Automated Diagnosis and Measurement of

Strabismus in Children

OREN YEHEZKEL, MICHAEL BELKIN, AND TAMARA WYGNANSKI-JAFFE

M

PURPOSE: Manual measurements of strabismus are EASURING THE TYPE AND MAGNITUDE OF
subjective, time consuming, difficult to perform in babies, ocular misalignment is essential for diagnosing
toddlers, and young children, and rely on the examiner’s strabismus, determining treatment, and
skill and experience. An automated system, based on eye following up.1 The conventional method for detecting
tracking and dedicated full occlusion glasses, was devel- the presence of manifest strabismus is the cover-uncover
oped to provide a fast, objective, and easy-to-use alterna- test (CUT). In addition, the conventional method for
tive to the manual measurements of strabismus. This measuring the total ocular deviation, combining manifest
study tested the efficacy of the system in determining and latent deviations when present, is the prism alternating
the presence of strabismus in children, as well as its cover test (PACT).2–6 The PACT’s disadvantages include
type and the amount of deviation, in addition to differen- its long test duration, the difficulty in performing the test in
tiating between phorias and tropias. young, uncooperative children, and the test’s dependence

DESIGN: A prospective, masked, inter-rater reliability on the examiner’s training, skill, and experience.
study. Consequently, this test is subject to high interexaminer

METHODS: A prospective, masked, cross-sectional variability.7–10
study included 69 children, 3-15 years of age. A cover- In order to develop standardization and reduce the
uncover test and a prism alternating cover test (PACT) variability of manual measurements, attempts were
for the primary gaze, at a distance of 50 cm, were made to develop objective automated strabismus mea-
performed by 2 independent, masked examiners and by surement devices. These include automated image anal-
the automated system. ysis software,11 videos based on infrared eye tracking,12

RESULTS: A high correlation was found between the video goggles to perform a Hess screen test,13 and binoc-
automated and the manual test results (R [ 0.9 and ular optical coherence tomography.14 Although good
P < 0.001 for the horizontal deviation, and R [ 0.91 agreement between the newly proposed technologies
and P < 0.001 for the vertical deviations, with 100% and the PACT measurements was reported, some tech-
correct identification of the type of deviation). The nologies are limited by the inability to incorporate spec-
average automated test duration was 46 seconds. The tacles with refractive error correction and others by the
Bland-Altman plot, used to compare the 2 measurement inability to block vision in all gaze positions or because
methods, showed a mean value of L2.9 prism diopters the participants’ cooperation is required; therefore, these
(PD) and a half-width of the 95% limit of agreement of technologies are unsuitable for use in young children.
±11.4 PD. None of the proposed methods could identify and differ-

CONCLUSION: The automated system provides precise entiate between phorias or latent eye deviations and
detection and measurements of ocular misalignment and tropias or manifest eye deviations. An eye tracker can
differentiated between phorias and tropias. It can be precisely detect saccadic eye movements, gaze position,
used in participants from the ages of 3 years old and fixation stability, smooth pursuit eye movements,
up. (Am J Ophthalmol 2020;213:226–234. Ó 2019 nystagmus, and alignment of the visual axis.15 Recently,
The Author(s). Published by Elsevier Inc. This is an a strabismus diagnosing system, based on eye tracking,
open access article under the CC BY-NC-ND license was developed; however, it is only able to detect the di-
(http://creativecommons.org/licenses/by-nc-nd/4.0/).) rection of the strabismus without quantifying the amount
of the deviation.12 This system also requires a pretest cali-
bration process, making it unsuitable for young children
and patients who are otherwise unable to undergo the
Supplemental Material available at AJO.com. conventional sensorimotor examination.
Accepted for publication Dec 11, 2019.
From NovaSight Ltd. (O.Y.), Airport City; Goldschleger Eye Research
The system described here is a newly developed objec-
Institute (M.B.), Sheba Medical Center, Tel Hashomer; Sackler Faculty of tive eye–tracking-based device that can automatically
Medicine (M.B., T.W-J.), Tel Aviv University, Tel Aviv; and Sheba detect the deviating eye and assess the extent and direction
Medical Center; Goldschleger Eye Institute (T.W-J.), Tel Hashomer,
Israel.
of both heterophorias and heterotropias. This study
Inquiries to: Tamara Wygnanski-Jaffe, Goldschleger Eye Institute, compared the performance of this eye–tracking-based test
Sheba Medical Center, Tel Hashomer, 52521, Israel; e-mail: Tamara. to those of the manual CUT and PACT methods.
[email protected]

226 © 2019 THE AUTHOR(S). PUBLISHED BY ELSEVIER INC. 0002-9394

https://doi.org/10.1016/j.ajo.2019.12.018
 SUBJECTS AND METHODS processing algorithms are then used to estimate the posi-
tion of the eye and the point of gaze.
PARTICIPANTS WERE PROSPECTIVELY RECRUITED FROM THE To perform the automated deviation measurement pro-
Goldschleger Eye Institute at the Sheba Medical Center, cedure, a dedicated pair of wireless glasses, controlled by
Israel (clinical trials government identifier MOH 2018- the system, was used to enable each eye to be covered sepa-
01-09001991). The study was approved by the Institutional rately. The participant wore the wireless glasses over pre-
Review Board of the Sheba Medical Center, Israel, and scription glasses, if required, and followed a visually
written informed consent was obtained from all guardians. stimulating target, presented on the screen, simultaneously
The study adhered to the ethical standards established by with audio augmentation. No language or verbal skills were
the Declaration of Helsinki for research involving human required. The first step in the procedure was to use an auto-
participants. Participants were consecutively recruited mated CUT to assess the presence of a tropia and, if pre-
from the hospital clinic, according to their appointment sent, to determine the deviating eye and the amount of
schedule, with no prior selection, until the required sample the deviation. The second step was to use an automated
size was obtained. All participants were screened by a pedi- ACT to measure the amount of total deviation and its di-
atric ophthalmologist before they entered the trial. Partic- rection, similarly to the manual PACT but without the
ipants were excluded if they did not match the inclusion physical presence of a prism. If a tropia was detected using
criteria, they were unable to cooperate, had low CUT, the deviating eye was reported (i.e., right, left, or
vision (< _20/200), or had abnormal anterior segment alternating). If a deviation was detected using only ACT
configuration. and not CUT, a phoria was reported, including the devia-
The testing distance of 50 cm was carefully controlled tion direction (i.e., esophoria, exophoria, hyperphoria, or
during both the automatic and the manual tests. The auto- hypophoria). In addition, if the ACT began with no tropia
mated testing distance was monitored constantly using 2 but a phoria developed during ACT and, by the end of
parallel and independent methods. The first method uses ACT, under binocular viewing conditions, a tropia was
the eye tracker’s ability to measure the distance of the measured, an intermittent deviation was reported. Once
eyes from the eye tracker camera positioned on the screen. the test was completed, the numerical and graphical results
The second method uses the designated marks on the full- of the deviation type, direction, and magnitude were
occlusion glasses that are part of the testing system; this also recorded.
enables accurate measurement of their distance from the
monitor. In addition, the operator’s screen indicates any
EYE-TRACKER SYSTEM: The eye–tracker-based system
change in the patient’s position. The orthoptists were includes the following components. 1) Tablet PC (Micro-
instructed to carefully monitor the testing distance soft, Redmond, Washington): Lenovo IdeaCenter AIO
throughout the test, using a 50-cm tape. 520-24IKL (60-Hz refresh rate with a resolution of
1,920 3 1,080 pixels) used for displaying targets, running

MANUAL EXAMINATION PROCEDURE: All participants the software, and for data storage. 2) Tobii with a sampling
underwent complete ocular assessment, orthoptic assess- rate of 90 Hz; EyeTracker 4C, Mississauga, Ontario, Can-
ment, and automated testing. All tests were performed ada. 3) Test controller: it displays the graphic user inter-
with the participants fixating on accommodative targets face, performs the test, and presents the test results. 4)
at 50 cm, using their habitual optical correction. Two ex- Occlusion glasses: these glasses occlude the vision of a par-
aminers (either 2 independent orthoptists or a pediatric ticipant’s eyes in order to simulate manual occluding, as
ophthalmologist and an orthoptist) performed the manual used in CUT and PACT. 5) Glasses controller: it allows
CUT and PACT on each participant. Both examiners were the wireless flow of information between the software and
masked to the results of the other examiner and to the auto- the glasses. The controller opens or occludes each lens ac-
matic test results. The two manual examinations and the cording to the software commands. 6) Testing distance: the
automated tests were performed during the same visit. eye tracker image measures the distance between the eye
tracker and the subject.

AUTOMATED EXAMINATION PROCEDURE: Automatic During the automated procedure (Supplemental
testing was performed twice by using the EyeSwift system Figure 1), a fixation target was presented to confirm that
(NovaSight Ltd., Airport City, Israel). This system is based the participant was observing the screen. The target dimen-
on eye tracking technology that provides continuous infor- sions were 1.2 3 1.6 cm (Supplemental Figure 2). For the
mation for the gaze position of each eye. Near-infrared illu- automated CUT, a short animated movie, with dimensions
mination is used to create reflection patterns from the 1.9 3 2.2 cm, containing accommodative fine details, was
cornea (glint) and pupil reflection; these reflections are presented (Supplemental Video 1). The test began with
detected by capturing images using an infrared camera. binocular viewing of the target (Supplemental Video 2).
Following detection, a vector formed by the angle between Once the first eye was covered, the algorithm searched
the cornea and the pupil reflections is calculated. Image for a movement in the noncovered eye, such a movement

VOL. 213 AUTOMATED STRABISMUS MEASUREMENT 227

 TABLE. Participant Characteristics

Eye-Tracking-Based
Eye-Tracking-Based Test APCT Horizontal Test Vertical APCT Vertical Right Eye Left Eye
Horizontal Deviation (PD) Deviation (PD) Deviation (PD) Deviation (PD) Prescription (D) Prescription (D) Age, y

Ortho deviation
<4 <4 0 0 7
<4 <4 0.00 to 1.00 3 105 þ0.50 to 1.25 3 60 6.5
<4 <4 þ2.00 to 1.50 3 90 0 6.5
<4 <4 0 0 5.5
<4 <4 þ0.50 þ0.50 12
<4 4.5 þ0.50 þ0.50 11
<4 6 0 0 4.5
<4 <4 0 0 8.5
Exo deviation
<4 <4 0 þ1.00 to 1.00 3 85 3
<4 <4 6 13 0.75 0.75 to 2.00 3 175 9.5
<4 <4 8.50 to 3.00 3 155 11.50 to 3.50 3 25 9.5
<4 <4 <4 5 0 0 3.5
<4 <4 0 0 3.4
<4 7 0 0 5
5 0 0 0 14
5 0 0 0 9
7 10 0 0 14
7 10 0 0 8.5
8 0 0 0 3
8 0 1.00 to 1.00 3 80 1.00 to 1.00 3 95 14
9 0 þ4.25 to 1.75 3 11 þ0.25 to 0.50 3 180 11
9 0 0 0 8
11 6 7.00 2.00 11.5
11 8 0.00 to 1.00 3 23 3.00 to 0.75 3 8 6
11 17 þ1.00 to 1.75 3 178 þ2.00 to 3.00 3 8 6
13 17 0 0 7
13 8 0 0 6.5
13 14 1.75 to 0.50 3 160 0.75 to 0.50 3 177 7.5
13 18 0 0 7.8
14 12 0 0 3
16 18 0 0 5.5
16 10 <4 5 þ0.25 to 0.50 3 70 þ0.75 to 0.50 3 70 9
18 24 1.00 1.00 8.5
19 16 0.50 0.50 6
20 18 þ0.75 to 0.75 3 180 þ0.75 to 0.75 3 30 9.5
20 17 0 0.25 11
20 20 0 0 7
23 30 þ0.75 to 0.75 3 176 þ0.75 to 0.75 3 176 9.5
27 21 0 0 6
30 25 0 0 8.5
31 31 <4 5 þ1.25 to 0.75 3 90 þ0.75 to 0.50 3 70 15
34 30 þ0.50 þ1.00 to 0.50 3 90 10
44 25 0.00 to 2.00 3 13 0.00 to 2.75 3 167 4.5
50 38 0 0 5
Esotropic deviation
<4 <4 7.5 6.5 þ3.25 þ4.50 6
5 0 þ2.50 to 1.75 3 3 þ3.00 to 2.00 3 180 10
5 11 þ3.00 þ3.00 7
8 6.5 0 0 7.5
8 9 þ0.50 þ1.75 to 1.00 3 2 7

Continued on next page

228 AMERICAN JOURNAL OF OPHTHALMOLOGY MAY 2020

 TABLE. Participant Characteristics (Continued )

Eye-Tracking-Based
Eye-Tracking-Based Test APCT Horizontal Test Vertical APCT Vertical Right Eye Left Eye
Horizontal Deviation (PD) Deviation (PD) Deviation (PD) Deviation (PD) Prescription (D) Prescription (D) Age, y

9 8 þ4.00 to 4.00 3 173 þ3.00 to 2.75 3 5 8

11 8 þ6.50 þ7.00 3
11 13 8 14 0 0 5.5
12 10 þ2.50 þ2.50 11.5
13 6 þ1.00 to 0.75 3 175 þ1.00 to 0.75 3 10 8
14 18 0 0 5.5
17 10 þ2.75 to 1.00 3 180 þ1.75 to 1.00 3 5 6.5
20 12 0 0 4
20 16 0 0 8
22 10 10 19 þ2.00 to 1.25 3 159 þ2.25 to 1.50 3 13 8
23 21 0 0 6
23 19 þ1.00 to 1.50 3 180 þ1.00 3.5
23 18 <4 6 0 0 5
27 30 0 0.25 to 1.25 3 5 5.5
28 30 þ5.75 þ6.00 6
30 25 0 0 8.5
35 40 1.00 0.50 13
44 39 0 0 4.5
49 43 0 0 6.5
50 31 þ0.25 to 0.50 3 162 0.50 to 2.00 3 23 3.4

APCT ¼ alternate prism cover test; D ¼ diopters; Eso ¼ esodeviation; Exo ¼ exodeviation; Ortho ¼ orthophoria; PD ¼ prism diopters.
Eye deviation metrics, stereo acuity (arc seconds), and demographics are listed for each patient. The first column represents the average of
the 2 measurements of the automated system. The second column represents the average of the 2 manual measurements in PD.

indicated the presence of a tropia. The procedure was then All tests were performed at a distance of 50 cm in the pri-
repeated for the fellow eye. For the automated ACT, 1 eye mary gaze position. Owing to the Eye Tracker’s accuracy of
was covered at a time, and the algorithm searched for a 1.00 prism dioptr (PD) and the target size presented on the
change in the fixation position of the noncovered eye. monitor, it is possible to detect artifact deviations in the
The eyes were alternately covered for 3 seconds at a time, range of 4 PD and below. In order to avoid reporting arti-
for a total of 30 seconds (i.e., repeated in 5 cycles), with facts in deviation measurements, eye deviations of 4 PD
no binocular viewing allowed between alternations. The and above are reported, and smaller deviations are regarded
resulting deviation between the eyes was recorded contin- as artifacts. The first manual test was used to determine
uously. Next, the maximal deviation between the gaze po- whether the participant meets the inclusion criteria. The
sitions of the 2 eyes was calculated and used to shift the order of the subsequent tests, that is, the second manual
position of the monocular targets accordingly. The proced- test and the automated test, were randomized to eliminate
ure was then repeated until no eye movement was detected. any bias in the results, such as fatigue.
Finally, the precise amount of the deviation was calculated
using the ratio of the maximal distance between the 2
OUTCOME MEASUREMENTS: Interexaminer variability
monocular targets on the monitor and the sitting distance. was calculated for the manual PACT and automated
The test duration depended on the ability of the participant ACT. Test-retest reliability and Pearson correlation coeffi-
to fixate to allow a sufficient amount of data to be collected. cients between the angles of deviation measured by the
The algorithm of the automated test is designed to identify different methods were analyzed.
fixation within a predefined region of interest in order to
confirm that the participant is looking at the target with
STATISTICAL ANALYSIS: Agreement between the
at least 1 eye. The measurement trial automatically pauses manual and automated tests was represented in Bland-
if the participant is not focusing within the target region of Altman plots and concordance correlation coefficients.
interest and automatically continues once refixation oc- Pearson correlation coefficient was obtained to measure
curs. In addition, the operator may observe the eye gaze po- the strength of the linear relationship between each test.
sition, which is continuously presented on the operator’s P values of <0.05 were considered statistically significant.
monitor throughout the test. A 2-way ANOVA was performed for group comparisons.

VOL. 213 AUTOMATED STRABISMUS MEASUREMENT 229

 RESULTS
SEVENTY-TWO SUBJECTS WERE PROSPECTIVELY RECRUITED
from the Goldschleger Eye Institute at the Sheba Medical
Center, Israel. Three patients were excluded from the study
due to their inability to complete the test, mistaken instil-
lation of cycloplegic eye drops prior to the test, or their
inability to fixate, inability to cooperate due to many fac-
tors including developmental delay, or were diagnosed
with mental health issues (Table). Sixty-nine eligible par-
ticipants were included, 3-15 years of age (mean 7.17 6
2.78 SD years of age). The participants had either congen-
ital or acquired forms of manifest or latent strabismus or no
deviation. Twenty-five had an esodeviation, 36 had an
exodeviation, 8 were orthophoric, and 8 exhibited a verti-
cal deviation.
In the first step, manual and automated CUT results were
compared. Nineteen participants had constant tropia, 15
had intermittent tropia, and 35 had no tropia (orthotro-
pia). One patient exhibited no eye movements during
the automated CUT and was diagnosed as orthophoric,
in contrast to the manual CUT, which detected a tropia.
Interexaminer variability for the manual PACT results
was tested. The Bland-Altman plot showed consistent vari-
ability across the graph. The mean value was 1.6 PD, and
the half-width of the 95% limit of agreement was 612.0
PD (Figure 1A). There was no overall tendency for the
values of one examiner to be higher or lower than the
values of the other.

HORIZONTAL DEVIATION: The PACT and automated

ACT results showed a highly significant positive correla-
tion (R ¼ 0.9; P < 0.002) (Figure 2). The Bland-Altman
plot, comparing the 2 measurement methods, is presented
in Figure 1B. The mean value was 2.9 PD, and the half-
width of the 95% limit of agreement was 611.4 PD. Com- FIGURE 1. Bland-Altman plots. Analysis of the agreement be-
tween 2 measurements within each method and between mea-
parison of the results showed no significant differences be-
surements using the 2 methods. (A) The difference between
tween the outcomes of the 2 tests F(1,272) ¼ 3.18; P ¼ 0.07 manual PACTs tested by 2 examiners plotted against the exam-
for the test effect; F(1,272) ¼ 0.33; P ¼ 0.56 for the repetition iners’ average of manual PACT 2. The upper and lower dotted
effect; F(1,272) ¼ 0.65; P ¼ 0.65 for the test 3 repetition lines represent the 95% limits of agreement; the solid line shows
interaction). Interdevice variability for the automated the mean of the differences. The mean value is 1.6 PD and the
ACT results was tested. The Bland-Altman plot showed half-width of the 95% limit of agreement is ± 12 pris diopters
consistent variability across the graph. The mean value (PD). (B) The difference between the manual PACT and the
was 0.2 PD, and the half-width of the 95% limit of agree- automated ACT plotted against the mean of the manual PACT
ment was 64.25 PD (Figure 1A). There was no overall ten- and the automated ACT. The upper and lower dotted lines repre-
dency for the values of one examiner to be higher or lower sent the 95% limits of agreement; the solid line represents the
than the values of the other. Comparison of the SD of the mean of the differences. The mean value is L2.9 PD and the
half-width of the 95% limit of agreement is ± 11.4. (C) The dif-
manual PACT and the automated ACT showed that the
ference between the repeated automated ACT plotted against the
repeatability of the automated test was significantly higher mean of the repeated automated ACT. The upper and lower
(P ¼ 0.008, paired t-test), with an almost 2-fold reduction dotted lines represent the 95% limits of agreement. The solid
in the average SD (11.3 6 0.5 vs 0.54 6 0.27, respectively) line represents the mean of the differences. The mean value is
for the automated ACT vs the manual PACT, respectively. 0.2 PD, and the half-width of the 95% limit of agreement is ±
Moreover, in the youngest subgroup, 3-5 years of age 4.25 PD. There is no overall tendency for the values of one
(n ¼ 16; 3.7 6 0.7 years of age), the PACT and automated test to be higher or lower than the values of the other. ACT [
ACT also showed a highly significant positive correlation alternating cover test; PACT [ prism alternating cover test.

230 AMERICAN JOURNAL OF OPHTHALMOLOGY MAY 2020

FIGURE 2. Correlation between manual PACT and automated measurements. (A) Horizontal eye deviations as measured by using
the automated eye-tracking-based test, individualized to the participant, are plotted against the measurements obtained using the
PACT. There is a high correlation between the 2 measurements (correlation coefficient R [ 0.9; P < 0.00001). (B) Same measure-
ment as in (A) for the vertical eye deviations (correlation coefficient R [ 0.9; P < 0.00001). PACT [ prism alternating cover test.

(R ¼ 0.9; P < 0.01), with no significant differences be- and compared to manual PACT results. The automated
tween the outcomes of the 2 tests: F(1,60) ¼ 1.94; P ¼ system identified the same direction of the deviation as
0.17 for the test effect; F(1,60) ¼ 0.42; P ¼ 0.52 for the repe- the manual PACT in all cases. The results of the automated
tition effect; F(1,60) ¼ 0.04; P ¼ 0.95 for the test 3 repeti- ACT showed excellent concordance with those of the
tion interaction. PACT for measuring eye deviation in all cases of horizontal
and vertical deviations.

VERTICAL DEVIATION: In addition to the horizontal de- Despite the fairly good agreement between the 2 manual
viation, 8 participants exhibited a vertical deviation using examiners in the current study, the PACT relies on subjec-
the automated ACT and 5 using the manual PACT. One tive interpretation and the examiner’s skill in observing
participant, with Brown syndrome, exhibited no vertical very small ocular movements while performing the test.8
deviation with the automated ACT, although a hypodevi- Thus, the accuracy of measurements depends largely on
ation was found using the manual PACT. Another partic- the examiner’s expertise and the participant’s level of coop-
ipant exhibited a vertical deviation with the automated eration.7–10,16,17 Studies showed a significant interexa-
ACT, which was not detected by the manual PACT. miner variability in PACT among highly experienced
There were no significant differences between the out- examiners, ranging between 6.9 and 12.5 PD.18 In this
comes of the PACT measurements performed by the 2 study, the interexaminer variability of the automated test
masked examiners: F(1,224) ¼ 2.94; P ¼ 0.08 for the test ef- was 4.25 PD, which is much lower than the variability of
fect; F(1,224) ¼ 0.001; P ¼ 0.96 for the repetition effect; the manual test, which was 11.95 PD. However, it is note-
F(1,224) ¼ 0.01; P ¼ (0.89 for the test 3 repetition interac- worthy that the 2 automated measurements were
tion). Comparison of repeatability showed an almost 2-fold performed consecutively, whereas the manual tests were
reduction in the SD for the automated test versus the not necessarily performed consecutively, which may
ACPT (0.05 6 0.03 vs 0.12 6 0.06, respectively), which explain the differences in the SDs in favor of the automated
did not reach significance due to a large portion of zero ver- test.
tical deviation (P ¼ 0.12, paired t-test). Of the 7 cases where the automated system detected an
The correlation between the automated and manual eye deviation, whereas the manual tests showed no eye de-
tests was 0.9 for esodeviations, 0.88 for exodeviations, viation, in 4 cases at least 1 of the examiners detected an
and 0.91 for vertical deviations. In orthophoric partici- eye deviation, whereas the second examiner reported no
pants, no deviation was recorded by the automatic system. eye deviation. Therefore, the average measurement be-
The average test duration of the automated ACT was 46.00 tween the 2 examiners indicated an eye deviation below
6 3.45 seconds (a range of 42-69 seconds). the threshold of 4 PD. In the other 3 cases, the 2 examiners
reported no eye deviation, whereas the automated system
repeatedly measured a deviation. It is highly unlikely to
obtain such repeatable results in the 2 automated measure-
DISCUSSION ments when no eye deviation exists (4 and 5, 8 and 8, 9
and 9). Of the 3 cases in which the manual examiners
IN THIS STUDY, A NOVEL AUTOMATED SYSTEM FOR detected an eye deviation and the automated system
measuring ocular misalignment, based on an EyeTracker detected a deviation smaller than 4 PD, in all cases at least
and a liquid crystal display active shutter glasses, was tested 1 of the examiners detected a deviation of 5 PD, which is

VOL. 213 AUTOMATED STRABISMUS MEASUREMENT 231

very similar to the threshold of 4 PD (5 and 5, 5 and 7, 5 and advantage of this test is that it can be used in patients
8), whereas the automated system detected a deviation of with visual suppression and who are not suitable for the
just below the threshold of 4 PD. In one case, 1 of the auto- Hess screen test. It was also reported that binocular anterior
mated results was just above the threshold (i.e., 4.5 PD), segment optical coherence tomography is suitable for
whereas the repeated measurement result was just under measuring the total amount of eye deviation. Importantly,
the threshold; therefore, their average was lower than the it has the ability to detect subtle differences in the stra-
threshold. In the 2 cases of large deviations, a discrepancy bismus size that may not be visible to the naked eye; how-
existed between the manual and the automated test results; ever, it is unable to determine whether the deviation is a
however, the difference between the repeated measure- phoria or tropia, and only a spherical equivalent can be
ments within each of the tests was small. In the case of a used for optical correction during the test.14
3.4-year-old participant, the 2 repetitions of automated The automated system has several potential advantages
measurements detected a deviation of 52 and 48 PD versus compared to the manual PACT. The automated system
26 and 35 PD, respectively, which were detected by the 2 provides measurements of the latent component of eye de-
examiners, resulting in an average result of 50 PD in the viation; however, this is possible only after disruption of
automated test vs 31 PD in the manual test. This may indi- fusion. The test procedure is similar to that of the PACT,
cate that the patient had a combination of recurrent esotro- but instead of using prisms to shift the target toward the
pia after strabismus surgery, convergence excess, a habitual eye’s position, the target is shifted on the screen. In addi-
bifocal prescription, and developmental delay with limited tion, the automated system does not require eye tracker
cooperation. In the second case, involving a 4.5-year-old calibration, in contrast to the other automated tests, which
participant, the 2 repetitions of automated measurements depend on calibration.12 Eye Tracker calibration is a pro-
detected a deviation of 43 and 44 PD versus 20 and 30 cess during which the geometric characteristics of a partic-
PD, which were detected by the two examiners, resulting ipant’s eyes are estimated as the basis for fully customized
in the average results of 44 PD in the automated test versus and accurate gaze point calculation. In the method
25 PD in the manual test. In one case of Brown syndrome, described here, the Eye Tracker is not directly used for
the participant did not exhibit any vertical eye movements detecting the eye’s position but rather for determining
during the automatic test, but hypodeviation was detected whether the eyes move to a target presentation at different
by the manual PACT. It is possible that this occurred due locations. The exact deviation is calculated by combining
to an anomalous head posture that the participant adopted the distance obtained between the 2 monocular targets
during the automated test. on the monitor in relation to the sitting distance. Hence,
Several objective automated ocular deviation measure- it is suitable for testing participants who have extreme
ment devices have been reported. These include automated ocular dimensions that deviate from normal variation,
image analysis software, video-based infrared eye tracking, such as high refractive errors. The test can be performed
video goggles to assess the Hess screen test,13 and binocular with or without optical correction, including spectacles
optical coherence tomography to evaluate strabismus in or contact lenses. An additional advantage of the auto-
the primary position.14 Another attempt to estimate the mated system is that the full occlusion glasses allow
binocular alignment from photographs acquired using a se- blockage of the patient’s vision but are transparent to the
lective wavelength filter and an infrared video camera was eye tracker; therefore, the gaze position of both eyes can
also reported.11 In that system a unique filter was used as an be monitored continuously during the entire test. This
occluder for blocking the participant’s view while selec- noninvasive test provides automatic results that do not
tively transmitting the infrared light, followed by analysis require expert operation or interpretation; therefore, these
of the corneal reflex position, which revealed the latent de- results can potentially be widely implemented in standard
viation. Although good agreement with the PACT was re- clinical practice by technicians having a lower skill level.
ported, this technology is limited by the need for normative The automated test takes up to 1 minute to complete,
ophthalmic biometry. Furthermore, using a nonaccommo- which is advantageous in children and patients with
dative (a spot of light) target and asymmetric kappa angles limited levels of cooperation.20 Moreover, in communities
increased the susceptibility to errors. Another recent pilot in which strabismus specialists or orthoptists are unavai-
trial using a digital video analysis methodology that inter- lable, the automated test provides a telemedicine solution
prets shadowing related to room light, used for detecting for detecting and evaluating strabismus, regardless of the
and diagnosing exotropia, was compromised by room light geographical location. The test can be performed by
conditions unsuitable for dark eyes.19 Furthermore, the personnel in any clinical setting, and the results can be re-
validity of strabismus testing using video goggles was ported to a remotely located specialist. It is known that the
recently assessed in comparison with the Hess screen test. measurement variability of the angle of strabismus may
A high agreement was found between the tests in all tested vary over time due to physiological factors, such as fatigue
gaze positions, but both are unsuitable for participants un- and the patient’s anxiety level during the examination.21
der the age of 6 years (the authors reported that the soft- Fast, automated measurements allow for frequently
ware and hardware still need refinement).13 The repeated testing, thus improving measurement accuracy

232 AMERICAN JOURNAL OF OPHTHALMOLOGY MAY 2020

and decreasing the influence of intrapatient errors. More- already includes measurements at 20 feet. The tested proto-
over, the algorithm of the automated test is designed to type version of the device could not distinguish dissociated
perform 10 repeated measurements in 5 cycles of inter- vertical deviation (DVD) from vertical hypertropia. In the
changing right-left eye occlusion, and only once was the next version, the system will identify cases of monocular
vertical deviation consistently detected in all repetitions; and binocular DVD. A limited level of cooperation is
with a consistent pattern of the deviating eye movement, required, for example, focusing on an engaging animated
it is registered as a vertical deviation. This prevents arti- target for a test duration of 46 seconds (ranging from 42-
facts of eye movements reported as a vertical deviation. 69 seconds). Deviations of 4 PD and above were reported.
This prevents artifacts of eye movements from being re- The current results show the possible use of the automated
ported as a vertical deviation. An additional benefits of test in children as young as 3 years of age, but further studies
the automated system are that prisms greater than 20 PD with a larger sample in this particular age group are needed.
are usually available only in step differences of 5 PD, To conclude, the automated system reported here mea-
whereas the automated system is suitable for measuring a sures strabismus based on a procedure similar to that of
full range of deviation in steps of 1 PD. the manual PACT test, while continuously monitoring
The eye–tracker-based method is based on eye move- the gaze position of each eye, and then calculates the dis-
ment detection; therefore, similar to the manual PACT, tance between the 2 images presented on the screen after
is it not feasible in patients with extraocular muscle paral- eye movements have been completely neutralized. This
ysis. For the same reason, this method cannot be applied to method may assist in detecting and quantifying eye devia-
patients with a large amplitude of nystagmus, as the system tions. It may be used in locations in which skilled eye spe-
may become misled by the repetitive, uncontrolled eye cialists or orthoptists are unavailable or in settings with a
movements. The automated system reported here was not large volume of patients. It can be used before or after sur-
designed to measure torsion. This study focused on gical or optical treatment, for monitoring patients and for
measuring horizontal and vertical eye deviation at 50 cm screening purposes. Since the system is entertaining and re-
and in the primary gaze; however, the next version of the quires no head constriction, it is suitable for children as
system will evaluate eye positions in 9 gaze positions and young as 3 years of age.

ALL AUTHORS HAVE COMPLETED AND SUBMITTED THE ICMJE FORM FOR DISCLOSURE OF POTENTIAL CONFLICTS OF INTEREST
and none were reported.
FUNDING/SUPPORT: This study was supported by NovaSightLtd. The testing device was borrowed from NovaSight to carry out this study.
Financial Support: O.Y. is an employee of NovaSight Ltd. M.B. is a consultant for and a scientific advisory board member of NovaSight Ltd; O.Y. and
M.B. are patent holders of ‘‘A System And Method For Measuring Ocular Motility’’; US patent 2018/0028057. T.W.J. has received financial support for
conducting this study. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose. O.Y. was respon-
sible for conceptualization, methodology, formal analysis, resource writing (original draft), software, data curation, and investigation. M.B. was responsible
for conceptualization methodology resources writing-reviewing and editing, supervision. T.W-J. was responsible for methodology, supervision, investiga-
tion resource writing (original draft), and investigation.

REFERENCES 9. Group PEDI. Inter-observer reliability of the prism and alter-

nate cover test in children with esotropia. Arch Ophthalmol
1. Von Noorden GK, Helveston EM. Strabismus: A Decision 2009;127:59.
Making Approach. St. Louis: Mosby; 1994. 10. Hatt SR, Leske DA, Liebermann L, Mohney BG, Holmes JM.
2. Grosvenor T, Grosvenor TP. Primary Care Optometry. St. Variability of angle of deviation measurements in children
Louis: Elsevier Health Sciences; 2007. with intermittent exotropia. J Am Assoc Pediatr Ophthalmol
3. Gamble JD. Identifying deviations by the cover test. Opt J Rev Strabismus 2012;16:120–124.
Optom 1950;87:31. 11. Yang HK, Seo J-M, Hwang J-M, Kim KG. Automated analysis
4. McKean HE, Wirtschafter JD, Marx D. Bias of the cover test of binocular alignment using an infrared camera and selective
in the diagnosis of alternating tropia. Ann Ophthalmol 1976;8: wavelength filterimage analysis of strabismus with infrared fil-
435–437. ters. Invest Ophthalmol Vis Sci 2013;54:2733–2737.
5. Romano PE, Von Noorden GK. Limitations of cover test in 12. Chen Z, Fu H, Lo W-L, Chi Z, Xu B. An eye-tracking aided
detecting strabismus. Am J Ophthalmol 1971;72:10–12. digital system for strabismus diagnosis. Healthc Technol Lett
6. Scott AB. Strabismus: beyond the cover test. Invest Ophthal- 2018;5:1–6.
mol Vis Sci 1973;12:719–720. 13. Weber KP, Rappoport D, Dysli M, et al. Strabismus measure-
7. Schutte S, Polling JR, van der Helm FCT, Simonsz HJ. Human ments with novel video goggles. Ophthalmology 2017;124:
error in strabismus surgery: quantification with a sensitivity 1849–1856.
analysis. Graefes Arch Clin Exp Ophthalmol 2009;247:399. 14. Chopra R, Mulholland PJ, Tailor VK, Anderson RS,
8. Holmes JM, Leske DA, Hohberger GG. Defining real change Keane PA. Use of a binocular optical coherence tomography
in prism-cover test measurements. Am J Ophthalmol 2008;145: system to evaluate strabismus in primary position. JAMA
381–385. Ophthalmol 2018;136:811–817.

VOL. 213 AUTOMATED STRABISMUS MEASUREMENT 233

15. Holmqvist K, Nyström M, Andersson R, Dewhurst R, cover test (APCT) measurements of strabismus performed
Jarodzka H, Van de Weijer J. Eye Tracking: A Comprehen- by 4 examiners. Strabismus 2014;22:158–166.
sive Guide to Methods and Measures. Oxford, UK: Oxford 19. Valente TLA, de Almeida JDS, Silva AC, Teixeira JAM,
University Press; 2011. Gattass M. Automatic diagnosis of strabismus in digital
16. Hrynchak PK, Herriot C, Irving EL. Comparison of alternate videos through cover test. Comput Methods Programs Biomed
cover test reliability at near in non-strabismus between expe- 2017;140:295–305.
rienced and novice examiners. Ophthalmic Physiol Opt 2010; 20. Yehezkel O, Spierer A, Oz D, Yam R, Belkin M, Wygnanski-
30:304–309. Jaffe T. An objective rapid system based on eye tracking for
17. Simonsz HJ, Kolling GH, Unnebrink K. Final report of the early eye deviation measurement in children and adults. Invest
vs late infantile strabismus surgery study (ELISSS), a controlled, Ophthalmol Vis Sci 2018;59:1024.
prospective, multicenter study. Strabismus 2005;13:169–199. 21. Yang HK, Hwang J-M. The effect of target size and accommo-
18. De Jongh E, Leach C, Tjon-Fo-Sang MJ, Bjerre A. Inter- dation on the distant angle of deviation in intermittent
examiner variability and agreement of the alternate prism exotropia. Am J Ophthalmol 2011;151:907–913.

234 AMERICAN JOURNAL OF OPHTHALMOLOGY MAY 2020