Journal of

Clinical Medicine

Article
Design and Usability Study of a Point of Care mHealth App for
Early Dry Eye Screening and Detection
Sydney Zhang * and Julio Echegoyen

Department of Clinical Research, Westview Eye Institute, San Diego, CA 92129, USA;
jechegoyen@wvoptometry.com
* Correspondence: sydneyz2266@gmail.com; Tel.:+1-858-883-3869

Abstract: Significantly increased eye blink rate and partial blinks have been well documented in
patients with dry eye disease (DED), a multifactorial eye disorder with few effective methods for
clinical diagnosis. In this study, a point of care mHealth App named “EyeScore” was developed,
utilizing blink rate and patterns as early clinical biomarkers for DED. EyeScore utilizes an iPhone for
a 1-min in-app recording of eyelid movements. The use of facial landmarks, eye aspect ratio (EAR)
and derivatives enabled a comprehensive analysis of video frames for the determination of eye blink
rate and partial blink counts. Smartphone videos from ten DED patients and ten non-DED controls
were analyzed to optimize EAR-based thresholds, with eye blink and partial blink results in excellent
agreement with manual counts. Importantly, a clinically relevant algorithm for the calculation of “eye
healthiness score” was created, which took into consideration eye blink rate, partial blink counts as
well as other demographic and clinical risk factors for DED. This 10-point score can be conveniently
measured anytime with non-invasive manners and successfully led to the identification of three
individuals with DED conditions from ten non-DED controls. Thus, EyeScore can be validated as a
valuable mHealth App for early DED screening, detection and treatment monitoring.

Keywords: dry eye disease; diagnosis; mHealth; point of care; smartphone app; eye blink; EyeScore;
eye healthiness score

Citation: Zhang, S.; Echegoyen, J.

Design and Usability Study of a Point 1. Introduction
of Care mHealth App for Early Dry
The eye is one of the most important organs, and about 80% of outside information
Eye Screening and Detection. J. Clin.
is processed by the visual pathway [1]. Human tears play a crucial role in the protection
Med. 2023, 12, 6479. https://
doi.org/10.3390/jcm12206479
and lubrication of the ocular surface, and its secretion is mainly through the lacrimal
gland [2]. The function of blinking is to bring fresh tears and nutrients over the ocular
Academic Editors: Kyung Chul Yoon surface, while also removing metabolic waste or debris [3,4]. Research has shown that the
and Brent Siesky eye blink rate normally ranges from 15~20 blinks per minute [5,6]. Dry eye disease (DED)
Received: 14 September 2023 is the most common eye disease with a prevalence ranging from 5% to 50% globally [7],
Revised: 27 September 2023 which is projected to significantly increase in the coming years [8,9]. Dry eye patients
Accepted: 10 October 2023 usually experience blurry vision with a dry and gritty feeling in their eyes [10,11], blinking
Published: 12 October 2023 functions as a compensatory mechanism for their dysfunctional and unstable tear film.
More than two times higher blink rates were widely reported in DED patients as compared
to the normal controls [3,12]. In addition, partial blinks were also a common blink pattern
in DED patients and featured no contact between the upper and lower eyelids. It was
Copyright: © 2023 by the authors.
reported that OSDI and the number of partial blinks were positively correlated [13]. Besides
Licensee MDPI, Basel, Switzerland.
blink rate and partial blink, comprehensive studies also showed that extended lid contact
This article is an open access article
time, interblink interval and maximum blink interval (MBI) can be critical biomarkers for
distributed under the terms and
the early diagnosis of DED [12–15].
conditions of the Creative Commons
Recent development in medical big data and mHealth offers a great opportunity for
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
DED patients to safeguard their eyecare [16]. Since early diagnosis is the key to preventing
4.0/).
DED from progression, creating a point-of-care (POC), real-time, user-centered digital tool

J. Clin. Med. 2023, 12, 6479. https://doi.org/10.3390/jcm12206479 https://www.mdpi.com/journal/jcm

J. Clin. Med. 2023, 12, 6479 2 of 14

targeting relevant biomarkers (such as eye blink rate and partial blink rate) can effectively
address the critical challenges and provide significant clinical values for personalized
diagnosis with early intervention and the treatment of DED [17].
However, very few mHealth Apps are currently available for dry eye diagnosis. In
Japan, efforts have been made recently to develop a mHealth smartphone App called
“DryEyeRhythm” or “DEA01” for dry eye diagnosis assistance [18] (DryEyeRhythm can be
downloaded from the Apple App Store with some contents in Japanese). This mobile App
mainly uses the Japanese version of the OSDI questionnaire (J-OSDI) as well as App-based
MBI measurement for DED diagnosis. The initial clinical assessment of DryEyeRhythm
exhibited positive results with satisfactory internal data consistency [19,20]. It was reported
that patient enrollment has started in Japan for a clinical trial of DEA01 in 2023 [21].
The goal of our current study is to design an iOS smartphone App called EyeScore,
which can be used to accurately measure blink rate and blink patterns through advanced
computational algorithms, followed by an App-based questionnaire for demographic infor-
mation and eye disease symptoms or history. Then, a clinically relevant and personalized “eye
healthiness score” can be reported in real-time after each test. This 10-point eye healthiness
score was designed to serve as an indicator of the eye conditions of the patient, which can
be non-invasively measured with the EyeScore App anytime at home and shared with the
patient’s eye doctor remotely in low-resource settings. The primary purpose of the EyeScore
App is to rapidly screen for patients at the early stage of DED or with mild DED conditions
through the frequent in-home monitoring of eye conditions at no cost. Moreover, EyeScore
can also be used to monitor eye conditions during DED treatment courses. Our initial test of
EyeScore with 20 participants (including 10 confirmed DED patients) demonstrated encour-
aging results, with 3 of 10 non-DED participants identified with mild DED conditions. This
proof-of-concept study paved the way for the EyeScore App to be further evaluated for early
DED screening and detection in large-scale clinical validation studies in the future.

2. Materials and Methods

2.1. Study Participant Selection and Group Allocation
A total of 20 test participants were included in this study with age ≥ 18 years old.
The written informed consent was obtained from each participant before taking a 1 min
video recording. All involved parties in this study made efforts to protect study patients’
personal information and privacy. The test participants were allocated at a 1:1 ratio into
the DED group (including 10 DED patients with confirmed diagnosis) and the non-DED
group (10 individuals with no known eye disease conditions). Table 1 shows the demo-
graphic data of 20 participants in this study, including their gender, sex and occupation. In
addition, daily electronics use time and pre-existing eye diagnosis or symptoms were also
documented. A 1 min video was recorded for each participant according to the standard
protocol described in the following section.

Table 1. Demographic data from 20 participants in this study.

Regular Contact Preexisting Eye Symptoms

Group Enrollee No. Gender Age Occupation Electronics Use †
Lens Wearing or Diagnosis/Treatment ‡
1 M 58 Researcher +++ No Normal
2 F 20 Student +++ Yes Normal
3 F 21 Student +++ No Normal
4 F 22 Student +++ No Normal

Normal 5 F 53 Homemaker + No Normal

6 F 47 Office worker +++ Yes Normal
7 F 56 Homemaker + Yes Normal
8 F 80 Homemaker + No Normal
9 M 65 Cook + No Normal
10 M 29 Engineer ++++ No Normal
J. Clin. Med. 2023, 12, 6479 3 of 14

Table 1. Cont.

Regular Contact Preexisting Eye Symptoms

Group Enrollee No. Gender Age Occupation Electronics Use †
Lens Wearing or Diagnosis/Treatment ‡
DED 1 F 48 Engineer ++++ Yes Confirmed DED
Confirmed DED/
DED 2 M 80 Clinician ++ No
under treatment
Confirmed DED/
DED 3 F 83 Homemaker + No
under treatment
DED4 F 21 Student +++ Yes Confirmed DED
DED DED 5 M 60 Financial planner +++ Yes Confirmed DED
DED 6 F 51 Researcher +++ Yes Confirmed DED
Confirmed DED/
DED 7 F 28 Clinician ++ Yes
under treatment
Confirmed DED/
DED 8 F 44 Homemaker + Yes
under treatment
DED 9 M 60 Engineer ++++ No Confirmed DED
DED 10 M 19 Game Designer ++++ Yes Confirmed DED
† Electronic use daily (hours): +, 0~3 h; ++, 3~6 h; +++, 6~8 h; ++++, >8 h. ‡ All enrollees in the DED group have
confirmed DED diagnosis from their eye doctors. Some of them were under DED treatment including artificial
tears, omega-3, warm eye compress and blinking exercises.

2.2. Video Recording Conditions and Digital Data Storage

Standardized iPhone video recording conditions were as follows: (1) Distance, the
viewing distance from the eyes to the iPhone camera was one arm length (~76 cm or
~30 inches). An adjustable iPhone stand was used in this study, while handheld iPhone
recording at arm’s length was also acceptable. (2) Lighting, normal indoor lighting with
a simple background. (3) Time, 1 min recording time. (4) Resolution, HD resolution was
used in our study, but 4k resolution was also acceptable for data processing. (5) Video data
storage, all in-app recording videos were securely stored both in the archived database
at Westview Eye Institute (San Diego, CA, USA) and the Apple iCloud private account.
J. Clin. Med. 2023, 12, x FOR PEER REVIEW 4 of 14
Figure 1 shows a study flow chart from participant enrollment to EyeScore App testing and
post-EyeScore DED diagnosis confirmation.

Figure 1. A study flowFigure

chart1.of
A study flow chart
EyeScore App of evaluation
EyeScore App evaluation
for DEDfor DED screening
screening andand detection.
detection.
2.3. Computer Programming Tools and Software Development Package
For EyeScore App development, Swift programming was performed with an Apple
MacBook Pro laptop installed with iOS 16.1.2 and Xcode 14.0. The Apple Developer doc-
umentation was adopted for downloading VNFaceLandmarks2D for the development of
eye landmark detection and data array acquisition (Apple Developer Documentation,
2023).

2.4. EAR-Based Algorithm Development for Blink Count

 J. Clin. Med. 2023, 12, 6479 4 of 14

2.3. Computer Programming Tools and Software Development Package

For EyeScore App development, Swift programming was performed with an Apple
MacBook Pro laptop installed with iOS 16.1.2 and Xcode 14.0. The Apple Developer docu-
mentation was adopted for downloading VNFaceLandmarks2D for the development of eye
landmark detection and data array acquisition (Apple Developer Documentation, 2023).

2.4. EAR-Based Algorithm Development for Blink Count

Soukupova and Cech first proposed an algorithm using six landmarks around the
eye, namely P1 to P6 (Figure 2A) from 2D images [22]. A single scalar parameter called
“eye aspect ratio” (EAR) was created as a determinant of the eye opening (Figure 2A,
left)/closing (Figure 2A, right) state with a computational formula shown in Figure 2B. The
EAR value remains relatively constant (~0.25, Figure 2C) when eyes are open, but rapidly
12, x FOR PEER REVIEW decreases to near 0 when eyes are closed (Figure 2C). As a result, the successful
5 of 14 use of
a fixed-value EAR threshold led to the development of eye blink counts for a 3D video
stream with OpenCV, Python, and Dlib [23–25].

Figure 2. Eye landmarksFigure 2. Eye landmarks

(P1~P6) and EAR (P1~P6) and EAR
calculation calculation
formula for 2Dformula for 2D
eye blink eye blink(A)
detection. detection.
The (A) The
locations of(P1
locations of six eye landmarks six eye landmarks
to P6) around(P1thetoeye
P6)during
around the
the eye
eye during
openingthe(left)
eye opening (left) or closing
or closing
(right) state. (B) The (right)
formula state.
for(B) The formula
calculation offor
thecalculation of the(C)
EAR values. EAR values.
The EAR(C) The EAR
values valueswhen
decrease decrease when
the eye is closed. the eye is closed.

The development of eye landmark recognition and EAR thresholds made it possible to
Figure 3 shows
usethe schematic
a similar programming
approach flowchartenvironment
in the Xcode-integrated of the EyeScore App,
for an iOS Appwhich
development.
includes three code
Theblocks labeled
authors with green,
in the current orange, and
study proposed purple
to use colors.
a 1-min iPhoneThe detailed
video so the reliable
eye blink with
programming procedures behavior
the of each participant
numbered steps can be captured
(1)–(16) through in-app
were described recording and
in Supple-
subsequently
mental Materials S1. stored in the iPhone photo library. For video processing, the EAR formula
from Figure 2B was utilized initially to calculate the EAR value of each video frame,
with the reporting of the individual EAR, Average EAR, MaxEAR, and MinEAR from all
processed frames.
Figure 3 shows the schematic programming flowchart of the EyeScore App, which includes
three code blocks labeled with green, orange, and purple colors. The detailed programming
procedures with the numbered steps (1)–(16) were described in Supplemental Materials S1.
 the eye is closed.

Figure 3 shows the schematic programming flowchart of the EyeScore App, which
includes three code blocks labeled with green, orange, and purple colors. The detailed
J. Clin. Med. 2023, 12, 6479 programming procedures with the numbered steps (1)–(16) were described in Supple-
5 of 14
mental Materials S1.

Figure 3. Programming flow chart with fixed EAR approach with the obtained data for dynamic EAR
approach. Three loops of code blocks were labeled with green, orange, and purple colors, with (1) to
(16) referring to the numbered step in the diagram.

However, a fixed-value threshold approach failed to correctly report the blink rate
and partial blink, particularly when tests were performed with dry eye patients. This was
because DED patients frequently show abnormally high partial blink rates. Using data
obtained from the purple block (Figure 3), a dynamic EAR-value-based algorithm was
developed as a personalized threshold to determine the eye opened or closed state for
this study.

2.5. Creation of an Eye Healthiness Score for Monitoring Eye Conditions

Based on the literature review on DED, a 10-point formula was proposed in this study
with five components for computing an eye healthiness score (Table 2). Two components
(blink rate and partial blink rate) were reported from video-based analysis with EyeScore
App, while the other three components were obtained from the questionnaire filled out by
the study participants. These three components were the most common risk factors known
for DED development [26,27]. The supporting references for each of the components are
also included in Table 2.
J. Clin. Med. 2023, 12, 6479 6 of 14

Table 2. Different factors for computing an eye healthiness score by EyeScore App.

Factor Point * References

Blink rate > 30 per min 3 [5]
Partial blink > 5 per min 3 [12,13]
Dry and gritty feeling 2 [26]
Women over 50 years old 1 [26]
Regular contact lens wearing 1 [27]
Total 10
Normal: 0–3; Mild DED: 4–6; DED: 7–10; Severe DED (will need a doctor’s visit)
* Point can be optimized with other factors [28] and statistical approaches [29].

Briefly, (1) Blink rate. The normal blink rate ranges from 15–20 per minute, while
a blink rate over 30 per minute is considered a high probability of DED [5]. It carries a
three-point weight in our formula. (2) Partial blink. The partial blinks are common in DED
patients due to their dry and dysfunctional ocular surface, while normal controls should
have partial blinks < 5 per minute [12,13]. It also carries 3 points in the formula. (3) Dry
and gritty feeling. clinically called “foreign body sensation” [26]. The patients start to
develop DED symptoms at this stage so it carries two points in the formula. (4) Women over
50 years old. This group of women shows a statistically higher risk of DED development
among all population groups [26]. It carries 1 point in the formula. (5) Regular contact lens
wearers. Many publications show that regularly wearing soft or hard contact lens have a
significantly higher risk of inducing DED [27]. Thus, a regular contact lens wearer (not
occasional use of contact lens) carries one point in the formula.
Although many other risk factors also contribute to the proposed eye healthiness
score for DED diagnosis [28], our initial formula focused on the most important factors,
with a score ≥ 4 being considered as mild DED conditions. When the eye healthiness
score ≥ 7, the EyeScore App will report it as severe DED and recommend the user schedule
doctor’s visit to confirm the DED diagnosis (Table 2). A healthy individual with normal
eye conditions should have a score ≤ 3. Although the weight/points for this formula have
significant limitations, the authors decided to use this scoring system for our small pilot
study. The proposed eye healthiness score can be further optimized in the future with other
important risk factors [28] for linear regression analysis [29].

2.6. Statistical Analysis

The student t-test was performed using GraphPad Prism 9.0 (GraphPad Software
Inc, San Diego, CA, USA). The statistical significance was determined by p-values as
follows: p-values of <0.05; <0.01; and <0.001. The mean ± standard deviation was obtained
from at least three separate experiments.

3. Results
3.1. EyeScore App Interface and Questionnaire
Selected EyeScore App functions and user interfaces are shown in Figure 4, which
include a home page, eye landmark detection, in-app recording, questionnaire, and eye
healthiness score report. The patient registration and account information will be available
in the future so the personalized EyeScore results can be securely stored and shared with
eye doctors for remote monitoring of eye conditions.
 3.1. EyeScore App Interface and Questionnaire
Selected EyeScore App functions and user interfaces are shown in Figure 4, which
include a home page, eye landmark detection, in-app recording, questionnaire, and eye
healthiness score report. The patient registration and account information will be availa-
J. Clin. Med. 2023, 12, 6479 7 of 14
ble in the future so the personalized EyeScore results can be securely stored and shared
with eye doctors for remote monitoring of eye conditions.

Figure 4. Representative
Figure 4. Representative EyeScore
EyeScore App
App functions
functions and
and user
user interfaces.
interfaces.

3.2. Limitations of a Fixed Value EAR Approach

3.2. Limitations of a Fixed Value EAR Approach
For EyeScore App algorithm development, videos from six controls and four DED
For EyeScore App algorithm development, videos from six controls and four DED
patients were used for the initial development of a fixed-value EAR (Table 3). To study
patients were used for the initial development of a fixed-value EAR (Table 3). To study the
the effects of masks on the blink count, two individuals (Enrollee 3 and 6) in the control
effects of masks on the blink count, two individuals (Enrollee 3 and 6) in the control group
group recorded the EyeScore videos without masks (3A, 6A) and with masks (3B, 6B). An
recorded the EyeScore videos without masks (3A, 6A) and with masks (3B, 6B). An initial
initial program code was written to allow the EyeScore App to iterate through all video
program code was written to allow the EyeScore App to iterate through all video frames
frames for eye landmarks., followed by the calculation of EAR for each frame. If there are
for eye landmarks., followed by the calculation of EAR for each frame. If there are three
three or more frames in a row with an EAR value ≤ a fixed-value EAR threshold, the App
or more frames in a row with an EAR value ≤ a fixed-value EAR threshold, the App counts
counts this sequence of frames as a single blink. The blink rate can be reported after the
this sequenceofofallframes
combination as ablinks
validated singleinblink. The blink
one minute. Asrate
seencan be reported
in Table after
3, a fixed the EAR
value combi-
of
nation of all validated blinks in one minute. As seen in Table 3, a fixed
0.163 was suitable for classifying open and closed eye states for three study participantsvalue EAR of 0.163
was
in thesuitable for classifying
six-member open and
control group (videoclosed
no. eye
1~6states
in Tablefor 3).
three
For study participants
example, in the
for videos 1,
six-member control group (video no. 1~6 in Table 3). For example,
2 and 4, blink count results from the EyeScore App with fixed value EAR vs. manual for videos 1, 2 and 4,
blink count results from the EyeScore App with fixed value EAR
count was 100% matched (Table 3). However, for all 4 confirmed DED patients (Video No.vs. manual count was
100% matchedand
DED1~DED4) (Table
three3). However,
controls (Video forNo.
all 3A,
4 confirmed DEDthepatients
3B, 5, 6A, 6B), (Video
blink count No.
results
DED1~DED4) and three controls (Video No. 3A, 3B, 5, 6A, 6B), the blink
were significantly lower due to the partial blinks not being counted (Table 3). As shown incount results were
significantly
Figure 2B, thelower due towas
EAR value the based
partialon
blinks not being
the distance counted
between eye(Table 3). As shown
landmarks on the in Fig-
upper
ure 2B, the EAR value was based on the distance between eye landmarks
and lower eyelids. The EAR values of partial blinks were significantly increased due to the on the upper
and lowerseparated
unclosed, eyelids. The EARthus
eyelids, values of partialthem
preventing blinks
fromwere significantly
being increased
counted. Another due to
variable
the unclosed,
was the mask.separated eyelids,
participants thus preventing
wearing a mask canthem from being
significantly counted.
interfere Another
with vari-
landmark
able was thefrom
recognition mask. participants wearing awith
VNFaceLandmarks2D, mask can significantly
many frames returned interfere
with with landmark
“no landmark
identified”. Video 3B and 6B represented examples of mask-wearing with significantly low
blink counts as compared to the manual counts. Thus, the authors decided to use videos
with no masks for the rest of the study. Moreover, wearing frame glasses was not allowed
when taking the video recordings due to the various refractive indices of the lens.
J. Clin. Med. 2023, 12, 6479 8 of 14

Table 3. EyeScore App count vs. manual count results using a fixed value EAR as a threshold.

EyeScore Data
Group Enrollee No. Manual Partial Blink Note
with Fixed EAR
Correct full blink count/no
1 10 10 yes
partial blink count
2 29 29 No Correct full blink count
Incorrect full blink count/no
3A (no mask) 29 30 Yes
partial blink count
Incorrect full blink count/no
3B (with mask) 20 30 Yes
Controls partial blink count
Correct full blink count/no
4 15 15 Yes
partial blink count
Incorrect full blink count/no
5 20 25 Yes
partial blink count
Incorrect full blink count/no
6A (no mask) 22 29 Yes
partial blink count
Incorrect full blink count/no
6B (with mask) 17 29 Yes
partial blink count
Incorrect full blink count/no
DED1 29 39 Yes
partial blink count
Incorrect full blink count/no
DED2 26 16 Yes
partial blink count
DED
Incorrect full blink count/no
DED3 28 35 Yes
partial blink count
Incorrect full blink count/no
DED4 25 36 Yes
partial blink count

In addition to the partial blink, different study participants (particularly DED patients)
exhibited a range of eyelid patterns when in the eye fully opened state or in the eye fully
closed state. As a result, the MaxEAR value and the MinEAR value can be significantly
different. The use of a single fixed value EAR as a threshold led to inaccurate results of
full blink and partial blink counts, as shown in Table 3. Thus, a dynamic, individualized
EAR threshold that considers these variables was essential to address these challenges for
improving the accuracy of full blink and partial blink counts, particularly in DED patients.

3.3. Development of a Dynamic EAR Approach with Two EAR Thresholds

Efforts were made to develop algorithms to two separate counts for full blink and
partial blink, respectively. Two EAR thresholds were essential for EyeScore to process the
3D videos for reporting separated full blink and partial blink rates. Of these two EAR
thresholds, one was defined as the full blink EAR threshold, while the other was called the
partial blink EAR threshold. After multiple tests of coefficient calculation for accurate blink
counts, the following algorithm was chosen to calculate the full blink EAR threshold value:

Full blink EAR threshold = (0.1281/0.27527) × (MaxEAR − MinEAR)

The MaxEAR and MinEAR values were obtained from steps (12) and (13) in Figure 3.
Meanwhile, the partial blink EAR threshold value was also defined as:

Partial blink EAR threshold = Full blink EAR threshold value × 1.4

In other words, a partial blink occurs when the frame EAR value is 40% higher than
the full blink EAR threshold. The two resulting dynamic threshold values were designed
to be more stringent and tied to the MaxEAR and MinEAR of the testing individual, which
allowed for the accurate reporting of full blink and partial blink count from all 20 recorded
videos (Table 4).
J. Clin. Med. 2023, 12, 6479 9 of 14

Table 4. EyeScore App results of 20 study participants with reporting of eye healthiness scores *.
Enrollee Preexisting EyeScore Eye Partial Video Open/Close Post-EyeScore
Group Age Gender Full Blink Rate
Number Diagnosis/Treatment Diagnosis Score Blink Rate Time (sec) Ratio (%) Confirmation
1 58 M Normal Normal 0 10 1 65 42.3
2 20 F Normal Normal 1 29 0 63 7.8
3 21 F Normal Normal 0 25 5 61 9.1
4 22 F Normal Normal 0 15 1 61 14.3
5 53 F Normal Mild DED 4 23 8 62 3.7 DED Confirmed
Control 6 47 F Normal Mild DED 6 21 6 62 2.4 DED Confirmed
7 56 F Normal Mild DED 5 31 3 60 7.3 DED Confirmed
8 80 F Normal Normal 3 22 1 60 4.0
9 65 M Normal Normal 2 28 0 62 7.9
10 29 M Normal Normal 0 18 0 58 18.4
Average 45 7F; 3M Average 2.1 22.2 2.5 61.4 11.7
DED1 48 F Mild DED Mild DED 4 39 1 62 2.6 DED Confirmed
DED2 80 M DED/Treated Mild DED 5 16 14 61 60 DED Confirmed
DED3 83 F DED/Treated Mild DED 6 35 3 62 6.8 DED Confirmed
DED4 21 F DED DED 7 36 8 61 3.4 DED Confirmed
DED5 60 M DED DED 7 40 14 61 19.3 DED Confirmed
DED
DED6 51 F DED DED 7 64 0 61 3.0 DED Confirmed
DED7 28 F DED/Treated Mild DED 6 13 6 61 9.9 DED Confirmed
DED8 44 F DED/Treated Mild DED 4 23 1 61 9.8 DED Confirmed
DED9 60 M Mild DED Mild DED 6 55 0 61 3.9 DED Confirmed
DED10 19 M DED DED 9 53 11 61 7.2 DED Confirmed
Average 49 6F; 4M Average 6.1 37.4 5.8 61.2 12.6
* (A) Enrollees with results highlighted in red color indicate DED diagnosis with abnormal eye blink rates, abnormal partial blink rates or abnormal eye scores. (B) Enrollees with
highlighted yellow color results indicated newly identified DED patients from the control group with abnormal eye scores, full blink rates or partial blink rates (C) Statistical analysis
with student t-test for data comparison between control and DED groups. (1). Full blink rate and eye score both showed p < 0.001 with statistical significance. (2). Partial blink and
age both showed p > 0.05 with no statistical significance. (3). When moving three “mild DED” patients from the control group to the DED group, both full and partial blink became
statistically significant (p < 0.05).
J. Clin. Med. 2023, 12, 6479 10 of 14

3.4. Accurate Blink and Partial Blink Results Obtained from the Dynamic EAR Approach
As shown in Table 4, a total of 20 videos from 10 confirmed DED (including four DED
patients under clinical treatments) and 10 normal controls (no DED diagnosis or symptoms)
were analyzed with the newly established dynamic EAR thresholds for full blink and partial
blink counts. The full blink counts from 20 videos were 100% matched with manual counts,
validating our dynamic EAR thresholds approach. Importantly, EyeScore can accurately
count partial blinks with the newly developed partial blink EAR threshold, which is much
more precise and efficient than the manual count. The accurate measurements of both
full blink and partial blink lay a solid foundation for reporting App-based eye healthiness
scores. The average full blink rate for control and DED groups were 22.2 and 37.4 blinks
per minute, respectively, while the average partial blink rate for control and DED groups
were 2.5 and 5.8 blinks per minute, respectively (Table 4). Although the partial blink rate
did not show significance statistically with a p value > 0.05, the three “mild DED” patients
with high partial blink rates in the control group technically interfered with the results of
Student’s t-test (Table 4). Once these three patients were moved to the DED group, the
partial blink rate became statistically significant (p < 0.05) between the control and the
DED groups.
The results from the EyeScore report confirmed all 10 known DED patients, with
four of them with scores ranging from 7 to 9 (severe DED category), while six of them
with scores ranging from 4 to 6 (Mild DED category). Interestingly, all four DED patients
with a high score of 7–9 showed longer daily electronic use time (DED 4~6, DED10 in
Table 1). On the other hand, four of six DED patients scoring 4~6 were undergoing clinical
DED treatment. Thus, the relatively lower eye healthiness score could be the treatment
results from the improved eye conditions. In comparison, 7 of 10 normal controls exhibited
eye healthiness scores from 0 to 3, confirming their normal eye healthiness conditions.
Importantly, 3 of 10 normal controls (i.e., patients No. 5, 6 and 7, with no previous DED
diagnosis from their eye doctors) demonstrated mild DED conditions. They either exhibited
significantly higher partial blink rates (in enrollees 5 and 6) or higher full blink rates
(enrollee 7) compared to other healthy individuals. As a result, their eye healthiness scores
were reported as 4, 6, and 5, respectively. Importantly, our post-EyeScore eye examinations
showed that these three participants had a shortened TBUT time (<10 s) and positive
fluorescent staining, confirming they were in the early phase of DED development (Table 4).
For patients with scores ≥ 4, the EyeScore App was designed to send an alert notification of
mild DED conditions to the user. These patients were recommended to change their daily
eye usage routines with healthy habits, such as taking breaks from extended computer use
and increasing outdoor activities.

4. Discussion
The multifactorial nature of DED pathogenesis has made its diagnosis and treatment a
significant challenge among common eye diseases [10,30]. Current DED diagnosis mainly
relies on clinical tests such as TBUT, OSDI, Schirmer’s test, and fluorescein staining, which
can only be performed in an eye doctor’s office [31]. This can significantly delay the early
diagnosis of DED, which often leads to irreversible chronic DED conditions. As a result,
a great number of DED patients remain undiagnosed and inadequately treated, partic-
ularly in our ever-expanding digitalized society. Thus, a POC, mHealth diagnosis tool
can bridge the gap and offer digital, low-cost, patient-centered solutions to the millions of
under-diagnosed DED patients [16,17]. Although the smartphone App DryEyeRhythm was
developed and tested for DED diagnostic assistance in Japan [18], our EyeScore App aims to
perform rapid, accurate, in-home DED diagnostic screening, detection, and treatment mon-
itoring. Most importantly, the eye healthiness score was uniquely created as a convenient
digital parameter for the real-time reporting of DED eye conditions. EyeScore mHealth
App can benefit both DED patients and their eye doctors with the following features.
First, eye blink and partial blink rates provide reliable biomarkers for early DED
diagnosis [12,13]. Comprehensive studies on eye blink patterns strongly support their
J. Clin. Med. 2023, 12, 6479 11 of 14

specific roles in the DED development process, either as a compensation mechanism for
supplying more tears to the inflammatory ocular surface or as resulting symptoms from
dry eye development. Thus, the proposed formula for eye healthiness score weighs heavily
on both full blink rate and partial blink, with a total of 6 points on the 10-point scale. The
unique approach for separated calculation of thresholds for full blink and partial blink
further ensures the accuracy of blink counts, which serves as the foundation for early DED
screening through the EyeScore mobile App. Our study strongly supports our eye blink
counts approach, with 8 of 10 known DED patients exhibiting >30 full blinks per minute. In
addition, five DED patients also demonstrated >5 partial blinks per minute. As a result, all
participating DED patients showed their eye healthiness score in the range of 4~9 (Table 4),
confirming that the EyeScore App and the eye healthiness score are excellent digital tools
for DED rapid screening and detection.
Second, the eye healthiness score was carefully designed to cover common causes or
symptoms of the multifactorial DED [26], with the inclusion of dry/gritty feelings (foreign
body sensation), gender/age and regular contact lens wearing [27]. Since our focus was to
perform App-based screening and early dry eye detection, the initial questionnaire was
designed to obtain objective information from the EyeScore users, such as age, gender and
eye conditions. Together with the accurate measurements of eye blink rate and partial
blinks, the combination of these factors (Table 2) can give an objective, clinically relevant
evaluation of eye conditions for rapid DED evaluation at home with low resource settings.
Third, the EyeScore App can be used to screen for patients with mild DED conditions at
the early stage of DED development. The fact that 3 of 10 individuals in our “normal” group
being identified having, with the post-EyeScore as confirmation, mild DED conditions
strongly suggests that DED is significantly under-diagnosed, with a large number of hidden
pre-DED and DED patients in the general population. Interestingly, the three mild DED
patients identified from our “normal group” were all females with their ages ranging from
47 to 56 years old (Table 4). This subcategory of the population is well known with the
highest prevalence of DED development [26]. Moreover, the use of electronics plays a
significant role in DED development [32]. More than 60% of DED patients in our study
have daily electronic screen time of 6 to >8 h (Table 1). With the widespread use of electronic
devices in our digitalized society, a POC, mobile diagnostic method such as the EyeScore
App is essential for early DED screening and prevention [17].
Lastly, the repeated in-home EyeScore exams over time can provide eye doctors with
the useful trend of individual eye healthiness conditions, thus facilitating timely DED
screening and detection at minimal cost. This EyeScore feature makes it possible for DED
patients to set routine tests with the real-time reporting of their eye conditions, similar to
other health parameters, such as BMI, heart rate and blood pressure. For DED patients
undergoing treatment (such as artificial tears, Omega-3 and warm compress), EyeScore also
allows doctors to remotely monitor DED treatment courses and make necessary changes of
regimens accordingly. In our study, four of ten DED enrollees were under various forms
of DED treatment (Table 4). All of them demonstrated decreased eye healthiness scores
over a 2-to-4-week period after treatment was initiated (internal data). Thus, the EyeScore
App can improve the prognosis of these DED patients without the need to visit eye doctors’
offices in person. In addition, the EyeScore mobile App can feasibly collect a large amount
of DED patient data that was previously inaccessible through traditional clinical methods.
Encouraged by these positive results, the advanced development of the EyeScore App
is also planned for this mHealth project, with possible machine learning algorithms and
statistical analysis to further optimize the accuracy of blink count and eye healthiness score,
with a particular focus on the identification of patients with pre-DED conditions.
Although the EyeScore App effectively identified all DED patients as well as three
patients with mild DED conditions from the control group, caution should be taken for
the data interpretation with our small sample size pilot study. Given the multifactorial
nature of DED pathogenesis and only three factors being considered for the current study,
our EyeScore results exhibited certain limitations and biases. In many cases, the DED
J. Clin. Med. 2023, 12, 6479 12 of 14

patients were under different clinical treatment methods, further complicating the data
interpretation. One possible solution is to perform a large trial study with three groups,
including DED untreated, DED treated, and controls. In addition, many other conditions
(such as smoking, alcohol, refractive surgery, glaucoma and autoimmune diseases) were
known to be important factors in DED pathogenesis [28]. These factors should be included
in eye health score calculations in the future when enough patient data from relevant
clinical subgroups becomes available with statistical significance.
The smartphone-based platform offers great clinical potential for future POC diag-
nostic methods [33]. As the patient data collection for digital recording grows along with
integrated user inputs, the eye blinking recognition signals can be further analyzed, quan-
titated, and categorized for fine-tuning the proposed algorithms [34]. Further machine
learning and artificial intelligence-based modeling can provide even deeper, hidden connec-
tions of eye blinking patterns of DED or other ocular diseases [35,36]. Such a data-driven
approach will ultimately lead to the establishment of DED diagnosis toward predictive,
preventive, personalized and participatory medicine [37].

Supplementary Materials: The following supporting information can be downloaded at: https:
//www.mdpi.com/article/10.3390/jcm12206479/s1, The source code of EyeScore can be provided
through GitHub. S1 showed the detailed step-by-step programming procedures of the EAR approach
of the EyeScore App.
Author Contributions: Study concept and design, S.Z.; analysis and interpretation of data, S.Z.
and J.E; statistical analysis, S.Z.; writing—original draft, S.Z.; writing—review and editing, S.Z.
and J.E.; access to data, S.Z. and J.E. All authors have read and agreed to the published version of
the manuscript.
Funding: This study was funded by Westview Eye Institute in San Diego, California.
Institutional Review Board Statement: Ethics approval and consent to participate: The study was
performed adhering to the tenets of the Declaration of Helsinki. The research protocol was exempted
because it was a retrospective study of archived videos. The personal information in all study videos
were deleted.
Informed Consent Statement: Written informed consent was obtained from all subjects involved in
this study.
Data Availability Statement: The data presented in this study are available on request from the
corresponding author. The recorded videos are not publicly available due to privacy restrictions of
study participants.
Acknowledgments: The authors would like to sincerely thank all the study participants for their
support. We also thank all research staff at Westview Eye Institute for their generous help during the
entire project.
Conflicts of Interest: The authors declare no conflict of interest.

Abbreviations
DED, dry eye disease; EAR, eye aspect ratio; iOS, iPhone operating system; MaxEAR, maximal
EAR value; mHealth, mobile health; MinEAR, minimal EAR value; POC, point of care; TBUT, tear
break up time; OSDI, ocular surface disease index.

References
1. Haupt, C.; Huber, A.B. How axons see their way—Axonal guidance in the visual system. Front. Biosci. 2008, 13, 3136–3149.
[CrossRef] [PubMed]
2. Pflugfelder, S.C.; Stern, M.E. Biological functions of tear film. Exp. Eye Res. 2020, 197, 108115. [CrossRef] [PubMed]
3. Rodriguez, J.D.; Lane, K.J.; Ousler, G.W., 3rd; Angjeli, E.; Smith, L.M.; Abelson, M.B. Blink: Characteristics, Controls, and Relation
to Dry Eyes. Curr. Eye Res. 2018, 43, 52–66. [CrossRef]
4. McMonnies, C.W. The clinical and experimental significance of blinking behavior. J. Optom. 2020, 13, 74–80. [CrossRef]
J. Clin. Med. 2023, 12, 6479 13 of 14

5. Tsubota, K.; Hata, S.; Okusawa, Y.; Egami, F.; Ohtsuki, T.; Nakamori, K. Quantivideo graphicsgraphic analysis of blinking in
normal subjects and patients with dry eye. Arch. Ophthalmol. 1996, 114, 715–720. [CrossRef]
6. Chidi-Egboka, N.C.; Jalbert, I.; Wagner, P.; Golebiowski, B. Blinking and normal ocular surface in school-aged children and the
effects of age and screen time. Br. J. Ophthalmol. 2022. [CrossRef] [PubMed]
7. Stapleton, F.; Alves, M.; Bunya, V.Y.; Jalbert, I.; Lekhanont, K.; Malet, F.; Na, K.S.; Schaumberg, D.; Uchino, M.; Vehof, J.; et al.
TFOS DEWS II Epidemiology Report. Ocul. Surf. 2017, 15, 334–365. [CrossRef] [PubMed]
8. Papas, E.B. The global prevalence of dry eye disease: A Bayesian view. Ophthalmic Physiol. Opt. 2021, 41, 1254–1266. [CrossRef]
9. Farrand, K.F.; Fridman, M.; Stillman, I.Ö.; Schaumberg, D.A. Prevalence of Diagnosed Dry Eye Disease in the United States
Among Adults Aged 18 Years and Older. Am. J. Ophthalmol. 2017, 182, 90–98. [CrossRef]
10. Messmer, E.M. The pathophysiology, diagnosis, and treatment of dry eye disease. Dtsch. Arztebl. Int. 2015, 112, 71–78. [CrossRef]
11. NIH/NEI. Dry Eye. 2023. Available online: https://www.nei.nih.gov/learn-about-eye-health/eye-conditions-and-diseases/
dry-eye (accessed on 11 September 2023).
12. Ousler, G.W., 3rd; Abelson, M.B.; Johnston, P.R.; Rodriguez, J.; Lane, K.; Smith, L.M. Blink patterns and lid-contact times in
dry-eye and normal subjects. Clin. Ophthalmol. 2014, 8, 869–874. [CrossRef] [PubMed]
13. Su, Y.; Liang, Q.; Su, G.; Wang, N.; Baudouin, C.; Labbé, A. Spontaneous Eye Blink Patterns in Dry Eye: Clinical Correlations.
Investig. Ophthalmol. Vis. Sci. 2018, 59, 5149–5156. [CrossRef] [PubMed]
14. Johnston, P.R.; Rodriguez, J.; Lane, K.J.; Ousler, G.; Abelson, M.B. The interblink interval in normal and dry eye subjects. Clin.
Ophthalmol. 2013, 7, 253–259. [CrossRef] [PubMed]
15. Inomata, T.; Iwagami, M.; Hiratsuka, Y.; Fujimoto, K.; Okumura, Y.; Shiang, T.; Murakami, A. Maximum blink interval is
associated with tear film breakup time: A new simple, screening test for dry eye disease. Sci. Rep. 2018, 8, 13443. [CrossRef]
16. Inomata, T.; Sung, J.; Nakamura, M.; Iwagami, M.; Okumura, Y.; Iwata, N.; Midorikawa-Inomata, A.; Fujimoto, K.; Eguchi, A.;
Nagino, K.; et al. Using Medical Big Data to Develop Personalized Medicine for Dry Eye Disease. Cornea 2020, 39 (Suppl. 1),
S39–S46. [CrossRef]
17. Inomata, T.; Sung, J. Changing Medical Paradigm on Inflammatory Eye Disease: Technology and Its Implications for P4 Medicine.
J. Clin. Med. 2022, 11, 2964. [CrossRef]
18. Okumura, Y.; Inomata, T.; Midorikawa-Inomata, A.; Sung, J.; Fujio, K.; Akasaki, Y.; Nakamura, M.; Iwagami, M.; Fujimoto, K.;
Eguchi, A.; et al. DryEyeRhythm: A reliable and valid smartphone application for the diagnosis assistance of dry eye. Ocul. Surf.
2022, 25, 19–25. [CrossRef] [PubMed]
19. Hirosawa, K.; Inomata, T.; Sung, J.; Nakamura, M.; Okumura, Y.; Midorikawa-Inomata, A.; Miura, M.; Fujio, K.; Akasaki, Y.;
Fujimoto, K.; et al. Diagnostic ability of maximum blink interval together with the Japanese version of Ocular Surface Disease
Index score for dry eye disease. Sci. Rep. 2020, 10, 18106. [CrossRef]
20. Fujio, K.; Nagino, K.; Huang, T.; Sung, J.; Akasaki, Y.; Okumura, Y.; Midorikawa-Inomata, A.; Fujimoto, K.; Eguchi, A.;
Miura, M.; et al. Clinical utility of maximum blink interval the measured by smartphone application DryEyeRhythm to support
dry eye disease diagnosis. Sci. Rep. 2023, 13, 13583. [CrossRef]
21. Nagino, K.; Okumura, Y.; Yamaguchi, M.; Sung, J.; Nagao, M.; Fujio, K.; Akasaki, Y.; Huang, T.; Hirosawa, K.; Iwagami, M.; et al.
Diagnostic Ability of a Smartphone App for Dry Eye Disease: Protocol for a Multicenter, Open-Label, Prospective, and Cross-
sectional Study. JMIR Res. Protoc. 2023, 12, e45218. [CrossRef]
22. Soukupova, T.; Cech, J. Real-Time Eye Blink Detection using Facial Landmarks. In Proceedings of the 21st Computer Vision
Winter Workshop, Rimske Toplice, Slovenia, 3–5 February 2016.
23. Rosebrock, A. Eye Blink Detection with OpenCV, Python, and Dlib. Blog in Pyimagesearch. 2017. Available online: https:
//pyimagesearch.com/2017/04/24/eye-blink-detection-opencv-python-dlib/ (accessed on 11 September 2023).
24. Hassan, M. Eye blink CountPythonng OpenCV Pythan/Computer Vision. 2022. Available online: https://www.youtube.com/
watch?v=-TVUwH1PgBs (accessed on 11 September 2023).
25. Dlib: Face Landmark Detection. Available online: http://dlib.net/face_landmark_detection.py.html (accessed on
11 September 2023).
26. Hakim, F.E.; Farooq, A.V. Dry Eye Disease: An Update in 2022. JAMA 2022, 327, 478–479. [CrossRef] [PubMed]
27. Kojima, T. Contact Lens-Associated Dry Eye Disease: Recent Advances Worldwide and in Japan. Investig. Ophthalmol. Vis. Sci.
2018, 59, DES102–DES108. [CrossRef]
28. Qian, L.; Wei, W. Identified risk factors for dry eye syndrome: A systematic review and meta-analysis. PLoS ONE 2022,
17, e0271267. [CrossRef]
29. Schneider, A.; Hommel, G.; Blettner, M. Linear regression analysis: Part 14 of a series on evaluation of scientific publications.
Dtsch. Arztebl. Int. 2010, 107, 776–782. [PubMed]
30. Kojima, T.; Dogru, M.; Kawashima, M.; Nakamura, S.; Tsubota, K. Advances in the diagnosis and treatment of dry eye. Prog.
Retin. Eye Res. 2020, 78, 100842. [CrossRef] [PubMed]
31. Thulasi, P.; Djalilian, A.R. Update in Current Diagnostics and Therapeutics of Dry Eye Disease. Ophthalmology 2017, 124, S27–S33.
[CrossRef] [PubMed]
32. Talens-Estarelles, C.; García-Marqués, J.V.; Cervino, A.; García-Lázaro, S. Use of digital displays and ocular surface alterations: A
review. Ocul. Surf. 2021, 19, 252–265. [CrossRef] [PubMed]
J. Clin. Med. 2023, 12, 6479 14 of 14

33. Hernández-Neuta, I.; Neumann, F.; Brightmeyer, J.; Ba Tis, T.; Madaboosi, N.; Wei, Q.; Ozcan, A.; Nilsson, M. Smartphone-based
clinical diagnostics: Towards democratization of evidence-based health care. J. Intern. Med. 2019, 285, 19–39. [CrossRef]
34. Babenko, B.; Mitani, A.; Traynis, I.; Kitade, N.; Singh, P.; Maa, A.Y.; Cuadros, J.; Corrado, G.S.; Peng, L.; Webster, D.R.; et al.
Detection of signs of disease in external photographs of the eyes via deep learning. Nat. Biomed. Eng. 2022, 6, 1370–1383.
[CrossRef]
35. Chase, C.; Elsawy, A.; Eleiwa, T.; Ozcan, E.; Tolba, M.; Abou Shousha, M. Comparison of Autonomous AS-OCT Deep Learning
Algorithm and Clinical Dry Eye Tests in Diagnosis of Dry Eye Disease. Clin. Ophthalmol. 2021, 15, 4281–4289. [CrossRef]
36. Storås, A.M.; Strümke, I.; Riegler, M.A.; Grauslund, J.; Hammer, H.L.; Yazidi, A.; Halvorsen PGundersen, K.G.; Utheim, T.P.;
Jackson, C.J. Artificial intelligence in dry eye disease. Ocul. Surf. 2022, 23, 74–86. [CrossRef] [PubMed]
37. Inomata, T.; Nakamura, M.; Sung, J.; Midorikawa-Inomata, A.; Iwagami, M.; Fujio, K.; Akasaki, Y.; Okumura, Y.; Fujimoto, K.;
Eguchi, A.; et al. Smartphone-based digital phenotyping for dry eye toward P4 medicine: A crowdsourced cross-sectional study.
NPJ Digit. Med. 2021, 4, 171. [CrossRef] [PubMed]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.