Research Article
Using Eye Tracking to Assess Reading Performance in
Patients with Glaucoma: A Within-Person Study
Nicholas D. Smith, Fiona C. Glen, Vera M. Mönter, and David P. Crabb
Division of Optometry and Visual Science, School of Health Sciences, City University London, London EC1V 0HB, UK
Correspondence should be addressed to David P. Crabb; david.crabb.@city.ac.uk
Received December ; Accepted March ; Published May
Academic Editor: Stefanie I. Becker
Copyright © Nicholas D. Smith et al. is is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.
Reading is oen cited as a demanding task for patients with glaucomatous visual eld (VF) loss, yet reading speed varies widely
between patients and does not appear to be predicted by standard visual function measures. is within-person study aimed to
investigate reading duration and eye movements when reading short passages of text in a patient’s worse eye (most VF damage)
when compared to their better eye (least VF damage). Reading duration and saccade rate were signicantly dierent on average
intheworseeyewhencomparedtothebettereye(𝑃 < 0.001) in patients with glaucoma that had median (interquartile range)
between-eye dierence in mean deviation (MD; a standard clinical measure for VF loss) of . (. to .) dB; dierences were not
related to the size of the dierence in MD between eyes. Patients with a more pronounced eect of longer reading duration on their
worse eye made a larger proportion of “regressions” (backward saccades) and “unknown” EMs (not adhering to expected reading
patterns) when reading with the worse eye when compared to the better eye. A between-eye study in patients with asymmetric
disease, coupled with eye tracking, provides a useful experimental design for exploring reading performance in glaucoma.
1. Introduction
Glaucoma is a leading cause of visual impairment and aects
a signicant number of the elderly populations []. e con-
ventional view of vision loss in glaucoma suggests disruption
of peripheral vision and minimal impact on tasks that require
good central vision, like reading. However, patients with
glaucoma regularly self-report diculties with reading [–
]. Furthermore, evidence is emerging from experimental
studies showing that some patients with glaucoma have
impaired reading performance when compared to their
visually healthy peers. ese impairments are particularly
evident for patients with advanced or bilateral visual eld loss
[–] when reading small size text []; when reading text
at low contrast []; or when reading for sustained periods
of time []. However, not all patients displayed reduced
reading speeds in these studies, with some patients appearing
to be much more aected than others. A limiting feature of
the studies that have generated these results is that reading
speed, as an experimental outcome measure, is subject to
much between-person variability: it is very dicult to isolate
the impact of the glaucomatous visual eld loss from all
the other factors, such as age, visual acuity, and cognitive
andreadingability,thatmightcontributetoslowerreading.
Furthermore, dierences in eye movement patterns may
also inuence reading speed. Eye movements supplement
information about how long a person takes to read, by
giving insight into how they are reading. Previous research
has considered eye movements in patients with glaucoma
compared to visually healthy controls when carrying out a
numberofothervisualtasks,suchasvisualsearch[], face
recognition [], viewing of photographs [], and watching
of driving videos []. In these studies, patients sometimes
displayed dierent eye movement patterns on average to
controls, although it was suggested that some patients may
“adapt” their eye movements in ways that enable them to
function better in the task [, ]. However, the case-control
design that featured in all these studies again made it dicult
to discern the nature of the contribution of visual eld loss to
changes in eye movement behaviour.
As yet no studies have considered performing a within-
person, or between-eye, reading study to examine the impact
Hindawi Publishing Corporation
Journal of Ophthalmology
Volume 2014, Article ID 120528, 10 pages
http://dx.doi.org/10.1155/2014/120528
Journal of Ophthalmology
of glaucomatous visual eld loss on reading performance:
the idea here would be that a more damaged eye could
be compared with a less aected fellow eye. An experi-
mental design such as this might proer advantages over
studies comparing patients to controls, where large numbers
of people are needed to demonstrate eects. In addition,
experimental studies of reading speed in glaucoma have been
constrained to those where reading “out loud” or timed silent
reading is simply the main, or only, outcome measure. One
recent study incorporated eye tracking when investigating
reading performance in glaucoma []: the ndings of that
case-control study, which measured the maximum and min-
imum sizes of eye movements made during a reading task
by patients compared to controls, hinted that glaucoma may
lead to some alterations in xation behaviour. However, to
date, no studies have used an eye tracker to measure more
task-specic saccades (i.e., rapid eye movements occurring
between locations on the text) to tease out the eects
thatmightresultfromglaucomatousvisualeldlosswhilst
reading short passages of text.
In this study, we explore the usefulness of comparing
monocular reading performance in patients with asymmetric
glaucomatous visual eld loss. e study measures reading
performance using eye tracking whilst participants silently
read very short passages of text. Our main hypothesis is that
patients will take longer to read short passages of text in what
is considered to be their worse eye (most visual eld damage)
when compared to their better eye (least visual eld damage);
weaimtodothisinjustasmallsampleofpatientsinorder
to demonstrate the eectiveness of the experimental design.
We also, as a secondary aim, test the idea of determining
dierent types of reading-specic saccadic eye movements,
in an automated fashion, specically eye movements that
occur in a forward direction (forward saccades), saccades
that “backtrack” over previously read text (regressions), those
that occur between the end of one line and the beginning
of the next (line change saccades), and eye movements that
do not t expected patterns (unknown saccades). Next we
investigate if any of these measurements from this automated
approach are associated with the size of between-eye decits
in standard measures of visual function.
2. Methods
Participants were recruited from a database of patients that
had taken part in previous studies conducted at City Uni-
versity London [, ].Allpatientshadaclinicaldiagnosis
of primary open angle glaucoma and had no other ocular
diseases. Patients were contacted if they had previously
presented with asymmetric visual eld loss between eyes as
measured using a central - SITA Standard Test on the
Humphrey Visual Field Analyzer (HFA, Carl Zeiss Meditec,
CA, USA). is was quantied by considering the HFA mean
deviation (MD); this summary measure expresses the average
reduction in the visual eld relative to a group of visually
healthy age-matched observers []. Participants were only
invited to the study if the MD diered by more than dB
between eyes. is value represents a clinically signicant
dierence as used in staging schemes for visual eld severity
[].
e study was approved by the Ethics Committee for
the School of Health Sciences, City University London.
All participants gave their informed consent and the study
conformed to the Declaration of Helsinki.
2.1. Standard Vision Testing. Fourteen patients were recruited
and all testing was carried out on one day. Visual acuity (VA)
as measured with the Early treatment diabetic retinopathy
study (ETDRS) chart and contrast sensitivity as measured
with the Pelli-Robson chart (PR Log CS) were assessed
monocularly. Astigmatic error was less than ±. dioptres in
all those recruited. Visual eld tests (central - and -
SITAStandard)wereconductedineacheyeusingaHFA.On
testing (central -), two of the patients had a between-
eye MD dierence of less than dB (. and . dB). We
decided that these patients should still be included in the
study. From this point we dene the patient’s eye with the
worse VF damage (worse MD) to be the “worse eye” and the
felloweyetobethe“bettereye.”
e reading experiment was performed on a cm CRT
computer monitor displaying at a resolution of by
pixels and a refresh rate of Hz (Iiyama Vision Master PRO
, Iiyama Corporation, Tokyo, Japan). Participants were
seated (with a head rest) in front of the computer screen.
Each participant was tted with a set of trial frames with
the appropriate refractive correction. One eye was randomly
selected and then occluded by inserting a blackout lens into
the trial frames. Participants were then presented with
dierent texts (trials) on the screen, one at a time, and
wereaskedtosilentlyreadthem“asquicklyandaccurately
as possible.” Once the participants had read the texts,
they had a short break before repeating the task using their
alternate eye with novel texts. Participants read the same
texts but in a randomised order. Each text consisted of
one sentence, distributed over two lines, using the “Courier
New” font at size in which each letter subtended a
maximum height of .
∘
visual angle and a constant width
of .
∘
. e standardised passages of text had an average
Flesch-Kincaid readability score of . and were the same as
those used by Kabanarou and Rubin []. e background
brightness was . cd/m
2
and the text was displayed at
. cd/m
2
. Each paragraph subtended
∘
width and
∘
in
height.
Eye movements were recorded simultaneously during
the reading task using an EyeLink (SR Research Ltd.,
Mississauga, Ontario, Canada) which was set to record the
participant’s eye location at Hz. It is claimed that the
EyeLink measures at an average accuracy of better
than .
∘
. e saccade detection thresholds were dened
by a velocity greater than
∘
/s and acceleration above
∘
/s
2
. Before the study commenced, a calibration was
performed and had to be classied as a “good” standard
as set by the instrument. Furthermore, between each trial
(each displayed sentence) a dri check was performed and,
if a substantial dri had occurred, a recalibration would be
carried out.
Journal of Ophthalmology
2.2. Analysis of Eye-Tracking Data. To prepare the eye move-
ment data for analysis, we developed a novel preprocessing
technique. ese methods adjusted for calibration errors in
the eye tracking and ensured that only those saccades relevant
to the reading task were included. Secondly, we report a
novel method of classifying reading-specic eye movements
according to their saccade type, that is, whether they occurred
from le to right (forward saccade), right to le (regression),
or between lines (line change) or did not conform to expected
reading patterns (unknown saccade). We report both of these
methods here as they may be relevant to other studies using
eye tracking to measure reading performance. Note that the
techniques described below do not require information about
the specic content of the underlying text, such as details of
thewordsandcharacters,butonlythelocationsofthestart
andendofthetext.
2.3. Preprocessing. Data from the eye tracker was used to
determine reading duration for each trial in addition to iden-
tifying the key eye movement patterns made whilst reading
the texts. e eye tracker was running before the display of
each text in order to ensure that all eye movements were
recorded, meaning that it was highly likely that some addi-
tional eye movements were made prior to beginning to read
each sentence that were irrelevant to the task. Furthermore,
the dri correction carried out before each trial meant that
the participant always began the trial by xating in the middle
of the screen, therefore introducing bias into subsequent eye
movement recordings. It was therefore necessary to pinpoint
the exact points at which the person actually began reading
the sentence and the point at which they nished reading.
Use of an automatic real-time start and end point has the
potential to misidentify when the person started or nished
reading, as this technique uses xed points on the screen
and therefore assumes perfect calibration of the eye tracker.
To address this issue, a novel “preprocessing” method was
therefore implemented and is reported in detail here because
it may be of use in other eye-tracking experiments. Some
examples of preprocessed scanpaths are shown in Figure (a),
showing additional saccades that occurred before and aer
the patient read the passage.
e rst stage of the preprocessing algorithm attempted
to correct any rotational errors in the eye movement data.
As the text was displayed centrally, small errors in edge
calibration were not of huge concern for this particular task;
however inspection of scanpaths revealed that data some-
times appeared to be rotated along the centre. To correct this,
it was assumed that all small saccades running ±
∘
along
the horizontal (approximation of reading between words)
should be corrected to correspond with the angle of the text
(average angle of horizontal or
∘
). erefore, the circular
median of all these ±
∘
angles of the saccades was calculated
per trial, and all saccades were rotated (corrected) by this
amount. Visual analysis of scanpaths also conrmed that,
on being rst presented with a text, participants sometimes
made several involuntary eye movements at locations on the
screen that were irrelevant to the task itself, before adjusting
their gaze position so that they could start reading from
the beginning of the sentence. In order that the analysis
would only include those eye movements that were relevant
to the task, an automated procedure was developed that
determined which eye movements coincided with the text’s
start and end location, thereby ltering out all other irrelevant
eye movements. is process involved a series of steps to
identify the start and end point locations signalling the start
and end point of reading each text. e standard preset SR
Research EyeLink parser (edfasc) results in sharp downward
movements being recorded at the point just before the pupil
disappears (i.e., during a blink). Sharp downward saccades
do not correspond with reading, so these were identied
and excluded specically any saccade with an amplitude
>
∘
and with an angle of between
∘
and
∘
.Next,we
aimed to detect the starting point of the saccade nearest to
therstwordofthetextandtheendpointofthesaccade
closest to the nal word of the text. However, this procedure
wascomplicatedbythefactthatthetextwasrectangular
in shape, with the height being substantially smaller in size
than the width, a factor that would bias end point detection.
For instance, the end point of a saccade made at the end
oftherstlineoftext(i.e.,toprightofthetext)could
be incorrectly classied as being nearer to the end of the
text than a saccade made on the line below. We therefore
normalised the locations of the saccade start and end points
in order to make the axes equal. Specically, the Euclidian
distance from (, ) (top le) for each saccade start point
and the distance from (, ) (bottom right) for each saccade
end point were calculated, creating two sets of distances.
An exponential weighting was applied to these two sets of
distances. As such, the more the distance value increases the
further the point is from the start location. e start saccade
wasthenselectedastheminimumdistancefrom(,)once
the weights have been applied. e purpose of this procedure
was to “encourage” the algorithm to select the rst element in
the set as the start of the sentence; however if, for example, the
distance of the rst saccade’s start point is larger than another
saccade, the smallest distance from (, ) will be selected to
be the start point. To select the end point, the same process
is applied to the saccade end points, except that the weights
arereversedto“encourage”thealgorithmtochoosethenal
value. An example of this process can be seen by viewing
Figure , Participant : when viewing the raw scanpath in
column (a) and the processed path in column (b), it can be
observed that two points are a similar distance from (, ).
Using the weighting, the algorithm is “encouraged” to choose
the earlier point as the cut-o.
e reading duration was then dened as the time
between the start of the rst saccade and the end of the
nal saccade (the rotation and the reading extraction stages
are shown in Figure (b)). Once this was complete, any trial
shorter than ms or less than saccades per second was
excluded as it is likely the trial was of poor quality.
2.4. An Automated Algorithm for Classifying the Reading
Eye Movements. Eye-tracking soware typically expresses
data with general measures, such as the size (amplitude) or
location of each saccade. However, in tasks such as reading,
Journal of Ophthalmology
Subject
Visual eld
1
2
3
4
(a)
(b) (c)
F : Four examples of scanpaths from four dierent glaucoma patients with their visual elds on the le. e start and end of each
saccade are represented by a circle. Column (a) shows the original scanpaths made by the four participants reading the text. Column (b)
shows the scanpath aer the rotation has been corrected and reading-specic saccades have been extracted using the preprocessing algorithm.
Column (c) shows the scanpath results from the clustering and classication algorithm. e number represents the order in which the saccades
occurred, and the colours represent the classication that was attributed to them by the automated clustering algorithm (blue: forward saccade,
green: between line saccade, red: regression, and brown: unknown).
the properties of each saccade will vary according to the
demands of the task. For instance, when reading, a person
will make small forward saccades (from le to right). It is
alsocommonforpeopleto“backtrack”torereadprevious
sections (referred to as a “regression”). e properties of
a saccade occurring between the end of one line and the
beginningofthelinebelow“linechange”willagaindier.
Finally, readers may also make saccadic eye movements
that do not conform to expected patterns (unknown). For
this experiment we developed an automated data analysis
algorithm for classifying the types of saccade made during
the task. Again we provide details of this method because
it may be of use in other eye-tracking experiments. At
the centre of this technique is a Gaussian mixture model
that mines for clusters in the data. is approach was only
possible due to the type of texts used, where line length was
consistent throughout, giving predictable expected saccade
angles and similar amplitudes per person. Specically, the
information needed to classify the eye movements is the
amplitude (in degrees) of each saccade and the angle of
each saccade, for all sentences (trials) read by the “better
eye” in each person. Next, it is necessary to acknowledge
that the angle of eye movements occurring in a forward
direction (from le to right) will occur at an average of
∘
;
for example, some forward saccades could occur at
∘
and
others at
∘
. e discrepancy between these values, whilst
indicating the same saccade type, will subsequently inuence
the success of the classication algorithm by yielding two
separate clusters that actually give the same information. To
avoid having to use circular statistics to compensate for such
a scenario, we adjusted all angle values by −
∘
,meaningthat
standardstatisticalmethodscouldbeused(Figure shows an
example of this procedure in action, whereby the blue forward
saccades are now located at approximately
∘
). e Netlab
pattern analysis toolbox []GaussianMixtureModelwas
then used to determine four clusters with predened start
points and priors (approximate proportion of points that each
cluster contains). Using this method, eye movements made by
Journal of Ophthalmology
30
25
20
15
10
5
0
0 50 100 150 200 250 300 350
Angle (
∘
)
Amplitude (
∘
)
(a)
30
25
20
15
10
5
0
0 50 100 150 200 250 300 350
Angle (
∘
)
Amplitude (
∘
)
(b)
30
25
20
15
10
5
0
0 50 100 150 200 250 300 350
Angle (
∘
)
Amplitude (
∘
)
(c)
30
25
20
15
10
5
0
0 50 100 150 200 250 300 350
Angle (
∘
)
Amplitude (
∘
)
(d)
F : Scatterplots showing the amplitude and angle of saccades made across the sentences for four examples of patients reading
with the better eye. is data is used by the GMM to detect the four clusters within the data that represent the type of saccades made by
the patients. e types of saccade are represented by the colours green (line change saccade), red (regression), blue (forward saccades), and
brown (unknown). e black cross represents the start point for the GMM for each of the four clusters. e small circle represents the centre
of the cluster and the surrounding larger ellipse represents a distribution of the data (calculated to be standard deviations) captured by that
cluster following the GMM process. Examples of outcomes from the GMM clustering are shown in Figure (c) for four dierent patients.
the better eye were grouped into four clusters, representing
regressions, lines changes, forward saccades, and unknown
saccades (Figure ). Data yielded when reading with the
worseeyewasthenclassiedinthesameway,sothatthe
proportion of saccades that fell into each of the four clusters
couldbecalculatedandcompared.
2.5. Data Analysis. A linear mixed eects ANOVA was
performed in R [] using the linear and nonlinear mixed
eects models (nlme) package to assess dierences in the
average reading duration and saccade rate between patients’
worse and better eyes. A mixed eects model was chosen
since dierent sentences were viewed by the worse and better
eye. e random eect was set as the patient. e ANOVA
was performed to test the null hypothesis, for each response,
that the means for the patients’ worse and better eyes are the
same.
For each eye we also calculated the percentage of eye
movements that were automatically classed as the four types
of saccade by the classication algorithm, namely, “forward
saccades,” “regressions (backwards saccades),” “between line
(line change) saccades,” and “unknown” across the trials
read by the better and worse eye, respectively. Statistical
dierences in these proportions between the worse and better
eye were then assessed (Wilcoxon’s test).
To investigate whether the magnitudes of the change in
the key measured variables between eyes for each person
were important, we next calculated the dierence between
Journal of Ophthalmology
T : Descriptive statistics (median and interquartile range [IQR]) for key measured variables in the worse and better eye.
Better eye Worse eye Wilcoxon’s 𝑃 value
- MD (dB, median, and IQR) −. (−., −.) −. (−., −.) <.
- MD (dB, median, and IQR) −. (−., −.) −. (−., −.) <.
CS (Log CS, median, and IQR) . (., .) . (., .) .
VA (log units, median, and IQR) . (−., .) . (., .) .
Reading duration (seconds, median, and IQR) . (., .) . (., .)
Saccade rate (sac/sec, median, and IQR) . (., .) . (., .)
eyes for reading duration and saccade rate (worse eye minus
better eye) to create novel “change” variables for each person.
e dierences between the worse and better eye were also
calculated for all the measured visual function parameters
(i.e., change in VF severity, VA, and CS between eyes) and
then each of these resulting variables was compared to the
changes in reading duration and saccade rate between eyes.
erefore, it could be determined whether larger reductions
in visual eld defect severity, contrast sensitivity, or visual
acuity were related to a greater change in reading duration
or eye movement behaviour when reading with the worse eye
compared to the better eye.
Finally, dierences in the median values for each of the
identied eye movement types between the worse and better
eyewerecalculatedforeachperson;thesewerethencom-
pared to the change in reading duration per trial and saccade
rate between eyes. Statistically signicant associations were
tested for using Spearman’s rank correlation (rho) and also
using R [].
3. Results
Fourteen patients with a median age of (interquartile range
[IQR] of to ) years took part in the study. All participants
wereCaucasianand%weremen.epatientshadarange
of visual eld defects, visual acuity, and contrast sensitivity
measures (shown in Table ). Participants’ worse eyes and
better eyes were, as expected, signicantly dierent in -
MD, - MD, and PR Log CS but not in visual acuity
(Wilcoxon’s test). For example, median (interquartile range)
between-eye dierence in - MD was . (. to .) dB.
In of the patients, the “worse eye” was the right eye.
Table also shows median (IQR) reading durations and
saccade rates for the patients’ worse and better eyes. A linear
mixed eects ANOVA indicated that on average patients
took longer to read the sentences with their worse eye than
with their better eye and this was statistically signicant
(𝐹 = 132.3, 𝑃 < 0.001). Furthermore, patients made fewer
saccades per second, on average, when reading with their
worse eye compared to their better eye (𝐹 = 84.9, 𝑃 < 0.001).
When considering statistical associations for the change
in reading duration and saccade rate between eyes, an average
increase in reading duration in the worse eye compared to the
better eye was closely related to an average decrease in the
saccade rate in the worse eye compared to the better eye (rho:
−.; 𝑃 < 0.001; Figure ). In other words, those who took
longer to read with their worse eye than the better eye also
Dierence in saccade rate (%)
Dierence in reading duration (%)
0
50
100
150
200
−30 −20 −10 0
rho: −0.83
F : Scatterplots depicting the statistically signicant relation-
ships between the percentage dierence in reading duration between
the worse eye and the better eye and the percentage dierence in
saccade rate between the worse eye and the better eye.
had a greater reduction in saccade rate than those who read
at a similar speed in each eye.
Associations for the change in visual function measures
between the better and worse eye compared with the changes
for reading duration and saccade rate are shown in Table .
ere was noteworthy association between change in saccade
rate and the extent of dierence in contrast sensitivity
between the better and worse eye. So those with a greater
reduction in contrast sensitivity in the worse eye were more
likelytohaveareducedsaccaderateintheworseeye
(Figure (a)). Furthermore, those patients with a greater drop
in visual acuity in the worse eye also showed a greater
reduction in saccade rate (Figure (b)). ere were no other
statistically signicant correlations (Table ).
Table shows the proportion of saccades classied as each
of the four eye movement types for the better and worse eyes,
respectively. ere were no statistically signicant dierences
in these values between eyes. However, a larger increase in
reading duration in the worse eye compared to the better
eye was associated with an increase in the percentage of eye
movements that were regressions in the worse eye compared
to the better eye (rho: .; 𝑃 < 0.03; Figure (a)). In
Journal of Ophthalmology
T : Spearman’s rho correlations comparing the dierence in reading duration between the worse eye and the better eye and the dierence
in saccade rate between the worse eye and the better eye, with key measured variables related to age and vision.
Dierence between eyes
- MD - MD Mean central VF points CS VA Age
Change in reading duration per trial rho −. . . −. . .
𝑃 value . . . . . .
Change in saccade rate rho . −. . .
∗
−.
∗
−.
𝑃 value . . . . . .
Statistically signicant associations are marked with an asterisk.
T : Proportion of saccades that were forward, between lines,
regressions, or unknown when reading with the best eye and worse
eye, respectively.
Better eye Worse eye
Forward saccades
(%, median, and IQR)
. (., .) . (., .)
Line change
(%, median, and IQR)
. (., .) . (., .)
Regressions
(%, median, and IQR)
. (., .) . (., .)
Unknown (%, median, and IQR)
. (., .) . (., .)
addition, a greater increase in reading duration in the worse
eye compared to the better eye was associated with making
more unknown eye movements in the worse eye compared
to the better eye (rho: .; 𝑃 < 0.03; Figure (b)).
4. Discussion
For reading, it is clear that some patients are more aected
by vision loss in glaucoma than others. Some patients with
glaucoma self-report diculties with reading [, , ]. In
addition, reading speed experiments indicate that patients
with glaucoma have more problems with reading than people
with normal vision but only “on average”; some patients
with visual eld loss performed similarly or better than
people with healthy vision [–, , ]. Reading speed
can vary considerably between people making it dicult to
make comparisons between patients and controls; in these
studies adjustments are needed for covariates for reading
speed such as education, cognitive ability, age, amount of
day-to-day reading, and ethnicity. Such studies also require
largesamplesizes[]. Our study examined an alternative
experimental design: comparing performance between eyes
in patients with asymmetrical visual eld loss. Principally we
demonstrated a statistically signicant dierence in the time
it took patients to read a short passage of text in what is con-
sidered to be their worse eye (most visual eld damage) when
compared to their better eye (least visual eld damage). is
was done in a small sample of patients that carried out the
reading task many times. e eect size was, however, small
and the dierence in reading duration between eyes was not
associated with the magnitude of the dierence in visual eld
loss between the two tested eyes. In other words, there was no
“dose” eect: larger dierences in severity of visual eld defect
between eyes were not associated with worse performance.
is was true for the MD from a standard clinical visual
eld test (- HFA) and a visual eld test of more central
areas (- HFA). It is therefore unclear if an overall summary
measure of visual eld defect severity can be predictive of
worsening reading performance in glaucoma. ere was no
signicant dierence between eyes for visual acuity when
considering the average of all patients; this nding likely
reects the fact that many patients with worsening glaucoma
maintain relatively good visual acuity while other aspects of
visual function decline. However, when considering within-
person dierences in visual acuity in the worse versus the
better eye, a larger decline in visual acuity was associated with
a greater reduction in reading speed in the worse eye. is
nding highlights the benets of considering performance
changes within each individual in addition to consider-
ing average eects across all participants. e magnitude
of the dierence in contrast sensitivity between eyes was
also related to dierence in reading performance between
eyes. e important role of contrast sensitivity in reading
performance in glaucoma has been emphasised elsewhere
[].
is experiment was novel in comparison with most
other studies investigating reading performance in people
with glaucoma because it took advantage of measurements
from an eye tracker. Patients had a reduced saccade rate
(making fewer saccades per second) on average when reading
with their worse eye compared to their better eye. Fur-
thermore, average saccade rate was strongly associated with
reading duration. ese ndings imply that saccade rate,
measured by an eye tracker, could be a useful surrogate
for reading performance. A reduction in saccade rate in
patients with visual eld defects has also been observed in
other studies involving dierent visual tasks [, ]and
other experimental results suggest that saccadic initiation in
patients with glaucoma is delayed relative to controls with
healthy vision []. It may be that visual function loss caused
by glaucoma impairs the ability of the visual system to process
the surrounding information during each glance, meaning
that it takes longer to initiate a saccade towards relevant infor-
mation. Nevertheless, although reduced reading duration and
saccaderatewereobservedonaveragefortheworseeye
compared to the better eye, the degree of change between eyes
Journal of Ophthalmology
Dierence in contrast sensitivity (log CS)
Dierence in saccade rate (%)
rho: 0.65
−30
−20
−10
0
−0.6 −0.4 −0.2 0.0
(a)
0.0
0.2 0.4
Dierence in logMar acuity (log units)
Dierence in saccade rate (%)
−30
−20
−10
0
rho: −0.56
(b)
F : Scatterplots depicting the statistically signicant rela-
tionships between (a) the dierence in contrast sensitivity (log)
and percentage dierence in saccade rate between eyes and (b) the
dierence in logMAR visual acuity and the percentage dierence in
saccade rate between eyes.
varied considerably across patients. For example, Figure
shows that certain patients had a much longer reading dura-
tionfortheworseeyeandalsotendedtoshowamorereduced
saccade rate. However, other patients appeared to be less
aected in terms of reading speed when reading with their
worse eye and these people also tended to maintain a similar,
or increased, saccade rate to the better eye. Typically when
reading, there will be a “window” of information that can be
absorbed during each xation, referred to as the “perceptual
rho: 0.60
0
50
100
150
200
0 100 200 300
Dierence in regressions (%)
Dierence in reading duration (%)
−100
(a)
rho: 0.59
0
50
100
150
200
0 100 200 300 400
Dierence in unknowns (%)
Dierence in reading duration (%)
−100
(b)
F : Scatterplots showing statistically signicant relationships
between the percentage dierence in reading duration between the
better and worse eye and the dierence between the better and worse
eye in (a) the proportion of regressions and (b) the proportion of
“unknown” eye movement.
span.” Visual degradation caused by visual eld defects can be
expected to reduce the number of characters that can be read
with each xation [, ], suggesting that more saccades
must subsequently be made in order to process the same
quantity of information. erefore, some patients may have
maintained an adequate reading speed when reading with
their worse eye by increasing their saccade rate in order to
overcome the impairment that would normally be expected
due to visual degradation. is result coincides, in part, with
a nding that suggests that glaucomatous visual eld loss
Journal of Ophthalmology
restricts saccades during other tasks such as visual search but
that increasing saccade rate is associated with maintaining
“good” performance []. It is unknown whether these eye
movements are adaptive behaviour, and so this topic should
be the subject of future investigation.
Eye tracking generates copious data that can be easily
misidentied or misinterpreted. Eye movement analysis so-
ware for reading experiments typically provides scanpath
data [, , ] that has to be manually delineated to extract
specic saccades like regressions (a backtracking saccade
sometimes observed during reading). So, for this study, we
developed some automated techniques for identifying the
dierent types of eye movements made during the reading
task. In this experiment, there was no statistically signicant
dierence in the types of eye movements identied by the
algorithmmadebytheeyewithmorevisualelddamage
compared to the eye with less visual eld damage. Still,
there was a relationship between increases in the proportion
of regressions and worse reading performance. e algo-
rithm also automatically identied unknown or “irregular”
eye movements that were associated with poorer reading
performance in the worse eye compared to the better eye.
Patients who followed more conventional reading patterns
(making a smaller proportion of regressions and unknown
eye movements compared to forward saccades) in both eyes
appeared to read equally quickly in both eyes. ese ndings
illustrate the utility of eye tracking in studies of reading
in glaucoma and hint at the design of future studies. For
example, recent research suggests that reading performance
in patients with glaucoma is particularly aected during
sustained reading as opposed to when reading short passages
of text []; it might be useful to use eye tracking in future
experiments of that type.
ere are limitations associated with our study. ere was
no assessment of comprehension of the texts and the nature
of the reading experiment—large font size and reading from
a computer screen—does not mimic everyday reading. e
sample size was not large enough to tease out any statistically
signicant dierences in the types of eye movements that
mightbeusedbyaneyewithworsevisualelddamagecom-
paredtoonewithlessvisualelddamage.Wecertainlydid
not have enough eyes to explore how reading performance
is aected by the precise location of a visual eld defect or
how a similar visual eld defect in the right eye as compared
to the le eye might inuence performance; this awaits
further study. Future research may also wish to consider
the performance of people with asymmetric visual eld loss
when reading bilaterally and whether this is comparable to
reading monocularly with the better or worse eye. It is also
importanttopointoutthatourmethodsforpreprocessing
the eye movement data and for automatically classifying their
properties have not been validated or compared with manual
methods. Nevertheless, the study still adds to the literature by
showing the potential of eye tracking for understanding how
patients with visual eld defects function in everyday tasks
such as reading.
In summary, this study has shown that patients with
glaucoma will take longer to read a short passage of text
in what is considered to be their worse eye (most visual
elddamage)whencomparedtotheirbettereye(leastvisual
eld damage). However, the eects were small. Unexpectedly,
reading performance did not worsen in the eye with most
visual eld damage as the between-eye dierences in visual
eld defect severity increased (as measured by a single
summary measure of the visual eld). We have also presented
novel analytical eye movement data analysis that might be
useful for other reading studies. e results suggest that
regressions and unknown saccades result in slower reading
speeds. In conclusion, we have demonstrated the utility of
a novel experimental design that might help unravel the
relationship between glaucomatous vision loss and diculties
with reading. For example, a future study comparing per-
formance between eyes and using eye tracking could help
determine the precise location of visual eld loss that inhibits
reading performance in glaucoma.
Conflict of Interests
e authors declare that there is no conict of interests
regarding the publication of this paper.
Acknowledgment
isworkisfundedinpartbyanunrestrictedgrantfrom
the Investigator-Initiated Studies Program of Merck Sharp &
Dohme, Corp.
References
[] H. Quigley and A. T. Broman, “e number of people with
glaucoma worldwide in and ,” British Journal of
Ophthalmology, vol. , no. , pp. –, .
[] P. A. Aspinall, Z. K. Johnson, A. Azuara-Blanco, A. Montarzino,
R. Brice, and A. Vickers, “Evaluation of quality of life and pri-
orities of patients with glaucoma,” Investigative Ophthalmology
and Visual Science,vol.,no.,pp.–,.
[] P. Nelson, P. Aspinall, and C. O’Brien, “Patients’ perception of
visual impairment in glaucoma: a pilot study,” British Journal of
Ophthalmology,vol.,no.,pp.–,.
[]A.C.Viswanathan,A.I.McNaught,D.Poinoosawmyetal.,
“Severity and stability of glaucoma patient perception com-
pared with objective measurement,” Archives of Ophthalmology,
vol. , no. , pp. –, .
[]E.E.Freeman,B.Mu
˜
noz, S. K. West, H. D. Jampel, and D.
S. Friedman, “Glaucoma and quality of life: the salisbury eye
evaluation,” Ophthalmology,vol.,no.,pp.–,.
[]D.P.Crabb,N.D.Smith,F.C.Glen,R.Burton,andD.F.
Garway-Heath, “How does glaucoma look? patient perception
of visual eld loss,” Ophthalmology, vol. , pp. –, .
[]K.Fujita,N.Yasuda,K.Oda,andM.Yuzawa,“Reading
performance in patients with central visual eld disturbance
due to glaucoma,” Nippon Ganka Gakkai zasshi,vol.,no.,
pp.–,.
[] P. Ramulu, “Glaucoma and disability: which tasks are aected,
andatwhatstageofdisease?”Current Opinion in Ophthalmol-
ogy, vol. , no. , pp. –, .
Journal of Ophthalmology
[] M. Ishii, M. Seki, R. Harigai, H. Abe, and T. Fukuchi, “Reading
performance in patients with glaucoma evaluated using the
MNREAD charts,” Japanese Journal of Ophthalmology,vol.,
pp.–,.
[] U. Altangerel, G. L. Spaeth, and W. C. Steinmann, “Assessment
of function related to vision (AFREV),” Ophthalmic Epidemiol-
ogy,vol.,no.,pp.–,.
[] R. Burton, D. P. Crabb, N. D. Smith, F. C. Glen, and D. F.
Garway-Heath, “Glaucoma and reading: exploring the eects of
contrast lowering of text,” Optometry and Vision Science,vol.,
pp.–,.
[] P.Y.Ramulu,B.K.Swenor,J.L.Jeerys,D.S.Friedman,and
G. S. Rubin, “Diculty with out-loud and silent reading in
glaucoma,” Investigative Ophthalmology & Visual Science,vol.
, pp. –, .
[] N.D.Smith,F.C.Glen,andD.P.Crabb,“Eyemovementsduring
visual search in patients with glaucoma,” BMC Ophthalmology,
vol. , article , .
[]F.C.Glen,N.D.Smith,andD.P.Crabb,“Saccadiceye
movements and face recognition performance in patients with
central glaucomatous visual eld defects,” Vision Research,vol.
, pp. –, .
[]N.D.Smith,D.P.Crabb,F.C.Glen,R.Burton,andD.F.
Garway-Heath, “Eye movements in patients with glaucoma
when viewing images of everyday scenes,” Seeing and Perceiving,
vol. , pp. –, .
[] D. P. Crabb, N. D. Smith, F. G. Rauscher et al., “Exploring eye
movements in patients with glaucoma when viewing a driving
scene,” PloS ONE,vol.,no.,p.e,.
[] A. Cerulli, M. Cesareo, E. Ciuoletti et al., “Evaluation of
eye movements pattern during reading process in patients
with glaucoma: a microperimeter study,” European Journal of
Ophthalmology,vol.,no.,pp.–,.
[]F.C.Glen,D.P.Crabb,N.D.Smith,R.Burton,andD.F.
Garway-Heath, “Do patients with glaucoma have diculty
recognising faces?” Investigative Ophthalmology & Visual Sci-
ence,vol.,no.,pp.–,.
[] P. H. Artes, N. O’Leary, D. M. Hutchison et al., “Properties of the
statpac visual eld index,” Investigative Ophthalmology & Visual
Science,vol.,no.,pp.–,.
[] E.Hodapp,R.I.Parrish,andD.R.Anderson,Clinical Decisions
in Glaucoma,Mosby,St.Louis,Mo,USA,.
[] S. A. Kabanarou and G. S. Rubin, “Reading with central
scotomas: is there a binocular gain?” Optometry and Vision
Science, vol. , no. , pp. –, .
[] I. Nabney, NETLAB: Algorithms for Pattern Recognition,
Springer,NewYork,NY,USA,.
[] R Development Core Team, R: A Language and Environment for
Statistical Computing, R Foundation for Statistical Computing,
Vienna, Austria, .
[] C. M. Mangione, S. Berry, K. Spritzer et al., “Identifying the
content area for the -item national eye institute visual function
questionnaire: results from focus groups with visually impaired
persons,” Archives of Ophthalmology,vol.,no.,pp.–,
.
[]K.Fujita,N.Yasuda,K.Nakamoto,andT.Fukuda,“e
relationship between diculty in daily living and binocular
visual eld in patients with glaucoma,” Nippon Ganka Gakkai
Zasshi,vol.,no.,pp.–,.
[]P.Y.Ramulu,S.K.West,B.Munoz,H.D.Jampel,andD.
S. Friedman, “Glaucoma and reading speed: the salisbury eye
evaluation project,” Archives of Ophthalmology,vol.,no.,
pp. –, .
[] R.Kanjee,Y.H.Y
¨
ucel,M.J.Steinbach,E.G.Gonz
´
alez, and N.
Gupta, “Delayed saccadic eye movements in glaucoma,” Eye
,vol.
, pp. –, .
[] G. Lous, L. Kaufman, T. Nishimoto, and E. Ruthru, “Eects
of visual degradation on eye-xation duration, perceptual
processing, and long-term visual memory,” in Eye Movements
andVisualCognition,K.Rayner,Ed.,Springer,NewYork,NY,
USA, .
[] M. A. Bullimore and I. L. Bailey, “Reading and eye movements
in age-related maculopathy,” Optometry and Vision Science,vol.
,no.,pp.–,.
[] M. D. Crossland and G. S. Rubin, “Eye movements and reading
in macular disease: further support for the shrinking perceptual
span hypothesis,” Vision Research,vol.,no.,pp.–,
.
[] G. S. Rubin and M. Feely, “e role of eye movements during
reading in patients with age-related macular degeneration
(AMD),” Neuro-Ophthalmology,vol.,no.,pp.–,.
Submit your manuscripts at
http://www.hindawi.com
Stem Cells
International
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
M EDIATO R S
INFLAM MATION
of
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Behavioural
Neurology
Endocrinology
International Journal of
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Disease Markers
Hindawi Publishing Corporation
http://www.hindawi.com
Volume 2014
BioMed
Research International
Oncology
Journal of
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Oxidative Medicine and
Cellular Longevity
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
PPAR Research
The Scientic
World Journal
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Immunology Research
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Journal of
Obesity
Journal of
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Computational and
Mathematical Methods
in Medicine
Ophthalmology
Journal of
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Diabetes Research
Journal of
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Research and Treatment
AIDS
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Gastroenterology
Research and Practice
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Parkinson’s
Disease
Evidence-Based
Complementary and
Alternative Medicine
Volume 2014
Hindawi Publishing Corporation
http://www.hindawi.com
