US 5,050,983 AGrant
Visual Testing Method and Apparatus
Issue Date:1991-09-24
•15 Claims
•10 Drawing Sheets
Abstract
Method and apparatus for visual testing. In one embodiment, for perimetry testing, the method uses threshold light-stimulus values determined at multiple seed locations in a visual field to calculate expected stimulus threshold values at locations which are nearest-neighbors of the seed locations. The expected values at the nearest-neighbor locations are compared with subject responses at these locations, and the correspondence between the two is used to refine and guide line stimulus level presentations at the nearest-neighbor locations. The test is continued until a selected confidence level of response at nearest-neighbor locations is reached.
Metadata
Assignee
- The Regents of the University of California
Inventors
- Chris A. Johnson
- Lionel R. Shapiro
Application Information
Application Number:US 5176603
Filing Date:1990-04-30
Priority Date:1990-04-30
Art Unit:257
Classifications
IPC:
A61B 302
Field of Search:
351224;225;226
Patent Drawings (10 sheets)
Description
FIELD OF THE INVENTION
The present invention relates to visual testing, and particularly, to a method and apparatus which employs an optimized test-location sequence for real time interactive visual testing.
REFERENCES
Anderson, D. R., Perimetry With and Without Automation, C. V. Mosby, St. Louis, 1987.
Hecht, E., Optics, Addison Wesley, 1987.
Heijl, A., "Computer test logics for automated perimetry," Acta Ophthalmoligica, 55:837, 1977.
Johnson, C. A., "The Test Logic of Automated Perimetry" in Acta: XXIV International Congress of Ophthalmology, Henkind (ed.), J. B. Lipincott, Philadelphia, 1983a.
Johnson, C. A., Keltner, J. L., "Incidence of Visual Field Loss in 20,000 Eyes and its Relationship to Driving Performance," Arch. Opthalmol., vol. 101, pages 371-375, 1983b.
Johnson, C. A., Adams, C. W., & Lewis, R. A., "Fatigue Effects in Automated Perimetry," Applied Optics, vol. 27, pages 1030-1037, 1988.
Keltner J. L., Johnson, C. A., "Comparative Material on Automated and Semi Automated Perimeters," Ophthalmology, vol. 93, pages 1-25, 1986.
Lewis, R. A., et al., "Preliminary Clinical Trials With the Humphrey Field Analyzer," Doc. Ophthalmol. Proc. Ser., vol. 42, pages 159-165, 1985.
Shapiro, L. R., Johnson, C. A., & Kennedy, R. L., "KRAKEN: A Computer Simulation Procedure for Static, Kinetic, Suprathreshold Static and Heuristic Perimetry," In A. Heijl (Ed.), Perimetry Update 1988/89: Proceedings of the VIIIth International Perimetry Society Meeting. Amsterdam: Kugler & Ghedini, pages 431-438, 1989.
Tyrrell, R. A., & Owens, D. A., "A Rapid Technique to Assess the Resting States of the Eyes and Other Threshold Phenomena: The Modified Binary Search (MOBS)," Behavior Research Methods, Instruments, & Computers, vol. 20, pages 137-141, 1988.
BACKGROUND OF THE INVENTION
Visual tests to determine a subject's response to such variables as light intensity, color or stimulus size or to spatial or temporal patterns are common in ophthalmic examinations. One widely used visual test, known as perimetry testing, is designed to determine the range and sensitivity of a subject's visual field (Anderson, 1987; Johnson, 1983a). Many diseases of the eye, e.g., glaucoma or optic neuropathies, affect field of vision, and often these diseases are evidenced at an early stage, when treatment may be most effective. Additionally, the type and extent of many eye diseases, and the location of the disease along the visual pathway (i.e. the eye, the optic nerve, or the visual centers of the brain) can often be determined by visual- field examination.
In standard static perimetery testing methods, the visual field is partitioned into a number of discrete locations and the subject's threshold to visual stimuli at each of the locations is determined. Heretofore, this has been done by one of two methods. In a first approach, known as the accending methods of limits, the subject is presented with a series of stimuli of step-wise increasing intensity levels, until an "unseen" stimulus is first seen. This first seen stimulus intensity determines the visual threshold. In a second approach, known as the staircase method, the subject is presented with a series of stimuli of step- wise increasing intensity levels, until an "unseen" stimulus is first seen. Thereafter, the direction of the "staircase" is reversed, until a lower limit of detection is determined. The threshold is then refined and confirmed by two additional reversals of stimulus intensity. The staircase method involves at least 5 stimulus presentations at each location. Because of the additional reversals of stimulus intensity the staircase procedure is more accurate and reliable then the accending method of limits.
A more accurate method for determining intensity threshold is by a Modified Binary Search (MOBS), a method proposed by Tyrrell et al. (1988). This procedure has been used for measuring visual thresholds, but has not previously been applied to visual field testing. In this method, a selected-intensity stimulus, typically near the expected threshold level of the subject is presented, and depending on whether the stimulus is seen, that stimulus becomes the upper or lower threshold boundry. The intensity range is then divided up into a series of increasingly smaller half- intervals until the upper (seen) and lower (not seen) threshold boundries are within a defined range, e.g., 4 dB.
In prior-art perimetry methods, threshold values--either by the accending method of limits or the staircase procedure--are determined at each location in the visual field. The test method is therefore relatively time-consuming, in that many stimuli must be presented at each location. Typically, the time required for complete testing is between 15-30 minutes. Studies have shown a loss of reliability with increasing test time, which can be accounted for by fatigue factors (Johnson et al., 1988). This problem is aggravated in older patients, the group for whom perimetry testing is generally most important. In addition, the point-by-point approach lacks an efficient method of error correction, in that bad guesses or patient mistakes during the test can only be estimated by additional "catch" trials. Suspicious test values at any location can only be checked by a full retesting of the location after the field has been completely tested.
SUMMARY OF THE INVENTION
It is one general object of the invention to provide a visual test method and apparatus which substantially reduces the number of visual stimulus presentations which must be conducted to generate a reliable visual test of a subject.
A more specific object of the invention is to provide such a method and apparatus which employs real-time control of test locations and test stimuli to enhance test efficiency and reliability.
The method of the invention is designed for testing a subject's threshold response level to a visual stimulus of the type which can be presented to the subject at a selected level and at selected locations in a two-dimensional field. The selected level of stimulus may encompass, for example, different light intensities, colors, spatial frequencies, or temporal frequencies. The two-dimensional field may encompass a two-dimensional spatial field (for perimetry testing). Alternatively, one or both of the dimensions of the field may encompass non-spatial stimulus parameters, such as color or temporal frequency of the stimulus.
In practicing the method, the two-dimensional field is partitioned into an array of N locations composed of S seed locations and N-S non-seed locations, where each seed location has multiple non-seed nearest-neighbor locations. At each seed location, threshold testing is employed. Using the threshold ranges determined at each seed location, the expected threshold of detection by the subject at each nearest-neighbor non-seed location is calculated, preferably from (a) the thresholds of nearby seed locations, (b) the distances of the nearest-neighbor location from the nearby seed locations, and (c), a position-related change in threshold which would be expected in an average subject.
The subject is then presented, at the non-seed locations, a visual stimulus which is a selected level above or below the expected threshold of detection for that non-seed location, depending on whether that non-seed location has been assigned to a super- or sub-threshold group, respectively. In one preferred embodiment, the field locations are in a regular checkerboard array, such that each location is bordered by four orthogonal nearest-neighbor locations which are assigned to one threshold group, and four diagonal nearest-neighbor locations which are assigned to the other threshold group.
The response of the subject to the stimulus presented at each nearest-location is recorded as "seen" or "not seen" by the subject. These responses are then used to classify the neighbor agreement between each location in the field and its nearest neighbors into a "discrepancy," "low-confidence," or "high-confidence" category. If the pattern of neighbor agreement at any location is in the "discrepancy" category, that location is retested for a "seen"/"not-seen" response. If the pattern of neighbor agreement for a non-seed location is in the "low-confidence" category, the expected threshold of that location is adjusted to a revised expected threshold value.
The invention further includes an apparatus for testing a subject's threshold response level to a visual stimulus, according to the above method. The apparatus includes a stimulus device designed to present a selected-level stimulus at each of N locations in a two-dimensional visual field array, and a recorder for recording a subject's response to the stimuli.
A control unit in the apparatus operates to: (1) store, at each of S seed locations in the N location array, a subject's upper and lower threshold boundries of stimulus detection, (2) calculate, for each of N-S non-seed nearest neighbor locations in the array, the expected threshold of detection by the subject, based on the values stored in said storing means, (3) present to the subject, at each non-seed location, a visual stimulus which is either a selected level above or below the expected threshold of detection for that non-seed location, depending on whether that non-seed location has been assigned to a super- or sub-threshold group, respectively, (4) classify the neighbor agreement between each location in the field and its nearest neighbors into a "discrepancy," "low-confidence," or "high-confidence" category, (5) if the pattern of neighbor agreement at any location is in the "discrepancy" category, retest that location for a "seen"/"not seen" response, and (6) if the pattern of neighbor agreement for non-seed locations is in the low-confidence category, adjust the expected threshold of that location to a revised expected threshold value, and recalculate the nearest-neighbor agreement between the revised- threshold location and its nearest neighbors.
These and other objects and features of the present invention will become more fully apparent when the following detailed description of the invention is read in conjunction with the accompanying drawings.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 is a schematic view of an apparatus for perimetry testing, in accordance with the invention;
FIG. 2 shows the partitioning of a two-dimensional spatial field for perimetery testing, in accordance with the method of the invention;
FIG. 3 is a flow diagram of steps for determining a threshold intensity range at each seed location;
FIG. 4 is a flow diagram of steps for calculating the expected threshold values of non-seed locations;
FIG. 5 is a flow diagram of steps for measuring the correspondence between expected and recorded threshold levels at all locations;
FIG. 6 is a flow diagram of the classification and retest steps in the method of the invention;
FIG. 7 is a flow diagram of the program control used for evaluating and adjusting recorded response information in a test procedure;
FIG. 8 is a flow diagram of the program control for evaluation and adjusting recorded response information for low confidence locations that remain unresolved at the completion of procedure shown in FIG. 7;
FIG. 9 illustrates the relationship among the various operations carried out by the control unit in the apparatus of the invention;
FIGS. 10A-10D show visual field maps of an individual with normal visual field (FIGS. 10A and 10B), and actual threshold values measured (FIGS. 10C and 10D) using a standard 30-2 test strategy (FIGS. 10A and 10C) and the method of the present invention (FIGS. 10B and 10D);
FIG. 10E shows the symbols used in FIGS. 10A and 10B; and
FIGS. 11A-11C show visual field maps of an individual with moderate-severe glaucoma measured by perimetry testing by the method of the present invention (FIG. 11A), upon retest by the method of the invention one week after the first test (FIG. 11B), and by a standard 30-2 test stra-tegy (FIG. 11C), and FIG. 11D shows the symbols used in FIGS. 11A-11C.
DETAILED DESCRIPTION OF THE INVENTION
FIG. 1 is a schematic illustration of a perimetry test employing the method and apparatus of the invention. Shown at 10 and 12 in the figure are fixed- and movable-position image sources, respectively, for presenting selected-image stimuli to a subject 14. In a preferred embodiment, images are viewed through an ocular 16 which allows the images to be viewed without corrective lenses, as described in co-owned U.S. patent application Ser. No. 482,278, filed Feb. 20, 1990.
The images observed by the subject are recorded by a response recorder 20 in the apparatus, and these responses are passed to a control unit 22 which evaluates the responses and controls the stimulus presentation to the subject, in accordance with the present invention, to determine a visual threshold map. The results may be displayed on an output 24. The visual test method, and its control by unit 22, will be described in Section B. Apparatus for carrying out the method, including a preferred stimulus-presentation system, recorder 20 and control unit 22 will be detailed in Section C. Section A gives a guide to the Appendices which include the software commands used in the apparatus for real-time evaluation of the testing results and static perimetru test presentations.
A. Guide to the Appendices
In the appendices, the source code for RIOTS and necessary support procedures are in the Turbo Pascal (version 5.0) language.
RIOTS (Appendix A)
RIOTS (Real-Time Interactive Optimized Test Sequence) includes the heuristic strategy for static perimetry, in accordance with the method of the invention, and as detailed in Section B. Details of the method will be given with reference to the various sections of the RIOTS program which are indicated in the text.
KRAKEN (Appendix B)
KRAKEN is a general simulation and control program for evaluation of perimetric strategies and devices (Shapiro et al., 1989). KRAKEN is the supervisory program from which support procedures are called to handle specific tasks. These tasks include: (1) user interaction for the determination of operational parameters; (2) file preparation; (3) device or simulation initialization; (4) execution of specific strategies; and (5) recording of operational variables as well as test results.
Routines for user interaction (that is, the control and display of operational parameters used by KRAKEN and its support routines), file preparation and device or simulation initialization include:
KRMENU
Initializes a menu system (Turbo Professional routines, TP-prefix, provide the specific control) for choosing options in the control of KRAKEN, see also SIMPARAM for implementation of choice options. KRMENU is called by KRAKEN only.
PLOTPAK
Provides for graphical output of operational parameters and variables. This code contains a library of routines for specific graphic output tasks. These routines may be called from any number of points within any of the routines which follow.
TIMELIB
Provides routines for accessing and displaying time and date variables. These routines may be called from any number of points within any of the routines which follow.
PADPAK
Provides routines for control and information retrieval from a graphics input device such as a Summagraphics Bit Pad. This allows the operator of KRAKEN to control placement and levels of stimuli when KRAKEN is in a manual mode of operation. These routines are called primarily from the RESPN library (see below).
TRIG
Provides routines for trigonometric calculations. These routines may be called from any number of points within any of the routines which follow.
DWIGHT
Provides routines for control of device described in co-owned U.S. patent application Ser. No. 482,278, filed Feb. 20, 1990. These routines are called primarily from the RESPN library (see below).
GLOBALS
Defines and initializes some operational and all functional parameters and structures used by all of the following routines.
RANDOMZ
An initialization routine to provide for uniformly distributed random number generation. This allows the simulation portions of KRAKEN and its supporting routines to generate a reasonable simulation of events with constrained random variation. These routines are called primarily from the RESPN library (see below).
RESPN
KRAKEN may either simulate a patient or collect responses from an actual patient; and either simulate a perimetric device or control and receive patient responses from an actual perimetric device. This library of routines contains two types of response procedures: simulation of patient o interfacing control with actual devices. These routines are called through a standard routine called RESPOND which diverts control to the appropriate support routine based upon the specified operational parameters.
GNAME
This simply allows for operator entry of file names.
HC30--2
This consists of a definition of a data structure for the 30-2 test pattern. This data provides a list of neighbor groups for each of 76 locations in the test pattern.
SIMPARAM
This code provides the remaining definitions and initializations of operational parameters. The code also contains a routine which allows for modification of a number of these parameters via a menu oriented system (see KRMENU above). Code defined here is called from the KRAKEN main routine only.
MIN MAX
Simple support routines for determination of minimum and maximum values given a specified pair of values.
USTATIC
Support routines for static-type perimetry operation. For example, the display of parameters and variables specific to static stimuli presentations.
PAUSE
A simple routine to allow for operator controlled pauses in program execution.
KREXIT
Program control code to be executed upon the call for program termination.
PLOTMAN
Provides a routine for graphical display of a manifold (visual field and related information) data structure.
INIMOD
Initializes file and control variables associated with patient and reference files (see also normal reference files below).
REPORT
Called to output, in a variety of formats, data obtained through the application of the perimetric strategies via either simulation or device control.
The following are strategy code modules for simulation of specific perimetric devices or procedures:
SIMPLEST
This is the strategy code for a simple stair case procedure for static
HUMCEN30
This is the strategy code for a simulation of the type similar to the Humphrey Full Threshold (staircase) procedure for static perimetry.
MOBSC30
This is the strategy code for a modified procedure for static perimetry
SUPRA30
This is the strategy code for a simple suprathreshold static screening perimetric procedure.
Files (Appendix C)
In order to run the KRAKEN package, access must be made to normal sensitivity reference files. These data are calculated from statistical summaries of population data for specific perimetric tests. Appendix C contains reference files (for 7 age groups) for a typical test pattern used in the present invention; this layout information is redundant with the support file: CEN30-b 2.GRD). Any remaining references to additional support files may be ignored as they do not pertain to the execution of the RIOTS procedure with non-simulated perimetric devices.
B. Visual Testing Method
1. Partitioning the Visual Field
The method is designed for visual testing by light stimuli which can be represented as selected value points in a multi-variable space. In the typical case of perimetry testing, the stimuli are static light stimuli which are presented at selected locations in a two-dimensional spatial field. The static stimuli may vary, for example, in intensity, size, or color. Thus, the multi-variable space includes two spatial dimensions and a stimulus-variable dimension. Alternatively, in perimetry testing, the stimuli may themselves may be defined by two or more variables, such as intensity and color, giving a four dimensional test space.
Another type of multi-variable visual testing which is adaptable to the present invention involves a subject's ability to distinguish bar patterns having different spatial and temporal frequencies, and intensities. For example, the subject may be presented with a bar pattern having a selected intensity, bar spacing and flicker frequency, and these three parameters are varied to determine threshold values at different positions in the three-dimensional test space.
More generally, the present invention is useful in visual testing in which the threshold response of one type of visual stimulus is displayed at selected "locations" in a two- or higher-dimension test space. The test space may be defined by spatial coordinates, such as a two-dimensional visual field, by non spatial coordinates, such as spatial and temporal frequencies in the bar pattern, or by a mixture of spatial and non-spatial coordinates, such as a two-dimensional visual field containing a third light-stimulus variable, such as color.
For purposes of illustration, the test space described herein is a two-dimensional visual field defined by the projection of x and y coordinates on a hemispherical cap region. An exemplary test field is shown at 30 in FIG. 2. The outer boundary of the field is defined by the perimeter of a cone formed at a 30.degree. angle from the horizontal line of sight of the subject, as illustrated by the angle alpha in FIG. 1. As seen, the cap region is divided into four quadrants containing an equal number of test locations, such as locations 40, 42, 44, 46, 48, 50, and 52. The locations 40-52 and the two locations on either side of location 52 form a nine-location array 53 of locations representative of the ones which are shown in the figure. Each of the locations are equally spaced from one another, along horizontal and vertical axes, in a two-dimensional projection, as illustrated.
With continued reference to FIG. 2, the array of locations such as location 40, are subdivided into S seed ("S") locations, and N-S non-seed locations. Here the number of seed and non-seed locations is 12 and 64, respectively. As seen, each location other than an edge location has eight nearest neighbors, four of which are horizontal or vertical nearest neighbors and four of which are diagonal nearest neighbors. For example, the nearest neighbors of seed location 40 (filled circle) in array 53 are locations 48, 42, 50, 46, and 44 which are indicated with thick circles (horizontal and vertical nearest neighbor locations) and thin circles (diagonal nearest neighbors).
The projection of the above array of locations in a two-dimensional plane can also be thought of as a checkerboard array in which each location, such as location 50, has four same-color nearest neighbors (the diagonal nearest neighbors) and four different-color nearest neighbors (the vertical and horizontal nearest neighbors).
It will be appreciated that other arrangements of locations in two-dimensional space are possible. For example, the locations may be arranged in a regular hexagonal array with each point surrounded by three immediate nearest neighbors and nine more removed nearest neighbors. More generally, the method requires a partitioning of the space into locations which each have multiple nearest neighbor locations which can be assigned to two threshold groups, as discussed below.
2. Determining Seed Location Values
In the first phase of the test method, the threshold range of detection of a visual stimulus by the subject is determined at each seed location in the test field. As noted above, the visual stimulus may relate to light intensity, color, size, or spatial or temporal frequency. Regardless of the nature of the stimulus, the threshold test is carried out to determine the lower and upper threshold boundries of perception of the stimulus by the subject. Thus, in the case of light-intensity stimuli, the testing is carried out to determine the least upper bound and greatest lower bound of intensity which is perceived by the subject, at a given location.
The method preferably employs a reliable method for threshold determinations, such as a staircase or modified binary search (MOBS ["MOBS" function on pp. 2-6, Appendix A]) mode of testing, as have been described (Tyrrell). A preferred test procedure, for use in determination of light intensity threshold, employs a variant of the MOBS approach, as outlined in FIG. 3. In the procedure, a light source is placed at a selected seed location in the field (Section C below). The source is controllable to output a selected-size image, typically, 1 to 4 mm, as viewed from a distance of 0.33 meters, for a selected time interval, e.g., 100 to 500 msec, over a selected intensity range, typically between about 0 to 40 dB.
The first light stimulus which is presented to the subject at a seed location has an intensity which is 2/3 along the interval between the absolute lower (L) and upper (U) boundries of light intensities of the source. More generally, the intensity is biased somewhat in favor of receiving a "seen" response from the subject. Assuming the result is seen, the first presented intensity becomes the current upper threshold boundry T.sub.u, and the intensity is reduced by 1/2 the interval between T.sub.u and the lower boundry. That is, the second stimulus intensity is 1/2 T.sub.u -T.sub.l, where T.sub.l is the current lower boundry, which is at this point the lower bound L. This step represents a first reversal in direction of intensity. If the first stimulus is not seen, this first stimulus intensity becomes the current lower bound T.sub.l and the next stimulus which is presented has an intensity halfway between this value and the highest intensity. This step is not counted as a change in direction. These steps are shown at the top in FIG. 3.
If the second stimulus is lower intensity, and not seen, this second-stimulus intensity becomes the new lower current bound T.sub.l, and the next intensity is similarly set halfway between the current T.sub.u and T.sub.l. This step represents a second change in direction. If the second stimulus is higher intensity, and seen, this second-stimulus intensity becomes the new upper current bound T.sub.u, and the next intensity is similarly set halfway between the current T.sub.u -T.sub.l. This step also represents a reversal in direction.
If at any time the lower boundry goes above the upper boundary then the upper boundary is reset to its previous value. Similarly, if the upper boundary goes below the lower boundary then the lower boundary is reset to its previous value. This is handled by a LIFO (last in, first out) stack queue data structure.
This procedure is carried out until a total of at least R (e.g., R=4) reversals in direction, and a defined specific interval between upper and lower stimulus boundaries is achieved. The four reversals require at least five different-intensity presentations, and typically less than 8 presentations to achieve a preferred upper and lower boundary interval of about 4 dB. The steps are indicated in flow diagram form in FIG. 3. A threshold value T.sub.i for each seed point S.sub.i is calculated as the average of the T.sub.l and T.sub.u values.
It will be appreciated how the above modified MOBS procedure can be applied for determining threshold values of other types of visual stimuli, such as temporal or spatial frequency of a spaced-bar stimulus.
The threshold range determined at each seed point is stored in the control unit, for use in calculating expected threshold values of neighboring non-seed locations, as will now be described.
3. Calculating Expected Non-Seed Threshold Values
After the threshold value of each seed point has been established, the control unit operates to calculate expected thresholds at each of the non-seed locations, based on the threshold values measured at the neighboring seed locations. The flow diagram in FIG. 4 ["open-neighbors" function on pp. 11-12, Appendix A] shows an exemplary procedure for calculating expected intensity thresholds for non-seed locations. The procedure looks at each non-seed location N.sub.j location, and retrieves the threshold values T.sub.i of each nearest neighbor seed point S.sub.i. With reference to FIG. 2, for example, the nearest-neighbor seed locations of non-seed location 50 are the seed points 40, 52. Other non-seed points may have only one immediately neighboring seed point.
The threshold T.sub.si of each such seed point is then compared with a "normal" value T.sub.ni for that i location in the visual field, based on a statistical sampling of individuals with normal field of vision. If the measured and normal values are equal or nearly equal, the expected threshold value for the location E.sub.j with respect to the seed point S.sub.i is set equal to the normal value E.sub.nj expected at the non-seed location.
If, however, the measured and normal values at a neighboring seed location are not substantially equal, the program calculates an expected value E.sub.j which is based on the measured seed threshold and the distance between the seed location S.sub.i and non-seed location N.sub.j. This calculation uses the empirical formula:
where the radii of N.sub.j and S.sub.i refers to the degrees of eccentricity of the locations away from the center of the visual field. Thus all points on the perimeter of the visual field array shown in FIG. 2 have an eccentricity of 30.degree.. The 0.3 dB/degree constant may be slightly greater or smaller if stimulus parameters are employed.
After the value E.sub.j has been calculated with respect to each nearest-neighbor of the S.sub.i location, as above, the expected threshold values are averaged ["average-neighbors" function on pp. 7-8, Appendix A] to produce an expected value E.sub.j for that non-seed location. These values are then stored for use in visual stimulus presentations.
4. Stimulus Presentation
Each location in the test field array has now been assigned a stimulus threshold value: For seed locations, that value was determined directly by testing the threshold value of stimulus detection of the subject. For the remaining, and more numerous non-seed locations, the threshold value was an expected value calculated from neighboring seed values.
In the next step of the method, described in flowchart FIG. 5 ["Pass 1 or 2" function, lines 14 to 56, Appendix A], a test stimulus is presented to the subject at each of the non-seed locations, to confirm or refine the expected threshold value at that location. According to an important aspect of the stimulus presentation step, the non-seed locations in the test array are divided into two groups: a sub- threshold group at which below-threshold stimuli will be presented to the subject, and a super-threshold group at which above-threshold stimuli will be presented. For purposes of obtaining reliable neighbor agreement patterns, each location in the test array, including both seed and non-seed points, is preferably surrounded by (nearest neighbors of) two or more non-seed locations in each group.
In a preferred configuration, illustrated in FIG. 2, each internal seed location has eight nearest neighbors, with the non-seed locations which are orthogonal to the central seed location being assigned to one group and the diagonal nearest neighbors being assigned to the other group. For example, seed location 40 is surrounded by orthogonal neighbors represented by thick circles and diagonal neighbors represented by thin circles. Viewing the array of locations as a checkerboard, all the locations of one "color" are assigned to one group, and the locations of another "color," to the second group. As will be seen below, the locations of the two groups are reversed in different test phases.
The stimulus presented at a non-seed location in a sub-threshold group is a selected level below the threshold calculated for that location. By way of illustration, where the stimulus is a light flash of a selected intensity, the intensity of the stimulus is a selected level, e.g., 4 dB, below the calculated intensity threshold of the given non- seed location. The selected level below threshold is such as to guarantee a strong likelihood of obtaining a confirming response, that is, a "non-seen" response, from a normal subject when presented with the sub-threshold stimulus at a given location.
Similarly, the stimulus presented at each non-seed location in a super-threshold group is a selected level above the calculated threshold for that location, and preferably the same selected level, e.g., 4 dB for intensity testing, as applied to the sub-threshold group.
The sequence of stimulus presentation at the non-seed locations is preferably random, at least from the point of view of the subject, so that successive presentations are typically widely spaced over the test field, and sub-threshold stimuli are mixed randomly with super-threshold stimuli. This prevents the subject from concentrating on any area within the field, or expecting a stimulus within any area of the field or within any time interval. In one preferred device for stimulus presentation, discussed in Section B below, the successive light flashes are presented on an average of every 1 to 2 seconds.
Each "seen" stimulus presentation is recorded by the subject, typically by means of a hand-held pushbutton switch connected to recorder 20. A non-response is of course recorded in the recorder as a "non-seen response." The responses are stored in the control unit until all of the non-seed locations have been tested.
The procedure for testing the non-seed locations just described applies to a first-phase test in which all of the non-seed locations are tested at a fixed, selected level above or below the estimated threshold for each location. As described below, the method preferably involves a second- phase test in which the sub- and super-threshold assignments are reversed and the selected stimulus levels above and below expected thresholds are tailored to the confidence level at each non-seed location. The method also contemplates a third test phase in which the non-seed locations are returned to their original group assignments, and only points outside of a given confidence level are tested. The stimulus presentation in these second-and third-phase tests will be clear from the description of the tests below.
5. Classifying Neighborhood Patterns
At the conclusion of each test phase in the method, the locations in the test array are analyzed for agreement with nearest neighbors, to confirm or refine the threshold value tested during that test phase. Depending on the pattern of neighborhood agreement which is observed at a location, the location is classified into one of at least two, and preferably three, "confidence" categories, according to the agreement between that location and the immediate nearest neighbors. The two basic categories are a "low-confidence" and "high-confidence" class used to characterize those locations in which the agreement between a location and its nearest neighbors is below and above a selected confidence limit, respectively. The third class is a "discrepancy" category in which the location is in disagreement, typically complete disagreement, with its nearest neighbors.
In one general procedure for classification of neighbor agreement, each location in the test field is evaluated for self-consistency between expected and recorded thresholds, and for pairwise-consistency with its nearest neighbors. A location is self-consistent if it is (a) a non-seed location which is expected seen (superthreshold group) and is recorded as "seen," (b) a non-seed location which is expected not-seen (subthreshold group) and is recorded as "not-seen," or (c) a seed location (which is self-consistent by definition). A location is self-inconsistent if it is (a) a non-seed location which is expected seen (superthreshold group) and is recorded as "not-seen," or (b) a non-seed location which is expected not-seen (subthreshold group) and is recorded as "seen."
A pair of locations are pairwise-consistent if the locations are both self-consistent or self-inconsistent. The pair is pairwise-inconsistent if one of the pair is self-consistent, and the other self-inconsistent. Thus for example, with reference to FIG. 2, if non-seed point 44 is self-inconsistent and nearest-neighbor seed location 40 is self-consistent (as it must be), the two locations are scored as pairwise inconsistent. On the other hand, if non-seed point 46 is also self-inconsistent, then locations 44, 46 are scored as pairwise consistent.
The scoring algorithm considers each location in the array, scores the neighborhood agreement, based on a total pairwise consistency score, classifies that location into one of the possible confidence classes, then moves on to the next location in the array. To illustrate, the operation of the algorithm at non-seed point 50 will be considered. Here it is assumed that this location was expected seen (super-threshold group), but was recorded as "not-seen," indicating that the original calculated expected threshold for this location is too high. That is, the location is self-inconsistent.
The algorithm then performs a pairwise comparison between location 50 and each of the eight nearest neighbor locations, such as locations, 52, 48, 40, and 42. For purposes of illustration, it will be assumed that locations 52 and 40 are self-consistent (because they are seed points), location 50 is self-inconsistent (because it was expected "seen" and recorded as "not-seen"), location 44 is self-consistent (because it was expected "seen" and recorded as "seen"), and location 46 is self-inconsistent (because it was expected "not-seen" and was recorded as "seen"). Therefore, the pairwise score for locations 50, 44, is pairwise- inconsistent; for locations 50, 40 is pairwise-inconsistent; and for locations 50, 46, pairwise-consistent.
Notice that in the present case where the center location 50 is self-inconsistent, a non-seed location, such as location 46 may be self-inconsistent but pairwise consistent or may be self-consistent and pairwise inconsistent, as is the case for location 44. Where the center location is self-consistent, each nearest-neighbor non-seed location will be either self-consistent and pairwise consistent or self-inconsistent and pairwise inconsistent.
After each nearest-neighbor pair has been scored at a location, the location is classified according to the confidence value, as indicated above. One exemplary classification is as follows: (a) If the location is in complete disagreement with its nearest neighbors, i.e., all pairs have been scored pairwise-inconsistent, then the location is assigned to a "discrepancy" class; (b) If the pairwise self- consistency score is greater than 75% (e.g., at least seven of the eight pairs in a rectangular array FIG. 2 are pairwise consistent), then the locations assign to a "high-confidence" class; and (c) If neither (a) nor (b) apply, assign to a "low-confidence" class.
Depending on the classification of each location, one of several programmed actions are taken before the next phase of the test, as illustrated in the flow diagram in FIG. 6 ["Heuristics" procedure, pp. 8-9, Appendix A]. If a non-seed location is in a discrepancy class, that location is retested at the original calculated stimulus value, to confirm whether the response at that location was in error. If a different response is obtained, the "expected" threshold at that location is upgraded to correspond with the retested response.
If a non-seed location is in a "low-confidence" class, the expected threshold at that location is adjusted to a new threshold expectation value which is calculated to increase the nearest-neighbor agreement of that location. In one preferred method, the adjustment is made on the basis of position in the visual field and the expected values of neighboring locations, weighted by the confidence limits of those locations. If the location is in a "not-seen" group, the expected threshold is increased halfway between the previous threshold and the maximum stimulus value.
Each non-seed point in the test field has now been assigned a confidence level with respect to neighboring points in the test field, and the expectation value at this location has been adjusted or retested if the original value is below a given confidence level.
6. Refining Confidence Limits
Following the first-phase presentation, in which all of the non-seed locations are tested for agreement with expected values, the test array is retested to improve confidence limits, particularly at locations which are originally scored as in "discrepancy" or "low-confidence" classes. In one preferred embodiment, all of the non-seed locations in the array are retested in a second-phase presentation to (a) refine the expected threshold interval of "high-confidence" locations, and (b) recalculate the confidence limits of non-seed locations which were originally scored as "low-confidence" and therefore have adjusted revised expected threshold values. Selected locations are then retested in a third-phase presentation to refine the confidence limits.
FIG. 7 ["Pass 1 or 2" function, pg. 16, lines 14-53, Appendix A] shows a flow diagram of a second-phase stimulus presentation and analysis. As a first step, the sub- and super-threshold groups of non-seed locations are reversed; that is, a non- seed location originally assigned a to a sub-threshold or super-threshold group is now placed in a super-threshold or sub-threshold group, respectively. At each non-seed location, the confidence level of the location (determined from above) is examined. If the location was originally scored "high-confidence," the level of stimulus presentation above or below the expected threshold is reduced from the selected level Y employed in the first phase, to Y/2. For example, if the first-phase presentation was 2 dB above or below the expected intensity at a non-seed location, the second-phase intensity at "high-confidence" locations is 1 dB above or below expected threshold. If the non-seed location has a lower confidence value, the stimulus presentation level above or below the expected threshold is unchanged from the first- phase presentation. e.g., 2 dB above or below the expected threshold level.
After all of the non-seed locations have been tested, the analysis performed after the first-phase presentation is repeated to determine the confidence level of threshold response at each non-seed location. Here it is noted that the agreement analysis with nearest neighbors can provide an additional high-confidence classification, namely, high-confidence as determined for nearest neighbors which were tested at the refined threshold stimulus values. Any locations which are classed as "low-confidence" with respect to their nearest neighbors are adjusted in expected threshold to a value calculated to produce greater neighbor agreement, as described above.
The second-phase of the test may reveal locations which are in complete disagreement with all nearest neighbors after a second retesting. These locations are assumed to be spatial anomalies, strongly unrelated to nearest neighbors, and are separately tested directly for threshold value, using the method applied to the seed locations to establish their threshold value. These final processing steps are indicated at the bottom in FIG. 7.
Following the stimulus presentation and analysis in the second phase, the expected threshold values in the test field may be refined with a third-phase presentation. Unlike the first- and second-phase presentations, the third phase is concerned only with refining the confidence level of those locations which are in a "low-confidence" class after the second phase; that is, those locations for which an expected threshold adjustment was made after the second phase. All locations in the "high-confidence" class are deemed to be reliable after the second phase.
The flow diagram of the third phase of the test procedure is given in FIG. 8 ["Pass" procedure, page 22, line 24, to page 23, line 24, Appendix A]. As indicated, the sub- and super-threshold groups are reversed once more, returning the non-seed locations to their original group assignments. Each non-seed location in a "low-confidence" class is now presented a stimulus at the same level above or below the expected threshold value as used in the first- phase, e.g., 2 dB in intensity level.
After all of the "low-confidence" locations have been tested in the third phase, each location can be assigned a threshold value with defined confidence limits given to every non-seed location. The threshold value is, of course, the originally calculated or adjusted threshold value assigned to each non-seed location. The confidence limit assigned to each non-seed point is determined from (a) the extent of neighborhood agreement and (b) the confidence level of self-consistency for that location and its nearest neighbors, i.e., the selected range above or below expected threshold at which the location and its neighbors were scored as "self-consistent."
The final results are preferably presented as a two dimensional map showing the location, threshold value, and confidence limit given at each point. From this output, the subject's ocular condition related to the tested parameters can be determined.
The test method described above can be easily abbreviated if a lower confidence level for non-seed locations is acceptable, for example, in field testing or for routine screening of younger subjects. In one abbreviated test, a first-phase presentation, carried out as above, is followed by a second-phase test which is designed only to retest and upgrade "low-confidence" non-seed locations until all locations have been classed as "high-confidence" points. This abbreviated method reduces by 25-50% the number of stimulus presentations which are typically made in a visual test.
C. Visual-Test Apparatus
1. Stimulus Presentation
Stimulus presentation in the method may be made by any of a variety of visual-test devices designed for presenting a given type of stimulus over a range of stimulus values, e.g., intensity, color, size, and spatial or temporal frequency. For perimetry testing, where light stimuli are presented at selected locations in a visual field, the device may be a conventional projector or light-diode array type stimulus device.
One preferred type of device for perimetry testing is a Badal-type optical device described in co-owned U.S. patent application for "Perimetry Test Device and Method," Ser. No. 482,278, filed Feb. 20, 1990. Briefly, with respect to FIG. 1, this device includes fixed-position image source 10 for directing a low-intensity light beam 28 at the subject, and movable image source 30 for directing a selected-intensity beam 32 at the subject from any of the locations in the test field. The movable image is a variable-intensity light field. The movable image is a variable intensity light source carried at the end of a two-arm assembly designed to move the light source between any two locations in the test field in a period of about 1-3 sec.
The ocular through which the light images are viewed is provided with diffuse, back-scattered light which serves to mask background and other low-level stray light.
2. Recorder and Control Unit
The subject "seen" responses are recorded in a conventional recorder 20 preferably activated by a handheld push button, as indicated above, and "Not-seen" responses are recorded after each light stimulus which is not seen by the subject.
The control unit includes a conventional microprocessor, such as a microcomputer, which receives digitized response signals from recorder 20, performs realtime test- control calculations, and outputs signals to the stimulus presentation device for presentation of selected-level stimuli to the subject at selected locations in the test field. The operations carried out by the control unit 22, and the interaction of the unit with the presentation device, recorder and output display device are shown in FIG. 9 where the microprocessor, and the software employed in the test method are indicated by solid lines 60. The source code of software referred to in the description of the control unit is provided in the Appendices described in Section A.
It will be understood that the functions indicated in FIG. 9 do not necessarily represent separate parts of the program carried out by the software, or separate hardware operations. Rather, they indicate
In the initial phase of the test, the control unit operates to (a) select a seed location selection routip. 11 - Get next target, (b) instruct the presentation device to move to the selected location, and (c) instruct the presentation device to present a stimulus of a given level, e.g., given intensity. ("MMOBS" presentation, pp. 9 to 10, Appendix A). These operations are embodied in a "present" algorithm indicated at 62.
The subject response to the seed-location stimulus is supplied to the program through a record function 64, and the stored value is passed to a "calculate" function 66 to determine (a) the current upper or lower bound and (b) the level of the next stimulus at the seed location. This information is supplied through the "present" function to the presentation device. These steps are repeated until the threshold range at that seed location has been determined, according to the algorithm outlined in FIG. 3. The control unit then advances the presentation device to the next seed point for threshold determination. The process is repeated until all seed points have been tested ("Resolve" comands in MMOBS"presentation at pg. 11, Appendix A).
"Calculate" function 66 in the control unit also uses the seed-location values to calculate expected threshold values, and corresponding sub- and super-threshold values, at each of the non-seed locations, employing the algorithm outlined in FIG. 4 and given on pages 11 to 12 of "Open Neighbors," Appendix A. These values are stored in the "store" function 67 until used by the "present" function to present the selected sub- or super-threshold value at the associated non-seed location.
The responses to the non-seed stimulus presentations are recorded, stored and passed to an "analyze" function 68, as indicated, for neighborhood agreement analysis ("Average-Neighbor-xpectation," pp. 7 and 8, Appendix A). The analysis algorithm calculates the confidence limit of each location in the test field and classifies the location on the basis of the confidence value, as described above.
Depending on the classification, the location may be retested (for a discrepancy-class), or adjusted in threshold value (for a "low-confidence" value), according to the algorithm given in the "Heuristics" function on pp. 8 and 9 of Appendix A. The "threshold adjust" function is shown at 70 in FIG. 9.
The control unit continues the test procedure until desired confidence limits are reached, as described. The results, presented in terms of thresholds and confidence limits at the locations in the field are displayed at the output. Program steps for carrying out the retest and second- and third-phase of the test method are given on pages 14 to 27 of the Appendix A.
The final threshold values calculated for the locations in the visual field are plotted at output 24 in a suitable form, such as the visual field maps shown in FIGS. 10 and 11.
D. Test Results
Quantitative automated static perimeters known in the prior art use rapid staircase procedures to measure visual field sensitivity. Previous studies (both computer simulations and clinical evaluations of patients) indicate that the accuracy and efficiency of these staircase procedures are at nearly optimal expected levels of performance.
Extensive computer simulation comparisons of the method of the invention, as applied to perimetry, with standard staircase perimetry testing methods (Humphrey Analyser 30-2 Full Threshold Test). The results show that the present invention is significantly more efficient, typically requiring one-third to one-half or more fewer presentations to complete testing. The present method also has more consistent test times. Normal individuals and patients with visual field loss are all tested in 230-275 presentations, over a total test time of 6-8 minutes. By contrast, the standard staircase presentation typically requires between about 350-500 presentations for normal individuals and between about 500-900 presentations for patients with moderate-severe visual field impairment, as seen below, with test times ranging from 12-27 minutes.
In addition, the method of the invention was found to have slightly better accuracy and test reliability, and is less susceptible to the influence of fatigue and boredom, due to the fewer test presentations.
FIGS. 10A-10D compare visual field tests performed 10 A and 10C) and the present method (FIGS. 10B and 10D) on an individual with normal visual field. The total number of presentations in the 30-2 test was 369, and in the perimetry method of the invention, 231. The threshold values determined by each test are shown at the bottom of the figure. The average difference between the locations measured in the two test was 0.51 dB. The corresponding visual field maps, expressed in 5-dB increments (FIG. 10E) are given at the top in the figure.
FIGS. 11A-11C show a similar test comparison preformed on a patient with moderate to severe glaucoma. FIG. 11A is the visual field map generated in a test with 237 presentations, in accordance with the present invention. The map show severe loss of visual field in the lower left quadrant (inferior nasal) of the visual field and moderate loss in the other three quadrants. The patient was retested one week later with the same method, using 256 test presentations, with the results shown at the right in FIG. 11B. A comparison of the visual fields in FIGS. 11A and 11B shows the degree of test consistency which is achieved.
The same patient was tested by the 30-2 standard staircase method, requiring a total of 591 presentations, i.e., more than twice the number of presentations required in the present method. The visual field map generated by the 30-2 test is shown in FIG. 11C. The average difference between the first test (FIGS. 11A) and the 30-2 test was 1.49 dB, and between the second test (FIG. 11B) and the 30-2 test, 1.82 dB.
The advantages in test efficiency can be readily appreciated from the clinical tests above. As discussed, the greater efficiency is achieved, in part, by updating threshold and confidence values assigned to each point during the test, i.e., in real time, to optimize the presentation of additional stimuli. The method also provides a means of immediate real-time detection and correction of response errors as the test is being performed.
Although the invention has been described with reference to particular embodiments and test procedures, it will be apparent that various changes and modifications can be made without departing from the invention. ##SPC1##
Claims
It is claimed:
1. A method of testing a subject's threshold response level to a visual stimulus which can be presented to the subject at a selected level and at selected locations in a visual field, comprising (a) partitioning the field into an array of N locations composed of S seed locations and N-S non-seed locations, where each seed location has multiple nearest-neighbor non-seed locations, (b) determining, at each seed location, a threshold value of detection of a visual stimulus by the subject being tested, (c) calculating the expected threshold values of detection by the subject at each non-seed location, (d) presenting at each non-seed location a visual stimulus which is either a selected level above or below the expected threshold value of detection for that non-seed location, depending on whether that non-seed location has been assigned to a super- or sub-threshold group, respectively, (e) recording the response of the subject to each presented stimulus as "seen" or "not seen," (f) classifying the neighbor agreement between each location in the field and its nearest neighbors into a "low- confidence" or "high-confidence" class, (g) if the pattern of neighbor agreement for non- seed location is in the low-confidence category, adjusting the expected threshold of that location to a revised expected threshold value, and retesting the nearest-neighbor agreement between the revised-threshold location and its nearest neighbors.
2. The method of claim 1, wherein the classes into which the neighborhood agreement can be classified further include a "discrepancy" class, which further includes a step (h) of retesting a location which is classified in the "discrepancy" class.
3. The method of claim 2, for use in visual field perimetry testing, wherein said two-dimensional spatial field which substantially overlaps the subject's visual field, and the field is partitioned into 50-100 regularly-spaced locations with each seed point being bordered by at least 6-8 neighbors.
4. The method of claim 3, wherein said field is partitioned into a regular rectangular array of locations, with any give location being bordered by four orthogonal nearest neighbors of the same expected response group four diagonal nearest neighbors of the opposite expected response group.
5. The method of claim 2, wherein said determining includes measuring upper and lower boundaries of detection by the subject of the intensity of a short-duration flash of light at each seed location.
6. The method of claim 5, wherein the expected threshold intensity value of light-stimulus detection at each nearest-neighbor location is determined from (i) the threshold values of nearby seed locations, (ii) the distances of the nearest-neighbor location from the nearby seed locations, and (iii), a position-related change in threshold range which would be expected in an average normal subject.
7. The method of claim 2, wherein the discrepancy category is defined by complete disagreement with a center location and its nearest neighbors.
8. The method of claim 7, which further includes reversing the assignment of the sub- and supra-threshold groups, and repeating steps (d)-(h), where the selected level of stimulus above or below threshold which is presented at each non-seed location is (i) the original selected level for non-seed locations which are nearest neighbors of locations classified as low-confidence, and (ii) a reduced selected level for non-seed locations which are nearest neighbors of locations classified as high-confidence levels.
9. The method of claim 8, which further includes reversing the assignment of the sub- and suprathreshold groups to their original assignments, and repeating steps (d)-(h) for nearest neighbors of locations which are classified as low-confidence.
10. The method of claim 8, wherein a non-seed location originally classified as a discrepancy location is classified as a discrepancy location after repeating steps (d)-(h), which further includes determining, at that location a threshold value of detection of a visual stimulus by the subject being tested.
11. The method of claim 1, wherein said determining is carried out by a optimized modified binary search procedure.
12. Apparatus for testing a subject's threshold response level to a visual stimulus which can be presented to the subject at a selected level and at selected locations in a two-dimensional field, comprising (a) presenting means for presenting a selected-level stimulus at each of N locations in a two-dimensional visual field array, (b) recording means for recording the response of the subject to the stimuli presented at locations in the field as "seen" or "not seen," and (c) a control unit which operates to (1) store, at each of S seed locations in the N location array, a subject's upper and lower boundaries of stimulus detection, (2) calculate, for each of N-S non-seed nearest neighbor locations in the array, the expected threshold value of detection by the subject, based on the values stored in said storing means, (3) present to the subject, at each non-seed location, a visual stimulus which is either a selected level above or below the expected threshold value of detection for that non-seed location, depending on whether that non-seed location has been assigned to a super- or sub-threshold group, respectively, (4) classify the neighbor agreement between each location in the field and its nearest neighbors into a "discrepancy," "low-confidence," or "high-confidence" category, (5) if the pattern of neighbor agreement at any location is in the discrepancy category, retest that location for "seen"/"not-seen" response, and (6) if the pattern of neighbor agreement for non-seed location is in the low-confidence category, adjust the expected threshold of that location to a revised expected threshold value, and retesting the nearest- neighbor agreement between the revised-threshold location and its nearest neighbors.
13. The apparatus of claim 12, wherein the visual field array is a hemispherical surface cap region, and said presenting means includes a means for producing a selected- intensity light stimulus at selected locations in the surface cap region.
14. The apparatus of claim 13, wherein said producing means is a swinging-arm assembly having a light- stimulus source, and the assembly is designed to move the light source between any two locations in the cap region within a period of less than about 10 sec.
15. The apparatus of claim 14, wherein said presenting means includes a second light source which is fixed in the field, and means for focusing the two light sources coordinately for a subject being tested, and the apparatus further includes a transparent viewing screen through which the two light sources are viewed by the subject.
Non-Patent Literature (8)
- Heijl, A. "Computer Test Logics for Automated Perimetry," Acta Ophthalmoligica, 55:837, 3/1977.
- Johnson, C. A. "The Test Logic of Automated Perimetry", Acta: XXIV International Congress of Opthalmology, Henkind (ed.), J. B. Lipincott, Philadelphia, 1983.
- Johnson, C. A., Keltner, J. L., "Incidence of Visual Field Loss in 20,000 Eyes and its Relationship to Driving Performance," Arch. Opthalmol, vol. 101, pp. 371-375, Mar. 1983b.
- Johnson, C. A., Adamsm C. W., & Lewis, R. A., "Fatigue Effects in Automated Perimetry," Applied Optics, vol. 27, pp. 1030-1037, Mar. 1988.
- Keltner, J. L., Johnson, C. A., "Comparative Material on Automated and Semi-Automated Perimeters," Opthalmology, vol. 93, pp. 1-25, 1986.
- Lewis, R. A. et al., "Preliminary Clinical Trials with the Humphrey Field Analyzer," Doc. Opthalmol. Proc. Ser., vol. 42, pp. 159-165, 1985.
- Shapiro, L. R., Johnson. C. A., & Kennedy, R. L., "KRACKEN: A Computer Simulation Procedure for Static, Kinetic, Suprathreshold Static and Heuristic Perimetry," In A. Heijl (ed.), Perimetry Update 1988/89: Proceedings of the VIIIth International Perimetry Society Meeting. Asterdam: Kuglar & Ghedini, pp. 431-438, 1989.
- Tyrrell, R. A., & Owens, D. A., "A Rapid Technique to Assess the Resting States of the Eyes and Other Threshold Phenomena: The Modified Binary Search (MOBS)," Behavior Research Methods, Instruments & Computers, vol. 20, pp. 137-141, 1988.