual field measurements represents a major source offrustration. Often there is a temptation to introduce shortcuts in the testing process, on the grounds that field testing is inexact at best. We should remember, however, that such shortcuts usually result in an increased uncertainty to the point that the field chart is almost impossible to interpret with any degree of confidence. STIMULUS v BACKGROUND INTENSITY
A background light level is used in perimetry so that the whole retina can remain in a light-adapted state. Otherwise a stimulus would produce local light adaptation which would take many minutes to return to baseline, preventing repeat testing at that location. The Goldmann perimeter is calibrated for 31.5 apostilbs (as b) background illumination, as are most of the currently available automated units. Such a level was adopted as a standard by the International Council of Ophthalmology. However, the Oculus (Tubingen) uses 10 asb, the Octopus 4 asb, and the Rodenstock Peritest 3.1 asb as a background. These instruments seem to function well in practice, since such background levels are still within the lower end of the photopic range. A stimulus is seen when it is sufficiently brighter than the background to enable detection. It is the difference in brightness between stimulus and background that actually elicits a patient response. (The smallest difference that can be appreciated 50% of the time is the threshold). There are physical factors in
AM INTRA-OCULAR IMPLANT SOC J-VOL 10, SUMMER 1984
bulbs and light emitting diodes which limit the maximum stimulus intensity that can be presented. Thus, there is a usable testing range for each instrument, limited on the lower end by the background intensity and on the upper end by the physical limits of stimulus brightness.
Table 1. Decibels (db) and apostilbs (asb). Stimulus Intensity (asb)
Decibel Value (db)
10,000 (maximum)
0
1,000
10
100
20
APOSTILBS, LOG UNITS, AND DECIBELS
32
25
Why not just provide everyone with Goldmann equivalences for stimulus intensity on different devices and forget about apostilbs, log units, and decibels? Most of us understand what a Goldmann I-2-e represents, so why change the terminology? The reason is simple. The Goldmann stimuli we understand are used mostly in a kinetic mode, but the new automated devices are static perimeters. There is poor correlation between Goldmann stimulus numbers and automated stimulus numbers across patient populations. Moreover, the Goldmann perimeter uses size as well as brightness change to vary the stimulus, while many automated units do not. Size and brightness are related in a general sense, but not precisely enough to allow use of Goldmann terminology in automated perimetry. An apostilb is a measure of light brightness, in a physical sense. The eye, however, perceives light within a moderate range of intensity on a basis that approximates a logarithmic scale. That is, an illumination change of a stimulus from 10 to 100 asb gives roughly the same change in sensation as an illumination change from 100 to 1000 asb. Because of variability in such a subjective grading of brightness, the relationship does not hold exactly for all patients, but it is a useful approximation. Therefore, stimulus intensities can be expressed in log units and approximate the way the eye actually sees the stimulus. Because the stimulus can be made infinitesimally dim, starting the log scale at the lower end is impractical. Consequently, the maximum intensity an instrument can deliver is set at 0 log units, and dimmer intensities are expressed as the number of log units less than the brightest possible stimulus. The higher the number, the dimmer the stimulus. We all remember the decibel from audiology as a logarithmic function of sound intensity. It was "invented" because the ear responds to sound in a logarithmic way, just as the eye responds to light. A decibel is nothing more nor less than one tenth of the log unit. In perimetry, the decibel is used to describe brightness in logarithmic terms. Like the log unit, it is set at zero for the maximum stimulus intensity the instrument can deliver, and the number of decibels refers to a reduction from peak intensity. For example, if the maximum stimulus intensity is 10,000 asb, and the chosen stimulus is 1,000 asb, the decibel number is 10. If the maximum intensity is 10,000 asb, and the
10
30
1
40
chosen stimulus is 100 asb, the decibel number is 20, and so forth (Table 1). Observe that since higher numbers in log units and decibels refer to dimmer stimuli, whe n printed on a visual field chart, a higher number means a greater retinal sensitivity. If we are thinking of the island of vision, a higher number (log unit or decibel) means a higher altitude on the island (Figure 7). Likewise, a zero reading means that we are at the waterlinewithin the sea of blindness. The "new" terminology is fairly easy to get used to, given a little practice and a willingness to abandon the " Gold" standard.
SCREENING, DIAGNOSTIC, AND QUANTITATIVE FIELDS In the days of tangent screens, Goldmann perimeters, and human examiners, we did perimetry by finding defects, characterizing them well enough to fit them in a diagnostic category, then quantifying the defects for later comparison. The process of doing a field was viewed as a continuum. It is not, as any automated perimeter will tell you. The functions of screening, making a diagnosis, and quantitating defects require quite different strategies of testing (Table 2). A human examiner can make the strategic transitions easily, but we have found that computers need to be told when to change their strategies to accomplish the different functions required of them. In screening, the object is to tell if a visual field is abnormal or normal, and nothing else. Does this patient with high intraocular pressure have a field defect or not? We demand from test strategies used in screening that they be sensitive enough to detect most of the Table 2. Screening, diagnostic, and quantitative fields. Visual Field Task
Question to Answer
Screening
Is there a d efect or not?
Diagnosis
What sort of defect is it?
Quantitative
What does the defect look like?
AM INTRA-OCULAR IMPLANT SOC J-VOL 10, SUMMER 1984
351
defects, yet selective enough so a lot of false positives do not appear. It helps a great deal if we know what sort of defect is expected. Screening in patients with ocular hypertension should look for nasal steps, paracentral scotomas, Bjerrum scotomas, and temporal wedge defects. Screening in a stroke patient should look carefully for hemianopic defects. Screening in a patient with "bone spicules" of pigment in the fundus should concentrate on ring scotomas. Reliable screening strategies produce detection rates of90% or better with false alarm rates below 10%, and are available on most automated perimeters. Incidentally, rates for manual perimetry as practiced in the community are no better, and are often worse. Unfortunately, perfection is impossible to achieve. If one tries too hard to improve detection rate, false alarms become a major problem. If one tries too hard to minimize false alarms, then detection rate falls. A screening visual field does contain diagnostically useful information, which is sometimes enough to categorize a defect. Often, however, we are left only with a determination that the field is abnormal in some poorly defined way. The next step in testing is to define some characteristics of the defect that will allow some diagnostic and therapeutic decisions to be made. Does it align itself along the vertical meridian, giving a hemianopic character? Does the central defect connect with the blindspot in a centrocecal pattern? Does the defect have an edge along the nasal horizontal meridian suggesting a nasal step? Does it have sharp edges or sloping margins? A computer-assisted perimeter must be told by its program that these questions must be answered for a diagnostic categorization. These questions are not part of the screening strategy, and represent a definite change in direction away from screening toward what the physician needs to know about this abnormal field. The next change in direction for strategy occurs when we ask the automated perimeter to quantify the defects it has found. In other words, to develop enough information about the size and depth of the defect so we can measure it again in the future to tell if it has changed. This is an extremely important feature in following glaucoma patients. It is helpful to note that in quantitative perimetry we do not ask to know everything about the defects, for such an exercise would take forever, but merely enough so we can confidently assess change in the future. Currently, there is considerable literature about the abilities of many automated perimeters to perform screening to acceptable standards. There is somewhat less literature to indicate that some automated perimeters can make reliable diagnostic decisions possible for the clinician. The literature on automated quantitative perimetry involves only a few devices. There is much 352
clinical validation work to be done before we can be confident that automated perimetry can replace superb manual perimetry in following the difficult patient. There is no theoretical reason why it cannot; the problem lies in developing the computer software programs to "think" like a good visual field technician. And that is not as easy as it sounds. PRINCIPLES OF SANE PERIMETRY Over the years, several rules have emerged which have made perimetry reproducible and reliable in clinical practice (Table 3). Just because those rules were made for the tangent screen and Goldmann perimeter does not mean that we can discard them in automated perimetry. In fact, they are probably even more important. Table 3. Principles of sane perimetry. Provide corrective lens for testing within the central 30 degrees Present stimuli randomly Calibrate instrument for each test Retest all missed stimuli
The most commonly neglected of these rules is to provide a corrective lens for testing within the central 30 degrees of the visual field. It is amazing how easy it is to create artifactual defects in the visual field by failing to provide a corrective lens and, worse, how easy it is to miss significant defects. An emmetropic patient without a corrective lens in a 30 cm radius bowl perimeter needs to provide 3.3 diopters of accommodation (without benefit of an accommodative target) for the 20, 30, or 60 minutes the test will take. Except for the youngest of patients, this is not going to occur, the stimuli will blur, and the results will be suboptimal. Even a tangent screen (100 cm distance) and an Octopus (50 cm distance) need some provision for corrective lenses in the majority of patients. If you learn nothing else from this article, remember the message about corrective lenses. Please instruct your technicians to use corrective lenses for visual field testing within the central 30 degrees on all patients. A reasonably recent refraction plus the recommended "add" from the manual that accompanies the perimeter are placed in the lens holder using cheap narrowrimmed trial lenses. It takes little time to do this, but it pays dividends in the enhanced quality of your visual fields. Presentations of stimuli at random locations is important to make perimetry reliable. If a patient knows where the stimulus will appear, he or she is likely to peek toward it or increase attentiveness to it. Such a requirement is more reliably met with automated than with manual units.
AM INTRA-OCULAR IMPLANT SOC
J-
VOL 10, SUMMER 1984
Calibration of the perimeter must be possible so that test conditions do not vary from patient to patient or from session to session. Most automated perimeters do this automatically, but it is important to know if this is being done. Finally, patient errors such as failing to push the response button in time and temporary lapses in concentration occur. These errors lead to missing a stimulus that should have been seen. Therefore, all missed spots should be retested to be sure that the stimulus is
not seen. Otherwise, false positive rates in any kind of perimetry are unacceptably high. Most technicians do this manually as a matter of course, but automated perimeters must be programmed to incorporate this feature, especially during suprathreshold testing. In Part II, we will examine some of the available devices for automated perimetry, some currently available strategies for visual field examination, and ways to interpret automated visual fields from your own or other practitioners' instruments.
AM INTRA-OCULAR IMPLANT SOC J-VOL 10, SUMMER 1984
353
A revolution in perimetry is occurring. This revolution is no less dramatic than the one which has occurred in cataract surgery, and it is developing along parallel lines. The use of intraocular lenses forced a reanalysis of extracapsular extraction, which made lens placement technically easier and seemed to produce fewer complications than intracapsular extraction. Similarly, the automated perimeter is forcing a reawakening of interest in the static method of exploring the visual field. It is technically easier to computerize than the kinetic method and is more sensitive to field defects. KINETIC PERIMETRY The tangent screen and the Goldmann perimeter are the time-honored devices for testing visual fields. Stimuli of various intensities are moved from the non seeing periphery toward the center, and the patient responds when he or she detects the stimulus. The border line between seeing and non seeing can be drawn by connecting the points derived from a single stimulus strength, the familiar isopter of kinetic perimetry. If the chosen stimulus is quite weak, the isopter may form a circle around fixation inside the physiologic blind spot. If the stimulus is strong, the maximum field area is outlined by the isopter at the Goldmann perimeter, whereas the limits of testing at the tangent screen are exceeded. An array of isopters can be plotted on a visual field chart and all ophthalmologists feel comfortable interpreting such charts. Unfortunately, however, kinetic perimetry suffers from four principal disadvantages that limit its usefulness. To explain these disadvantages requires reference to the old Traquair analogy of the island of vision in a sea of blindness (Figure 1). The highest points on the island represent good sensitivity, at which weak stimuli can be detected. Down near the waterline, only the strongest stimuli can be seen, and sensitivity is poor. The hill is of course enshrouded in fog (Figure 2) and our job is to map it.
- -<::::::: :::;::
:::: ::: ::::::::::::
:::::
-~:;;;::::::::::::::
Fig. 1.
(Mills) Traquair's island of vision in a sea of blindness. The highest point, representing the fovea, has the greatest sensitivity to test stimuli. At the water level the sensitivity is zero and no stimulus can be seen. The slope of the normal island of vision varies, being steepest close to the fovea and at the edge of the island, and more gentle in the intermediate sections.
Fig. 2.
(Mills) The problem in clinical perimetry is that the island of vision is enshrouded in a fogbank and we must use indirect methods of mapping it. The peak is exposed, since we know something about it from visual acuity determinations, but the rest of the island is uncharted.
Reprint requests to Richard P. Mills, M. D., Department of Ophthalmology,
RJ -10, University ofWashington,
AM INTRA-OCULAR IMPLANT SOC J-VOL 10, SUMMER 1984
Seattle, Washington 98195. 347
In kinetic perimetry, a stimulus of known strength is moved centrally, like an airplane flying toward the island at known altitude. The airplane "stimulus" will crash when it hits the island. After a number of airplane stimuli have crashed at one altitude, we can draw an isopter (Figure 3). After drawing several isopters, we will have a topographic map of the island of vision.
~llil ~
Fig. 3.
flll'll'"
",7, "" "'" \ \,~
(Mills) One difficulty with kinetic perimetry is that depressions of the interior of the island of vision are easy to miss. The airplanes will all crash on the periphery of the island or fly over the depression to crash centrally. Only if we are lucky enough to start an airplane in the center of the depression will we discover its presence and be able to map it.
Fig. 5.
(Mills) Another difficulty with kinetic perimetry is that flat sections of the island are difficult to map. Airplanes flying toward the island at one altitude may crash centrally, while those flying at a slightly different altitude will crash peripherally. Moreover, undulations of the flat section will be undiscovered.
(Mills) One method of mapping the island is to fly many airplanes toward it at the same altitude. We could record where they crashed, and knowing the altitude, we could draw an isopter. After drawing several isopters we would have an idea of the shape of the hill of vision. The same principle is used in kinetic perimetry as practiced on the Goldmann instrument or tangent screen. Stimuli are moved centrally until seen, and all points discovered by the same stimulus are connected by an isopter line.
The first of the four disadvantages of kinetic perimetry is that hollowed-out valleys in the interior of the island of vision are difficult to detect (Figure 4). Stimuli (airplanes) coming from the outside will miss them. We could start the stimulus within the valley, of course, if we knew where to start. Since we do not, we would be forced to select places to start the stimulus randomly, hoping we would start inside a valley (scotoma). The chance of missing isolated valleys (scotomas) is high, especially with small or shallow ones. The second disadvantage is that kinetic perimetry does a poor job of mapping flat sections of the island of vision (Figure 5). An airplane stimulus coming from outside at one altitude may crash quite a distance away from one coming at a slightly different altitude. Undulations of the flat section, which may be quite important, will be unmapped. Most patients have at least some relatively flat areas in their visual island, like a plateau, but in some patients the entire island resembles a mesa. In such patients, kinetic perimetry tends not to be a very productive exercise. Third, kinetic perimetry by definition uses moving stimuli. It has long been known that moving stimuli are easier to detect than stimuli that are standing still, 348
Fig. 4.
especially within defective areas of the visual field. This is known as statokinetic dissociation, and contrary to popular belief, the phenomenon occurs with lesions at all levels of the visual pathways, though it may be most prominent in occipital lobe disease. The fact that the stimulus is moving may cause a visual field defect to be missed. Finally, kinetic perimetry has proven very difficult to automate. For example, the Coherent Perimetron had difficulty mapping scotomas, particularly those which "broke through" to the periphery. Inordinate amounts of time were consumed as the computer seemed to get lost inside the scotoma. Production of the instrument was subsequently discontinued. Thus, no competitively priced automated kinetic perimeter is currently available commercially.
AM INTRA-OCULAR IMPLANT SOC J-VOL 10, SUMMER 1984
Before the advent of automated perimeters, Armaly developed and Drance modified a method of screening for glaucomatous field defects which lessened the effect of the first three disadvantages of kinetic perimetry. They did so by including in the test strategy a series of stimuli presented statically within the central visual field. So the Armaly-Drance method of glaucoma screening is actually a mixture of kinetic and static perimetry. STATIC PERIMETRY Let us return to the island-of-vision analogy to explain how static perimetry works. We could drop parachutes with altimeters onto the island and record the height at which each one landed (Figure 6). Thus, at a number of discrete locations in space, we would know the height (sensitivity). This method is known as threshold static perimetry since it determines the threshold between seeing and nonseeing at a number of preselected locations .
45' o-~----f§~-+------22 '
(
~----~~~;;;--'.:--35· Ifl lilit / I I i i I \ \ \ \ , \ \ \' \, \
Fig. 6. (Mills) Another me thod of mapping the island of vision is to drop parachutes at known coordinates toward the island, recording the height at which they land. The principle is the same for static threshold determinations in clinical perimetry. While time consuming, we know all there is to know about each point and can map the island if we test enough points.
Observe that threshold static perimetry overcomes the disadvantages of kinetic perimetry. Isolated valleys (scotomas) in the island of vision are less likely to be missed if one uses a large enough number of parachutes (stimuli). Flat areas of the island are accurately defined, and undulations unmasked. The stimulus is not moving, and the process is easy to computerize. Unfortunately, threshold determinations are very time consuming, and long testing sessions are required to develop adequate point density. The reluctance of ophthalmologists and patients to commit to long testing sessions is a problem, but this becomes even worse if screening of significant patient numbers is required. In an effort to save testing time but retain some of the advantages of static perimetry, suprathreshold static testing methods were developed. Helicopters are employed to fly to certain locations over the island of
Fig. 7.
(M ills) Anothe r m ethod of mapping the island of vision is to employ helicopte rs to fly to known coordinates and altitude and report simply whether they are in the air or have lande d . The same principle applies for suprathreshold testing in clinical perimetry. A large number of points can be tested in a short time, but we know only whether each point is seen or not seen. With a sufficient variety of stimulus strengths, a map of the island of vision can be generated.
vision and hover at a specified altitude (Figure 7). They radio whether they are still in the air or whether they have landed. A large number of points in space at different altitudes can be tested in relatively short order. If we select our points well, we can detect valleys (scotomas) in the interior of the island. We will still have trouble defining flat sections of the island (field) and determining undulations. But we are not moving the stimulus, and automation of the process is easy. One real problem with suprathreshold static perimetry is that it is easy to waste time testing with stimulus strengths that are either way above threshold or way below it. Consider the example of a single stimulus strength tested at about 150 locations, a simple strategy employed by many automated perimeters. Toward the periphery of the field, the stimulus is far too weak to be seen, and toward the center it is too strong. If our objective is to detect subtle scotomas, we will find them only in areas where the stimulus is presented close to the threshold between seeing and nonseeing. We can get around this problem somewhat by testing these 150 locations at multiple stimulus strengths. In other words, use several "runs" with the automated perimeter, at the cost of added test time (and more visual field charts to shuflle). If we only knew what the expected threshold between seeing and nonseeing was at any point on the island, we could select our stimulus strength close to the expected threshold. This would allow us to detect scotomas across the whole field because we would be using different stimulus strengths at different places in the field: weaker toward the center and stronger toward the outside. To accomplish this, we could use a threshold static method at a few points in the field, determining the actual threshold at those points. We could then select
AM INTRA-OCULAR IMPLANT SOC J-VOL 10, SUMMER 1984
349
stimuli for 'our suprathreshold test related to (interpolated from) those few threshold determinations. Such a strategy is called threshold-related suprathreshold static perimetry. Other threshold-related strategies for suprathreshold static perimetry select stimuli based on a "normal" visual field (derived from large populations and corrected for patient age) or the "normal" gradual decrease in sensitivity from the foveal threshold at more peripheral visual field locations. Obviously, with the many types of threshold-related strategies, it is important to inquire which type was used before interpretation is attempted. In summary, three types of static perimetry are commonly used. Threshold static perimetry determines the threshold stimulus strength at specified locations in the field. Suprathreshold static perimetry presents a preselected stimulus at specified locations and determines whether it is seen or not. Thresholdrelated supra threshold static perimetry presents stimuli at specified locations, with stimulus strength chosen on the basis of a few known threshold points. THE UNCERTAINTY PRINCIPLE IN PERIMETRY Up to this point, we have been talking about perimetry as though the threshold of a point on the island of vision is an absolute number. Stimuli stronger than threshold would always be seen and stimuli weaker than threshold would never be seen. U nfortunately, this is not the case. Threshold is actually a relative number, referring to the level at which a patient will see 50% of the stimuli presented. A stimulus stronger than threshold increases the chance of seeing the stimulus and a stimulus weaker than threshold decreases the chance of seeing it (Figure 8). This uncertainty is common in biologic systems. For example, recall the LD50 concept from toxicology, a dose level at which half of the animals receiving a drug die from its toxic effects. The uncertainty principle plays a role no matter which type of perimetry is being performed, but it is most measurable by threshold static testing. We can measure threshold at a single point over and over again, and obtain different numbers within a range. Presumably the "true" threshold lies somewhere near the middle of that range. The uncertainty is magnified by lapses in patient attentiveness caused by distractions, fatigue, or disease. It is also magnified in areas of defective visual field. Some evidence is accumulating that the earliest sign of glaucomatous field loss is a local widening of the zone of uncertainty. Stated differently, this early sign is represented by an increased local fluctuation in multiple threshold determinations. This may happen while the average measured threshold is still normal. 350
Q)
U
c
0
..r:.
-
U c
Q)
u
.... Q) a..
VI :;) :;)
E
100
(/)
0>
c
'(ij Q)
(/)
--0
0 50
55
59
Brightness of Stimulus Fig. 8.
(Mills) The uncertainty principle in perimetry. As a stimulus becomes brighter, the chance that it will be seen by the patient increases. The 50% level is called the threshold. In this example, a stimulus of brightness 55 is seen half the time and not seen half the time. Brightness 55 thus represents the threshold for that point in this patient. However, in attempting to measure threshold, a stimulus of 50 will be seen 5% of the time and a stimulus of 58 will be missed 5% of the time. Multiple trials could establish a mean threshold, but time expenditure would be excessive if we tested many points.
For the clinician, the uncertainty of the visual field measurements represents a major source offrustration. Often there is a temptation to introduce shortcuts in the testing process, on the grounds that field testing is inexact at best. We should remember, however, that such shortcuts usually result in an increased uncertainty to the point that the field chart is almost impossible to interpret with any degree of confidence. STIMULUS v BACKGROUND INTENSITY
A background light level is used in perimetry so that the whole retina can remain in a light-adapted state. Otherwise a stimulus would produce local light adaptation which would take many minutes to return to baseline, preventing repeat testing at that location. The Goldmann perimeter is calibrated for 31.5 apostilbs (as b) background illumination, as are most of the currently available automated units. Such a level was adopted as a standard by the International Council of Ophthalmology. However, the Oculus (Tubingen) uses 10 asb, the Octopus 4 asb, and the Rodenstock Peritest 3.1 asb as a background. These instruments seem to function well in practice, since such background levels are still within the lower end of the photopic range. A stimulus is seen when it is sufficiently brighter than the background to enable detection. It is the difference in brightness between stimulus and background that actually elicits a patient response. (The smallest difference that can be appreciated 50% of the time is the threshold). There are physical factors in
AM INTRA-OCULAR IMPLANT SOC J-VOL 10, SUMMER 1984
bulbs and light emitting diodes which limit the maximum stimulus intensity that can be presented. Thus, there is a usable testing range for each instrument, limited on the lower end by the background intensity and on the upper end by the physical limits of stimulus brightness.
Table 1. Decibels (db) and apostilbs (asb). Stimulus Intensity (asb)
Decibel Value (db)
10,000 (maximum)
0
1,000
10
100
20
APOSTILBS, LOG UNITS, AND DECIBELS
32
25
Why not just provide everyone with Goldmann equivalences for stimulus intensity on different devices and forget about apostilbs, log units, and decibels? Most of us understand what a Goldmann I-2-e represents, so why change the terminology? The reason is simple. The Goldmann stimuli we understand are used mostly in a kinetic mode, but the new automated devices are static perimeters. There is poor correlation between Goldmann stimulus numbers and automated stimulus numbers across patient populations. Moreover, the Goldmann perimeter uses size as well as brightness change to vary the stimulus, while many automated units do not. Size and brightness are related in a general sense, but not precisely enough to allow use of Goldmann terminology in automated perimetry. An apostilb is a measure of light brightness, in a physical sense. The eye, however, perceives light within a moderate range of intensity on a basis that approximates a logarithmic scale. That is, an illumination change of a stimulus from 10 to 100 asb gives roughly the same change in sensation as an illumination change from 100 to 1000 asb. Because of variability in such a subjective grading of brightness, the relationship does not hold exactly for all patients, but it is a useful approximation. Therefore, stimulus intensities can be expressed in log units and approximate the way the eye actually sees the stimulus. Because the stimulus can be made infinitesimally dim, starting the log scale at the lower end is impractical. Consequently, the maximum intensity an instrument can deliver is set at 0 log units, and dimmer intensities are expressed as the number of log units less than the brightest possible stimulus. The higher the number, the dimmer the stimulus. We all remember the decibel from audiology as a logarithmic function of sound intensity. It was "invented" because the ear responds to sound in a logarithmic way, just as the eye responds to light. A decibel is nothing more nor less than one tenth of the log unit. In perimetry, the decibel is used to describe brightness in logarithmic terms. Like the log unit, it is set at zero for the maximum stimulus intensity the instrument can deliver, and the number of decibels refers to a reduction from peak intensity. For example, if the maximum stimulus intensity is 10,000 asb, and the chosen stimulus is 1,000 asb, the decibel number is 10. If the maximum intensity is 10,000 asb, and the
10
30
1
40
chosen stimulus is 100 asb, the decibel number is 20, and so forth (Table 1). Observe that since higher numbers in log units and decibels refer to dimmer stimuli, whe n printed on a visual field chart, a higher number means a greater retinal sensitivity. If we are thinking of the island of vision, a higher number (log unit or decibel) means a higher altitude on the island (Figure 7). Likewise, a zero reading means that we are at the waterlinewithin the sea of blindness. The "new" terminology is fairly easy to get used to, given a little practice and a willingness to abandon the " Gold" standard.
SCREENING, DIAGNOSTIC, AND QUANTITATIVE FIELDS In the days of tangent screens, Goldmann perimeters, and human examiners, we did perimetry by finding defects, characterizing them well enough to fit them in a diagnostic category, then quantifying the defects for later comparison. The process of doing a field was viewed as a continuum. It is not, as any automated perimeter will tell you. The functions of screening, making a diagnosis, and quantitating defects require quite different strategies of testing (Table 2). A human examiner can make the strategic transitions easily, but we have found that computers need to be told when to change their strategies to accomplish the different functions required of them. In screening, the object is to tell if a visual field is abnormal or normal, and nothing else. Does this patient with high intraocular pressure have a field defect or not? We demand from test strategies used in screening that they be sensitive enough to detect most of the Table 2. Screening, diagnostic, and quantitative fields. Visual Field Task
Question to Answer
Screening
Is there a d efect or not?
Diagnosis
What sort of defect is it?
Quantitative
What does the defect look like?
AM INTRA-OCULAR IMPLANT SOC J-VOL 10, SUMMER 1984
351
defects, yet selective enough so a lot of false positives do not appear. It helps a great deal if we know what sort of defect is expected. Screening in patients with ocular hypertension should look for nasal steps, paracentral scotomas, Bjerrum scotomas, and temporal wedge defects. Screening in a stroke patient should look carefully for hemianopic defects. Screening in a patient with "bone spicules" of pigment in the fundus should concentrate on ring scotomas. Reliable screening strategies produce detection rates of90% or better with false alarm rates below 10%, and are available on most automated perimeters. Incidentally, rates for manual perimetry as practiced in the community are no better, and are often worse. Unfortunately, perfection is impossible to achieve. If one tries too hard to improve detection rate, false alarms become a major problem. If one tries too hard to minimize false alarms, then detection rate falls. A screening visual field does contain diagnostically useful information, which is sometimes enough to categorize a defect. Often, however, we are left only with a determination that the field is abnormal in some poorly defined way. The next step in testing is to define some characteristics of the defect that will allow some diagnostic and therapeutic decisions to be made. Does it align itself along the vertical meridian, giving a hemianopic character? Does the central defect connect with the blindspot in a centrocecal pattern? Does the defect have an edge along the nasal horizontal meridian suggesting a nasal step? Does it have sharp edges or sloping margins? A computer-assisted perimeter must be told by its program that these questions must be answered for a diagnostic categorization. These questions are not part of the screening strategy, and represent a definite change in direction away from screening toward what the physician needs to know about this abnormal field. The next change in direction for strategy occurs when we ask the automated perimeter to quantify the defects it has found. In other words, to develop enough information about the size and depth of the defect so we can measure it again in the future to tell if it has changed. This is an extremely important feature in following glaucoma patients. It is helpful to note that in quantitative perimetry we do not ask to know everything about the defects, for such an exercise would take forever, but merely enough so we can confidently assess change in the future. Currently, there is considerable literature about the abilities of many automated perimeters to perform screening to acceptable standards. There is somewhat less literature to indicate that some automated perimeters can make reliable diagnostic decisions possible for the clinician. The literature on automated quantitative perimetry involves only a few devices. There is much 352
clinical validation work to be done before we can be confident that automated perimetry can replace superb manual perimetry in following the difficult patient. There is no theoretical reason why it cannot; the problem lies in developing the computer software programs to "think" like a good visual field technician. And that is not as easy as it sounds. PRINCIPLES OF SANE PERIMETRY Over the years, several rules have emerged which have made perimetry reproducible and reliable in clinical practice (Table 3). Just because those rules were made for the tangent screen and Goldmann perimeter does not mean that we can discard them in automated perimetry. In fact, they are probably even more important. Table 3. Principles of sane perimetry. Provide corrective lens for testing within the central 30 degrees Present stimuli randomly Calibrate instrument for each test Retest all missed stimuli
The most commonly neglected of these rules is to provide a corrective lens for testing within the central 30 degrees of the visual field. It is amazing how easy it is to create artifactual defects in the visual field by failing to provide a corrective lens and, worse, how easy it is to miss significant defects. An emmetropic patient without a corrective lens in a 30 cm radius bowl perimeter needs to provide 3.3 diopters of accommodation (without benefit of an accommodative target) for the 20, 30, or 60 minutes the test will take. Except for the youngest of patients, this is not going to occur, the stimuli will blur, and the results will be suboptimal. Even a tangent screen (100 cm distance) and an Octopus (50 cm distance) need some provision for corrective lenses in the majority of patients. If you learn nothing else from this article, remember the message about corrective lenses. Please instruct your technicians to use corrective lenses for visual field testing within the central 30 degrees on all patients. A reasonably recent refraction plus the recommended "add" from the manual that accompanies the perimeter are placed in the lens holder using cheap narrowrimmed trial lenses. It takes little time to do this, but it pays dividends in the enhanced quality of your visual fields. Presentations of stimuli at random locations is important to make perimetry reliable. If a patient knows where the stimulus will appear, he or she is likely to peek toward it or increase attentiveness to it. Such a requirement is more reliably met with automated than with manual units.
AM INTRA-OCULAR IMPLANT SOC
J-
VOL 10, SUMMER 1984
Calibration of the perimeter must be possible so that test conditions do not vary from patient to patient or from session to session. Most automated perimeters do this automatically, but it is important to know if this is being done. Finally, patient errors such as failing to push the response button in time and temporary lapses in concentration occur. These errors lead to missing a stimulus that should have been seen. Therefore, all missed spots should be retested to be sure that the stimulus is
not seen. Otherwise, false positive rates in any kind of perimetry are unacceptably high. Most technicians do this manually as a matter of course, but automated perimeters must be programmed to incorporate this feature, especially during suprathreshold testing. In Part II, we will examine some of the available devices for automated perimetry, some currently available strategies for visual field examination, and ways to interpret automated visual fields from your own or other practitioners' instruments.
AM INTRA-OCULAR IMPLANT SOC J-VOL 10, SUMMER 1984
353