Novel 3D computerized threshold Amsler grid test

NOVEL 3D COMPUTERIZED THRESHOLD AMSLER GRID TEST

WOLFGANG FINK1,2 and ALFREDO A. SADUN2

1California Institute of Technology, Pasadena; 2Doheny Eye Institute, Keck School of Medicine, University of Southern California, Los Angeles; CA, USA

Abstract

Purpose: To introduce a novel 3D computerized threshold Amsler grid test, developed by Fink and Sadun, that allows for a novel characterization of the structure of visual field defects in three dimensions (see http://www.wfbabcom5.com/wf335.htm). Methods: With one eye covered, patients are placed in front of a computer monitor displaying an Amsler grid. While focusing on a varying central fixation marker, the patients trace the areas, on a touch sensitive screen overlaying the displayed grid, that are missing from their field of vision, with their finger. The procedure is repeated with varying degrees of grid contrast. The results are recorded and later displayed as topographical contour rings, in a 3D depiction of the central hill-of-vision. Results: Several clinical pilot studies have been conducted at the Doheny Eye Institute and over 200 patients have been examined so far. Conditions such as optic neuritis, anterior ischemic optic neuropathy, age-related macular degeneration, glaucoma, and ocular hypertension have been examined with the 3D visual field test. The tested visual field is limited to the central 20-25 degrees (depending on the computer monitor size). The physiological blind spot cannot be detected with this test due to the cortical ‘fill-in phenomenon’. Conclusions: The 3D computerized threshold Amsler grid test is an innovative and fast (four to five minutes per eye) visual field test. It provides several novel features, including: a. additional information through immediate 3D rather than 2D depiction of scotomas, such as location, extent, slope, depth, and shape; b. simple test-setup; and c. good patient compliance. In light of results from pilot studies, the 3D visual field test appears to have the potential for the early detection and monitoring of various diseases, in particular glaucoma and macular degeneration.

Introduction

Visual field testing (perimetry) has always been an important part of an ophthalmological evaluation. In 1947 Amsler introduced a suprathreshold grid for evaluating the central ten degrees of the visual field.1,2 The high contrast of the standard Amsler test may fail to detect subtle field defects, such as relative scotomas.

In 1986, threshold Amsler grid testing was introduced by Wall and Sadun.3 This test utilizes near threshold, rather than suprathreshold, visual stimuli. Contrast be-

Address for correspondence: Wolfgang Fink, PhD, Visiting Associate in Physics at Caltech, California Institute of Technology, Mail Code 106-38, Pasadena, CA 91125, USA. Email: wfink@krl.caltech.edu

Perimetry Update 2002/2003, pp. 207–212

Proceedings of the XVth International Perimetric Society Meeting, Stratford-upon-Avon, England, June 26–29, 2002

edited by David B. Henson and Michael Wall

© 2004 Kugler Publications, The Hague, The Netherlands

tween the target and the background was controlled with cross-polarizing filters. With this modification, it was possible to detect subtle scotoma and relative visual field defects.

Taking this a step further, Fink and Sadun computerized threshold Amsler grid testing in 2000.4,5 The examination results were used to generate a three-dimensional (3D) map of the central visual field that illustrates the location, extent, slope, depth, and shape of visual field defects.

The 3D computerized threshold Amsler grid test (see Fig. 1) is the outcome of close collaboration over the past three years between the California Institute of Technology and the Doheny Eye Institute, Keck School of Medicine at the University of Southern California (for further details see http://www.wfbabcom5.com/wf335.htm).

Examination method

The 3D computerized threshold Amsler grid test uses an IBM compatible PC with a 17inch monitor and touch sensitive screen (Fig. 1). Each patient is positioned 30 cm in front of the computer monitor. An eye-patch is used to cover the eye that is not being examined. Refractive correction is used with the patient’s contact lenses or spectacles when necessary.

Fig. 1. The 3D computerized threshold Amsler grid test, using a computer monitor with touch sensitive screen.

Fig. 2. Screenshot of Amsler grid at a mid grayscale level with changing central fixation marker and filled in area of scotoma.

An Amsler grid, at a preselected contrast level, is displayed on the monitor (Fig. 2). The patient is first asked to focus on a changing stimulus at the center of the grid. In order to suppress the central Troxler effect and keep the patient’s attention, the stimulus is regularly changed (changing set of characters such as letters and numbers; see Fig. 2). The patient is asked to mark the areas, on the Amsler grid, that are missing from his or her field of vision by tracing this region with their finger on the touch screen. Areas that are ‘missing’ are defined as areas where the grid lines are missing, bent, wavy or distorted, or areas where the grid contrast differs from the rest of the grid. The test is repeated with varying degrees of contrast. The results are recorded for later conversion into 3D format at the end of the test (Fig. 3). The third dimension being screen contrast.

Results

Each eye, depending on the patient compliance, required a total of approximately four to five minutes to be tested.

Examples from several ophthalmic disorders are depicted in Figure 3. In Figure 3a, the 3D plot exhibits a more scalloped-shape visual field defect in optic neuropathy, with islands of relative sensitivity. This is in contrast to a clean-cut division between vision and no vision in a case of anterior ischemic optic neuropathy (Fig. 3b) (see also Fink et al.5). This feature may have potential for differential diagnosis.

Figures 3c and 3d depict typical cases of age-related macular degeneration (see also

Fig. 3. 3D display of visual field with a diagnosis of: a. optic neuropathy; b. anterior ischemic optic neuropathy; c. age-related macular degeneration; d. glaucoma; and e. unimpaired central hill-of-vision; recorded by the 3D computerized threshold Amsler grid test. The x/y-axis denotes the horizontal/vertical coordinates of the visual field in degrees with (0, 0) being the center of fixation. The z-axis denotes the screen contrast expressed as a percentage.

Nazemi et al.6) and advanced glaucoma (see also Fahimi et al.7). The 3D representation of visual field defects caused by these two diseases is both intuitive (macular degeneration being a ‘central hole’ and advanced glaucoma being a ‘confined mesa’) and informative. The 3D representation can be further characterized by a slope along the circumference of the defects. Figure 3e shows a normal central hill-of-vision at 100% contrast, for the central 20 degrees.

The 3D computerized threshold Amsler grid test has been in use since 2000 at the Doheny Eye Institute, Keck School of Medicine at the University of Southern California. Over 200 patients have been examined or screened for glaucoma, age-related macular degeneration, optic neuritis, anterior ischemic optic neuropathy, and ocular hypertension (see also Nazemi et al.8). In these pilot studies, the 3D test has proven to be reliable, fast, and accurate.5-8

Subtle scotoma, hard to detect with standard automated perimetry (mainly because of limited spatial resolution, e.g., light stimulus every six degrees both horizontally and vertically), were repeatedly identified by the 3D test,8 suggesting a potential role as a screening instrument.

The physiological blind spot was not detected with this test because of cortical ‘fillin phenomenon’ for a steady grid, which is a known limitation of all Amsler gridbased tests.

Further (larger scale) clinical studies are needed to corroborate and extend the initial findings gathered with the 3D computerized threshold Amsler grid test.

Conclusions

The 3D computerized threshold Amsler grid test is a novel approach to visual field testing that provides several additional features over conventional perimetry. The test delivers 3D rather than 2D depictions of scotomas (the z-axis reflecting contrast sensitivity) and provides insight into visual field defects, including shape and slope information about the scotoma boundaries (see Fig. 3). This test provides characterization of the 3D structure of scotoma, thus laying the foundation for monitoring the development of scotoma profiles. The 3D computerized threshold Amsler grid test only requires a touch sensitive screen, the test software, and an off-the-shelf computer system. The new test is simple to use and tests are performed quickly (four to five minutes per eye), making frequent testing feasible.9

References

1.Amsler M: L’examen qualitatif de la fonction maculaire. Ophthalmologica 144:248-261, 1947

2.Amsler M: Earliest symptoms of diseases of the macula. Br J Ophthalmol 37:521−537, 1955

3.Wall M, Sadun AA: Threshold Amsler grid testing: cross-polarizing lenses enhance yield. Arch Ophthalmol 104(4):520−523, 1986

4.For further information on the 3-D Computer-Automated Threshold Amsler Grid Test: http:// www.jpl.nasa.gov/releases/2001/release_2001_215.html and http://www.wfbabcom5.com/wf335.htm

5.Fink W, Hsieh AK, Sadun AA: Computer-automated 3-D visual field testing in distinguishing paracentral scotomas of Optic Neuritis versus AION. (ARVO Abstract No. 1643). Invest Ophthalmol Vis Sci 41(4):S311, 2000

6.Nazemi PP, Fink W, Lim JI, Sadun AA: Paracentral scotomas of age-related macular degeneration detected by means of a novel computer-automated 3-D visual field test. (ARVO Abstract No. 3799) Invest Ophthalmol Vis Sci 42(4):S705, 2001

7.Fahimi A, Sadun AA, Fink W: Computer automated 3D visual field testing of scotomas in glaucoma. (ARVO Abstract No. 796). Invest Ophthalmol Vis Sci 42(4):S149, 2001

8.Nazemi PP, Fink W, Sadun AA, Minckler D, Francis B: Early detection of glaucoma by means of a novel computer-automated 3-D visual field test. Poster and Abstract, American Academy of Ophthalmology Meeting in New Orleans 2001. Scientific Poster 33, p 159 (http:/www.scientificposters. com/aao/viewabstract.cfm?id=1&CFID=32449&(FTOKEN=99797202)

9.Fink W, Sadun AA: Prospects for autonomous visual field testing on space missions. Lecture and Poster, NanoSpace 2001, Exploring Interdisciplinary Frontiers, The International Conference on Integrated Nano/Microtechnology for Space and Biomedical Applications, March 13th-16th, 2001, Houston, TX. Abstract, Conference Program, p 102 (http://www.wFbabcom5.com/ftp/nanospace200/ abstract.txt)

CLOSING PERIMETRY’S SENSITIVITY GAP: A RAREBIT APPROACH

LARS FRISÉN

Institute of Clinical Neuroscience, University of Göteborg, Sweden

Abstract

The primary task of clinical perimetry is to identify and grade neurovisual system damage. Clinicopathological and experimental studies have highlighted a wanting sensitivity to low-to-moderate degrees of damage. A possible explanation is that perimetric test targets carry an excess of information, both in space and time. Another possible explanation relates to the dependence on external, empirical references for normality: internal references should produce tighter limits. A new visual field test was devised under these premises. The test depended on rarebit test targets, in the form of briefly exposed microdots of high contrast, and the expectation that close to 100% of such targets should be visible normally. A large number of circumscribed test areas were probed repeatedly, with ever-new microdot positions. On average, normal subjects responded to 95 ± 3% of probes. Patients with known visual field defects from a variety of causes missed larger numbers of probes within the defective areas, and the deeper the defects, the larger the number of misses. Detailed comparisons were made with high-pass resolution perimetry (HRP) in ten patients with minor mid-chiasmal lesions. On average, microdot perimetry revealed twice as extensive defects (p = 0.002), indicating superior sensitivity to low-degree neurovisual damage.

Introduction

Given that retinal receptive fields sample visual images discretely, it would appear that the most direct way to identify damage to receptive fields, or their upstream connections, would be to sample receptive field function discretely, with test targets adapted to receptive field sizes. Current perimeters use much larger targets (Fig. 1), which may explain why their capacity to reveal low-degree damage is quite limited. Receptive field-size targets might perform better, but numerous obstacles can be listed. These include the difficulty of generating targets that are small enough, optical imperfections of the eye, the large variety of receptive field types, and interference from eye movements. However, unaware of these difficulties, normal people can actually see minuscule images in their peripheral vision, e.g., the stars in the night sky. Hence, small-target probing of the neuroretinal matrix may actually be both practicable and useful.

This report provides a brief overview of the implementation of microdot (‘Rarebit’)

Address for correspondence: Lars Frisén, Neuro-oftalmologi Blå 7, SU/S, SE-413 45 Göteborg, Sweden. Fax ++ 46 31 82 01 43. Email: lars.frisen@neuro.gu.se

Perimetry Update 2002/2003, pp. 213–219

Proceedings of the XVth International Perimetric Society Meeting, Stratford-upon-Avon, England, June 26–29, 2002

edited by David B. Henson and Michael Wall

© 2004 Kugler Publications, The Hague, The Netherlands

Fig. 1. A simple scheme to illuminate the relative dimensions of neural input substrate and various perimetric test targets. Not to scale. DLS: differential light sense.

perimetry and presents data that reflect its capacity to reveal low-degree neural damage. A full report will appear elsewhere.1

Outline of test principles

Target size

The ideal of literal point size is difficult or impossible to realize in clinical work. Instead, an acceptable upper size limit has to be found. One solution is to use existing data on minimum angles of resolution (MAR) at high contrast. Normal MAR thresholds typically range from 2-6 seconds within 5-30° of visual field eccentricity. However, even smaller targets are visible, if shown with sufficient contrast and separation, and so it was decided to try targets scaled as one-half of normal MARs.

Target generation

Computer graphics offers many advantages. Liquid crystal displays (LCD) offer better definition of single picture elements (pixels) than the ubiquitous cathode ray tubes. A commercial 15-inch diagonal LCD with a native resolution of 1280 x 1028 pixels was used. Pixel size was 0.239 mm. The screen was set to produce a background of 1 cd/m2. Maximum luminance was 160 cd/m2. These luminances were similar to those used in visual acuity testing. However, to maximize retinal image contrast, the current test used the reverse conditions, i.e., bright targets on a dark background.

In order to keep the targets’ subtended angles at specification, it was necessary to use two testing distances: 1 m for the paracentral field, and 0.5 m for the more peripheral areas.

Exposure time

Brief exposure times are required to negate effects of the eye’s perpetual micro movements. These movements do not normally interfere with vision. However, eyes with partially depleted neural matrices may benefit from movement by sweeping functional receptive fields across the target’s retinal image, and so cause an over-rating of performance. For this reason, exposure times should be short enough to ‘freeze’ the retinal image relative to the moving receptive field array, i.e., of the order of flashes. Unfortunately, currently available LCDs require tens of milliseconds to toggle maximum brightness. Here, 200 msec nominal presentation time was used.

Estimating functional status

The combination of perpetual eye movements and small-sized test targets works against repeated testing of one and the same retinal area. Hence, thresholding is not meaningful. Instead, the new test simply probed for the presence of vision, and instead of repeating probes in the same position, ever-new positions were selected. The anticipated outcome for normal eyes was all probes seen, with some allowance for the blind spot, angioscotomata, and blinks (hit rates ≈ 100%). Eyes with partially depleted neural matrices were expected to miss additional probes, and the larger the number of matrix defects, the larger the expected number of misses (hit rates << 100%).

Test procedure

Because probes must be directed to ever-new retinal areas, test locations were not defined as discrete points but as circles centered on discrete coordinates (Fig. 2, right panel). The circle diameter was set to five degrees and probes could appear anywhere inside this region, with the restraint that previously probed positions must not be re-visited. In the interest of saving on test time, probes were presented in pairs, with a minimum separation of four degrees. Hence, the test task was to signal perception of one probe, or two probes, or none. Responses were recorded by clicking a mouse button, once for one probe seen, twice for two. No click was to be made if no probe was seen.

Thirty circular test areas inside 30° of eccentricity were probed in a pseudo-random sequence. The fixation mark was moved across the screen so as to allow access to all test locations. First, one pass was made over the 26 non-central locations, at 0.5-m test distance. This required less than one minute of testing time. A brief pause was allowed before running the next pass. In all, five passes were made before changing the test distance to 1 m. The four central-most test areas were then traversed the same number of times, again in a pseudo-random sequence.

Blank and single-dot presentations were included for control purposes. The examiner, sitting to the side of the subject, monitored fixation. Auditory and visual feedback was provided automatically.

The sizes and numbers of test areas were selected from clinical experience, with the aim of meeting the clinical needs of spatial detail. The number of passes was not hardcoded, but was left to the examiner’s judgment, to suit the individual patient’s need and capacity. Accumulated results could be reviewed after each pass.

Fig. 2. Examples of HRP (left) and Rarebit perimetry (right) results from the same eyes. HRP plots resolution thresholds to scale. The upper pair of charts shows the right-eye results of a normal 52-year-old subject, which are well within normal limits in both tests. The bottom pair shows the left-eye results of a 54-year-old subject with moderate damage from open-angle glaucoma, particularly in the lower nasal quadrant. The central pair shows the results of the same subject’s right eye, with less visual field damage. Inset: decimal acuity.

Result presentation

Results were graphed using a nested circle format (Fig. 2, right panel). Each test location was identified by an outer, open circle representing the tested area. Any missed probes were represented by an inner, filled circle, whose diameter reflected the missed proportion of probes. Hence, a miss rate (100 – Hit Rate) of 70% was shown as a filled circle with a

70% diameter relative to the outer, reference circle. Further, various Hit Rate statistics were calculated.

Reference test

All subjects were examined by high-pass resolution perimetry (HRP), which is a welldocumented procedure that performs at least on par with conventional perimetry. HRP was always done before rarebit perimetry. Most subjects were naive to the former test, and all were naive to the latter.

Results

The test task proved to be easy to master and even somewhat entertaining. Twenty-seven normal subjects aged between 27 and 70 years obtained mean Hit Rates averaging 95 ± 3% (SD), excluding the test area closest to the blind spot. There was a slight decline with age, averaging 0.1% per year. Except for the vicinity of the blind-spot poles, missed presentations did not cluster in meaningful patterns. A typical example is shown in Figure 2 (top right panel). Test times averaged five minutes 14 seconds. The modal number of errors was 0. Reproducibility and tolerance to blur induced by optical over-correction were good. Learning effects were mostly small.

More than 100 patients have been examined so far. They were not selected on any statistically representative basis, but primarily on grounds of clinical or HRP evidence of minor neural damage from various, but mostly, neuro-ophthalmological causes. Cataract and ametropia exceeding three diopters were grounds for exclusion. All subjects with abnormal HRP results obtained abnormal Hit Rates. The spatial distribution of abnormal test areas consistently overlapped with those shown by HRP, but tended to spill over the HRP defect borders. Illustrative results in a patient with asymmetric glaucoma damage are shown in Figure 2 (lower panels).

Sensitivity to low-degree neural damage was studied in patients with mid-chiasmal lesions. These lesions constitute a useful model disorder for this type of evaluation for the reason that the associated field defects evolve in a predictable way, sloping fairly smoothly around the fixation axis. Hence, from their deepest portion in the upper temporal quadrant, defects taper out into the lower temporal quadrant, the lower nasal quadrant, or the upper nasal quadrant, depending on the severity of damage. Consequently, severity of damage can be defined as the number of quadrants involved. This number may be different for different techniques, depending on their ability to identify the shallowest part of the slope, i.e., their sensitivity to low-degree damage. Ten patients were examined. Decimal visual acuities ranged from 0.8-1.2, attesting to small degrees of neural damage. Deriving normal quadrant limits from the normal subjects described above, HRP and rarebit perimetry produced quite discordant results: in the right eyes, HRP identified a mean of 1.7 quadrants involved, rarebit perimetry 3.4 (p = 0.002). Figure 3 shows the number of quadrants involved for each technique. Left eye results were similar (not shown).

Fig. 3. Numbers of visual field quadrants falling outside normal quadrant limits in HRP (left bars) and rarebit perimetry (right), in the right-eye results of ten patients with light-to-moderate mid-chiasmal compression. UT: upper temporal; LT: lower temporal; LN: lower nasal; UN: upper nasal.

Discussion

Although microdot perimetry appears to meet the primary goal of raising perimetry’s sensitivity to low-degree damage, it is an open question whether sensitivity can be raised additionally. For example, other combinations of microdot size, contrast, presentation time, and probe numbers might prove better. Outcomes can neither be predicted from current knowledge of visual physiology, nor illuminated by the study of normal subjects. What is required is the study of subjects with subtle degrees of neurovisual damage. The model disorder introduced here, mid-chiasmal compression, seems to be quite useful for such purposes. The tapering nature of the field defects provides a straightforward indicator of severity of damage. Further, direct imaging of the retinal nerve fiber layer and the optic chiasm can provide additional, objective indicators of severity of involvement.

A striking feature of microdot perimetry is that it does not depict normal visual fields in a format similar to the classical hill of vision. Instead, the normal response surface is essentially flat, and only occasionally marred by missed probes. Misses can be viewed as discrete defects in the neuroretinal matrix, or micro holes in the visual field. Such micro defects aggregate in abnormal numbers within established visual field defects, and the

deeper the defect, the larger the density of the micro holes. Hence, microdot visual field defects show similar spatial distributions to those of ordinary perimetry, but defects containing sloping borders tend to appear larger in rarebit perimetry (Figs. 2 and 3), as if the latter depicted a later stage of disease.

Another striking feature of the normal response surface is that its height does not vary between subjects. Unlike conventional perimetry, microdot perimetry does not show much variation in performance between normal subjects. Instead of gauging level of performance, it is submitted that microdot perimetry probes the integrity of the neural architecture of the visual system. This architecture is normally complete, in the sense that functional receptive fields fully and seamlessly tile the retina. Although normal subjects certainly differ in their numbers of receptive fields, which may somehow translate into different levels of performance in conventional tests, their neural architecture shares the invariant characteristic of complete coverage. In this regard, there is no room for individual differences. Therefore, probes for completeness of coverage normally should show little variation, i.e., normal limits should be narrow.

A recurring topic in the perimetric debate is the targeting of specific neural channels, e.g., magnoand parvocellular channels. Rarebit perimetry was not devised to target any specific channel type. It may be argued that the small size of its test targets should favor the small receptive fields of parvocellular channels. However, it may also be argued that the perception of microdots belongs to the spatial primitives of vision, and may well depend upon distributed neural processing. Presently, it would appear wise to defer from channel attribution.

The rarebit perimetry software (in Microsoft Windows format) is available from the author free of charge.

Reference

1.Frisén L: New, sensitive window on abnormal spatial vision: rarebit probing. Vision Res 42: 19311939, 2002