Ophthalmol Ther (2021) 10:445–464

https://doi.org/10.1007/s40123-021-00353-2

REVIEW

The Role of Telemedicine, In-Home Testing

and Artificial Intelligence to Alleviate an Increasingly
Burdened Healthcare System: Diabetic Retinopathy
Janusz Pieczynski . Patrycja Kuklo . Andrzej Grzybowski

Received: April 12, 2021 / Accepted: May 15, 2021 / Published online: June 22, 2021
Ó The Author(s) 2021

ABSTRACT the following key words: telemedicine in dia-

betic retinopathy screening, diabetic retinopa-
In the presence of the ever-increasing incidence thy screening, automated diabetic retinopathy
of diabetes mellitus (DM), the prevalence of screening, artificial intelligence in diabetic
diabetic eye disease (DED) is also growing. retinopathy screening, smartphone diabetic
Despite many improvements in diabetic care, retinopathy testing. We have included 118
DM remains a leading cause of visual impair- original articles meeting the above criteria, dis-
ment in working-age patients. So far, preven- cussing imaging diabetic retinopathy screening
tion has been the best way to protect vision. The methods. We have found that fundus cameras,
sooner we diagnose DED, the more effective the stable or mobile, are most commonly used for
treatment is. Thus, diabetic retinopathy (DR) retinal photography, with portable fundus
screening, especially with imaging techniques, cameras also relatively common. Other possi-
is a method of choice for vision protection. To bilities involve the use of ultra-wide-field (UWF)
alleviate the burden of diabetic patients who imaging and even optical coherence tomogra-
need ophthalmic care, telemedicine and in- phy (OCT) devices for DR screening. Also, the
home testing are used, supported by artificial role of smartphones is increasingly recognized
intelligence (AI) algorithms. This is why we in the field. Retinal fundus images are assessed
decided to evaluate current image teleophthal- by humans instantly or remotely, while AI
mology methods used for DR screening. We algorithms seem to be useful tools facilitating
searched the PubMed platform for papers pub- retinal image assessment. The common use of
lished over the last 5 years (2015–2020) using smartphones and availability of relatively
cheap, easy-to-use adapters for retinal pho-
tographs augmented by AI algorithms make it
J. Pieczynski (&)
P. Kuklo
A. Grzybowski possible for eye fundus photographs to be taken
Chair of Ophthalmology, University of Warmia and
by non-specialists and in non-medical setting.
Mazury, Zolnierska 18, 10-561 Olsztyn, Poland
e-mail: [email protected] This opens the way for in-home testing con-
ducted on a much larger scale in the future. In
J. Pieczynski
P. Kuklo conclusion, based on current DR screening
The Voivodal Specialistic Hospital in Olsztyn,
Olsztyn, Poland techniques, we can suggest that the future
practice of eye care specialists will be widely
A. Grzybowski supported by AI algorithms, and this way will be
Institute for Research in Ophthalmology, Poznan,
Poland, Gorczyczewskiego 2/3, 61-553 Poznan,
more effective.
Poland
446 Ophthalmol Ther (2021) 10:445–464

Keywords: Artificial intelligence in diabetic patients who present with early stages of dia-
retinopathy screening; Automated diabetic betic retinopathy. Therefore, a natural solution
retinopathy screening; Diabetic retinopathy to the problem of diabetic eye diseases seems to
screening; Smartphone diabetic retinopathy be diabetic retinopathy screening. The rules of
testing; Telemedicine in diabetic retinopathy screening in medicine were established in 1968
screening by Wilson and Jungner and were accepted by
the World Health Organization (WHO) [3, 4].
The basic principles for disease screening
Key Summary Points
include the following: the condition sought
should be an important health problem; there
The future role of the ophthalmologist in
should be an accepted treatment for patients
diabetic retinopathy (DR) care will be
with recognized disease; facilities for diagnosis
focused on consultations of difficult and
and treatment should be available; there should
complicated cases and their treatment.
be a recognizable latent or early symptomatic
Telemedicine augmented by artificial stage; there should be a suitable test or exami-
intelligence (AI) will make the DR nation; the test should be acceptable to the
screening system more effective and population; the natural history of the condi-
cheaper, with better coverage of the tion, including development from latent to
diabetic population. declared disease, should be adequately under-
stood; there should be an agreed policy on
The screening of DR will be done by eye whom to treat as patients; the cost of case
technicians, general practitioners or by finding (including diagnosis and treatment of
patients themselves supported by AI. patients diagnosed) should be economically
balanced in relation to possible expenditure on
medical care as a whole; and case finding should
be a continuing process and not a ‘‘once and for
all’’ project. Despite huge improvement in dia-
DIGITAL FEATURES betic retinopathy detection due to screening,
there is still a problem with efficiency. A useful
This article is published with digital features, tool for diabetic retinopathy screening is artifi-
including a summary slide, to facilitate under- cial intelligence (AI) used together with tele-
standing of the article. To view digital features medicine techniques.
for this article go to https://doi.org/10.6084/ The term ‘‘telemedicine’’ was defined in the
m9.figshare.14589417 1970s by Strehle and Shabde and meant ‘‘heal-
ing at a distance’’ [5]. WHO introduced a stan-
dardized definition of telemedicine as ‘‘the
INTRODUCTION delivery of healthcare services, where distance is
a critical factor, by all healthcare professionals
Diabetes is a prevalent global disease. According using information and communication tech-
to estimates, the number of diabetic people nologies for the exchange of valid information
worldwide was a staggering 415 million in 2015, for diagnosis, treatment and prevention of dis-
and this is still expected to rise and very likely to ease and injuries, research and evaluation, and
reach 642 million by 2040 [1]. Diabetic for the continuing education of healthcare
retinopathy (DR) is one of the leading causes of providers, all in the interests of advancing the
vision loss in working-age patients [2]. The dis- health of individuals and their communities’’
ease usually remains asymptomatic until visual [6]. Telemedicine relies on information and
acuity decreases, but in most cases, it can be communication technology (ICT), defined as a
detected with retinal imaging techniques even ‘‘diverse set of technological tools and resources
in its early stages. It is known that the best used to transmit, store, create, share or
treatment options and prognosis are for exchange information. These technological
Ophthalmol Ther (2021) 10:445–464 447

tools and resources include computers, the and was initially based on single-field Polaroid
Internet (websites, blogs, and emails), live fundus photographs, and later on digital single-
broadcasting technologies (radio, television, field retinal images [11].
and webcasting), recorded broadcasting tech- The aim of this paper is to evaluate currently
nologies (podcasting, audio and video players available imaging teleophthalmology schemes
and storage devices), and telephony (fixed or for the detection of diabetic retinopathy and to
mobile, satellite, visio/video-conferencing, discuss the existing screening possibilities as
etc.)’’ [7]. An innovative combination of well as the role of artificial intelligence as a
screening by means of fundus cameras, OCT diagnostic tool. The paper also aims to define
and other devices with telemedicine ushered in the advantages and disadvantages of this
the era of teleophthalmology, which could be examination method.
applied both in ophthalmology offices and in
non-eye care settings, including primary care
offices. This comes with the possibility of METHODS
remote grading and appropriate follow-up eye
care. Growing global enthusiasm for the use of We searched the PubMed database for papers
telemedicine in screening of diabetic retinopa- published over the last 5 years (2015–2020)
thy has led to the appearance of many publi- using the following key words: telemedicine in
cations over the last few years. diabetic retinopathy screening, diabetic
Of all diabetic patients worldwide, 75% live retinopathy screening, automated diabetic
in low- or middle-income countries [1]. For retinopathy screening, artificial intelligence in
patients living in rural environments, eye tele- diabetic retinopathy screening, smartphone
screening may sometimes be the only way to diabetic retinopathy testing.
gain access to professional examination and We selected 118 original English-language
treatment. Several studies found that this articles on the use of imaging methods in DR
method produced the same clinical results as screening which met the above criteria for
direct ophthalmological examination [8]. In inclusion in this paper.
public healthcare systems, the waiting time for This article is based on previously conducted
a professional ophthalmologist appointment studies and does not contain any new studies
can be very long, owing to shortage of special- with human participants or animals performed
ists and partially due to lack of cooperation by any of the authors.
between a patient and a doctor. Of all diabetic
patients, only 34.8% have ever received a dila-
ted fundus examination by an ophthalmologist
RESULTS
[9]. In England, diabetic retinal screening with
Techniques
two-field digital mydriatic photographic exam-
ination for all diabetic patients successfully
reduced the prevalence of blindness in working- Stable, Classic Non-mydriatic Fundus
age patients [10]. An additional advantage of Cameras
telemedicine is its safety of use during the SARS- Digital fundus cameras, usually non-mydriatic
CoV-2 pandemic, for both patients and medical ones, have become the classic screening diag-
professionals. We have previously reviewed nostic systems. These systems of non-mydriatic
current methods and programs used in diabetic fundus cameras can be stably mounted in one
retinopathy screening adopted in different parts location and were also the most popular in our
of the world [11]. One of the most efficient review. This is probably because they are rela-
national screening programs is in the United tively simple to operate, can be maintained by
Kingdom, where color fundus two- or one-field trained technicians, and so can be successfully
images are graded in specially dedicated grading used in the primary care setting. Non-mydriatic,
centers. The first diabetic retinopathy screening non-stereo cameras were also among the first to
program ever was started in Singapore in 1991 be used in diabetic retinopathy screening, and
448 Ophthalmol Ther (2021) 10:445–464

their use is well established in the field [12–60]. Optical Coherence Tomography (OCT)-Based
More accurate but at the same time more com- Diabetic Retinopathy Screening
plicated methods have involved the use of Some authors recommend the use of OCT
seven-field ETDRS standard stereoscopic pho- devices for more accurate and reliable detection
tographs of the eye fundus. Those images had to of macular edema [23, 44, 58, 72–76].
be taken with a wide pupil, by very experienced
photographers; grading was also technically Portable Fundus Cameras
more difficult [61–63]. Six pairs of stereo images The use of a portable fundus camera allows
of each eye were taken by Park et al. [64], three patients from rural parts of the country to be
pairs of stereo images were taken by Silva et al. tested for the presence of diabetic retinopathy
[65], and two pairs of two-field stereo images [77]. This method involves the use of a portable,
were taken by Bursell et al. [66] and by Mans- non-mydriatic, handheld, lightweight digital
berger et al. [67]. fundus camera with a 45° field of view. It is
supported by a trained nurse or technician
Mobile Classic Non-mydriatic Fundus using an educational pamphlet translated into
Cameras (moved from location to location) local languages, highlighting the importance of
A more effective solution, generating lower regular eye screening. These images can be used
costs and allowing for wider coverage (up to in initial screening for the presence of disease,
70% of the diabetic population), is the classic and can help determine whether patients
non-mydriatic fundus camera which can be require a dilated exam by an ophthalmologist
moved from one location to another place of [32, 78–81]. Web-based portable fundus cameras
examination. This usually follows a previously were presented by Keshvardoost et al. [82].
planned scheme [68]. Zhang et al. [83] concluded in their study that
handheld fundus cameras were sufficiently
Mobile, On-Vehicle Hard-Mounted Diagnostic efficient in DR screening.
Sets
In order to optimize the use of current diag- Smartphone-Based Retinal Imaging
nostic resources, diagnostic sets can be hard- Smartphones are nowadays the most common
mounted on vehicles. Ultra-wide-field imaging portable devices used worldwide. In ophthal-
is more accurate than classic non-mydriatic mology they can be used for ophthalmological
cameras and, when hard-mounted on a vehicle, examination. There are several types of smart-
is an even more perfect screening set [69]. phone adapters, making it possible to see and
Mobile screening sets have even been specially take a picture of the eye fundus, with or without
named ‘‘virtual clinics’’ and are based on ultra- pupil dilation; these include Peek Retina [84]
wide-field and classic non-mydriatic cameras and D-Eye adapter, adapters for smartphones
[50, 70]. Mobile DR screening vans have also such as the iExaminer, or all-in-one devices
been used in China to take care of diabetic such as the Horus Scope and Smartscope PRO
patients [27]. [85], a wide-field smartphone fundus video
camera, and CellScope Retina, a retinal imaging
Ultra-Wide-Field (UWF) Diagnostic Sets system [87, 88]. Smartphones are also used for
A more effective and more accurate method of monocular indirect ophthalmoscopy with
diabetic retinopathy screening is assessment by application of the 20D Volk lens, and a plastic
means of ultra-wide-field fundus cameras, in adapter to hold the lens in one piece with the
some cases combined with macular OCT, to phone (EyeArt) [89]. Various healthcare workers
improve the detection of macular edema can potentially operate a smartphone-based
[41–44, 41–44, 36–39, 67, 69, 71, 72]. UWF as a retinal imaging device [84, 89, 90], as their use
stereo pair was reported by Silva et al. [65]. is not restricted to qualified staff [84, 86–88, 90].
The effectiveness, defined as the number of
changes in the posterior segment identified by
Ophthalmol Ther (2021) 10:445–464 449

ophthalmological technicians as compared to from place to place. Diabetic retinopathy grad-

ophthalmologists, was found to differ in 42.4% ing is difficult when based on pictures taken
[90]. However, in terms of examination of the with this camera because of their quality, but it
anterior part of the eye and qualification for is a useful instrument to detect changes in the
cataract surgery, as well as the diagnosis of fundus. The available piggyback camera has a
absence of any changes in the eye, the effec- 45° field of view, 33 mm working distance, and
tiveness is more comparable between techni- ?20 to –20 D adjustment; it has optical mag-
cians and ophthalmologists (k = 0.732, 95% CI nification of 912, and its battery lifetime is up
0.65–0.81; k = 0.642, 95% CI 0.53–0.75) [90]. to 7 h. A special application allows patients to
The patients themselves or their families can store data in a folder. Pictures can be sent for
also perform retinal imaging with a smartphone evaluation using the WhatsApp application,
camera as in-home testing [86, 87]. The only which is a free and encrypted application [7].
condition is the need to buy a special adapter Smartphone-based semiautomatic DR detec-
for a smartphone that is lightweight and easy to tion was also proposed by Saeed et al. [97], using
use; no other changes in the telephone are the d-Eye sensor mounted on a mobile phone to
necessary. Good-quality pictures can be easily give patients the possibility of in-home testing.
shared via telemedicine and analyzed remotely The obtained images are sent via the web to an
by an ophthalmologist or other trained grader algorithm which detects hard exudates.
[84, 86–88]. As a diagnostic and monitoring Another example of smartphone use in DR
tool, smartphone detection can be particularly telescreening was proposed by Toy et al. [63],
important in rural areas and in developing who used a special adapter mounted on a
countries, where access to healthcare and qual- mobile phone to obtain retina pictures, which
ified staff is limited and medical equipment is could be assessed remotely.
lacking [90, 91]. Children can also be easily Smartphone testing is usually easy and can
examined with a smartphone device, which is be performed by non-specialists and in a non-
lightweight and can be easily moved in front of professional setting, so it can take the form of
a person’s eye [84, 88, 92, 93]. The quality of in-home testing.
smartphone imaging is lower than that of con- Methods of retinal image acquisition are
ventional fundus imaging. More artifacts presented in Table 1.
appeared in smartphone imaging when the
mid- and far periphery of the retina was inclu- The Process of Photograph-Taking
ded in pseudophakic eyes [94]. The quality of
images depends on the type of smartphone Several DR telescreening models have been
adapter, but not on the experience of the implemented. Retinal images are generally
investigator [91]. It is possible to take a picture acquired by trained primary care physicians,
of the fundus without dilation of the pupil, but technicians, ophthalmic photographers and
it is more time-consuming to obtain a clear nurses [7, 10, 12, 13, 15, 17, 18, 20, 22, 23,
image [94]. A picture of the optic disc and the 25–27, 27–31, 53–60, 63–65, 68–72, 51–53,
macula can be taken without special adapters, 53, 97–101, 57, 58, 58, 59, 26–30, 64, 64, 65,
after pupil dilation, using the 20D Volk lens in 65, 48–55, 58–60, 63–67, 81, 82, 82, 83, 79–83].
front of the patient’s eye [95]. A video can be
taken and then a screenshot from it or a direct
Grading Methods
picture during examination [95]. Smartphone
imaging is easier to perform by an unqualified
examiner compared to direct ophthalmoscopy Human Grading
(82.3% vs. 48.5%, P \ 0.0001) [96]. Another The data collected in medical centers from dia-
device is a special piggyback-design Fundus on betic patients during camera-based fundus
Phone retina camera, which is fitted to com- examinations are sent for diabetic retinopathy
mercially available smartphones. It is a low- detection to experts including ophthalmolo-
weight camera (900 g), so it can be easily carried gists or retinal specialists [13, 16, 18, 20, 21,
450 Ophthalmol Ther (2021) 10:445–464

Table 1 Methods of retinal image acquisition

Method Main features References
Stable fundus camera Fundus camera (narrow angle) located in one place [56–62]
Mobile fundus camera Fundus camera (narrow angle) moved from location to location; [63]
mounted in offices; more effective
Mobile, on-vehicle hard- Diagnostic sets (fundus camera [narrow- and wide-field] and [22, 45, 64, 65]
mounted diagnostic sets software) hard-mounted on specially adapted vehicles; moved
from one location to another
Ultra-wide-field (UWF) Ultra-wide fundus cameras hard-mounted in offices or on [36–39, 60, 61, 64, 66, 67]
diagnostic sets vehicles
OCT based diabetic Screening with use of OCT (method of choice for diabetic [18, 39, 53, 67–71]
retinopathy screening macular edema)
Portable fundus cameras Screening with portable handheld camera (low cost, possible in- [27, 72–78]
home testing)
Adopted smartphones Screening with commercially available smartphones with proper [44–53, 58, 72, 79]
adapters (low cost, possible in-home testing

23, 24, 26–29, 32, 33, 35, 42–45, 47–52, 54, future. AI can be divided into three types:
56, 64, 67–75, 79, 81–83], certified technicians supervised, semi-supervised and unsupervised
and nurses; in more complicated cases, the data [100]. The taxonomy is based on the possibility
are reevaluated by retinal specialists [12, 15, 20, of adjustment of some parameters during the
31, 41, 44, 59, 63, 66, 78, 102]. training phase, which are collected in response
To reduce problems with under- or over- to a specific performance, until acceptable com-
evaluation, two or more readers can participate pliance is achieved. AI can use multiple retinal
in examinations [53, 80]. There can also be two detectors to find special features on the retina
or more masked reviewers, who check the same which were pathognomonic for diabetic
data, and a third reviewer or more for incon- retinopathy, for example hemorrhages, exu-
sistent decisions [18, 52, 80]. Prior to com- dates, microaneurysms, and nonlinear remod-
mencement of work, some readers have eling of the outer retinal layers. The detected
completed a special training program or changes are then classified by the system as
obtained a certificate for retinal evaluation normal or abnormal and the final output is
[17, 18, 48, 52]. generated. This type of AI is categorized as
lesion-based, because it is used for detecting
Artificial Intelligence relevant abnormal lesions. Artificial intelligence
AI is a technology created to mimic the per- can also integrate one or more multilayer neural
ception and information processing of the networks which are trained to associate diag-
human brain by a machine to make objective nostic outputs on disease level, for instance for
decisions [73]. It can improve the quality and a retinal image. Each image is analyzed and
efficacy of ophthalmic examinations as well as compared to a large corresponding output in a
reduce the costs. Over the past 5 years, artificial training set. Based on this pixel data of changes
intelligence systems have undergone dynamic characteristic of diabetic retinopathy, the sys-
changes and are expected to be evaluated in the tem learns how to grade images. In most
Ophthalmol Ther (2021) 10:445–464 451

studies, AI is based on color fundus examina- The EyeArt system, a CE-marked IIa medical
tion imaging [73]. Optical coherence tomogra- device developed by Eyenuk, Inc., Los Angeles,
phy (OCT) and OCT angiography (OCTA) have CA, USA, is cloud-based with telemedicine
also been integrated into AI systems, but it was software. The algorithm automatically excludes
very challenging to create a database for a large, pictures of inadequate quality and offers the
multicenter system [73]. possibility for macroaneurysm turnover assess-
In order to classify images using artificial ment. Its screening sensitivity is 91.7% (95% CI
intelligence, various quantitative parameters 91.3–92.1%) and specificity is 91.5% (95% CI
are needed. In one study, OCTA vessel maps 91.2–91.7%) [5]. Eyenuk, Inc. also offers
and skeletal maps were extracted from OCT another algorithm, EyeMark, for macroa-
scans and vascular features were measured, neurysm turnover assessment. This software can
including blood vessel tortuosity (BVT), blood also work on smartphone app-based images
vascular caliber (BVC), vessel perimeter index (tested on a Remidio Fundus on Phone device),
(VPI), blood vessel density (BVD), foveal with 95.8% sensitivity for any DR and speci-
avascular zone (FAZ) area (FAZ-A) and FAZ ficity of 80.2% [6].
contour irregularity (FAZ-CI) [73]. All these Google Inc. developed a convolutional neu-
parameters were recalculated and compared ral network-based algorithm for automatic DR
between healthy patients and patients diag- detection [7]. This system can be tweaked for
nosed with diabetic retinopathy and sickle cell higher specificity (93.9%) and sensitivity
retinopathy [82]. The optimal feature combi- (96.1%) for referable DR prediction.
nation directly correlated with the most sig- Singapore SERI-NUS, presented by Ting et al.
nificant morphological changes in the retina. [109], is a deep learning-based algorithm for DR
It also had limitations, however, including detection, with sensitivity of 90.5% and speci-
severe artifacts, segmentation errors and ficity of 91.6% for referable DR. This software
errors in reconstruction, which had to be was also used for the detection of suspected
identified and eliminated. glaucoma and age-related macular degeneration
The IDX-DR was developed from the Iowa (AMD).
Detection Program (IDP) and is based on con- The Bosch DR algorithm is a convolutional
volutional neural networks [103–105]. Through neural network-based AI system used on the
the addition of new deep learning features to Bosch Mobile Eye Care fundus cameras with the
IDP, specificity improved from 54.9% for IDP to following output: disease/no disease and picture
87% for IDX-DR, while the high sensitivity of quality assessment. This system showed sensi-
IDP and IDX-DR remained unchanged [103]. tivity of 91% and specificity of 96% [61].
This AI algorithm also achieved Food and Drug RetinaLyze, a CE-marked Class I device soft-
Administration (FDA) approval as the first FDA- ware system with end-to-end encryption, is an
approved fully autonomous AI diagnostic sys- algorithm for automatic eye fundus image
tem [106]. This algorithm is designed to work analysis for DR, glaucoma and AMD detection
with the Topcon non-mydriatic NW400 fundus with website-based assessment. Its sensitivity
camera to take macula- and disc-centered pic- for DR detection is 93.1% and specificity is
tures of each eye. 71.6% [110–112].
Retmarker DR is a Portuguese machine Gargeya and Leng presented their DR detec-
learning algorithm for DR detection as ‘‘disease’’ tion deep learning-based algorithm [113]. They
or ‘‘no disease,’’ and requires subsequent human used this system both on desktop hardware and
verification [107]. This system is able to com- on an iPhone 5 and achieved sensitivity of 93%
pare current retina pictures with previously and specificity of 87% for disease and no disease
taken pictures and to assess worsening or diagnosis.
improvement of DR state [108]. Its sensitivity Li et al. tested their deep learning-based
for referable DR is 85%. This algorithm is a CE- algorithm for DR detection. They presented
marked Class IIa medical device. external validation tests for referable DR with
92.5% sensitivity and 98.5% specificity [114].
452 Ophthalmol Ther (2021) 10:445–464

Alam et al. [73] proposed a supervised for image evaluation, with shorter time for bad
machine learning-based approach in which a image quality and longer time for good quality.
support vector machine (SVM) classifier was The current results can be accessed at http://
trained to evaluate diabetic retinopathy with demo.dicoogle.com/screen-dr.
multi-task AI classification, using quantitative Walton et al. [60] tested the efficacy of an
OCTA features. There were three steps to automated algorithm based on a neural network
achieve validation of the OCTA image. The first (Intelligent Retinal Imaging System, IRIS) for
was image data acquisition. The second step was DR detection, reported as ‘‘referral’’ or ‘‘obser-
the hierarchical backward elimination tech- vation.’’ The authors concluded that their soft-
nique supported by SVM, which was used to ware had high sensitivity (66% compared to
identify an optimal feature combination for the human grading), with specificity of 72.8% and a
best diagnostic accuracy and the most efficient very low rate of false negatives (2%), so the
classification. As the third step, multilayer diagnosis could be an effective alternative to
hierarchical tasks were performed to create human grading.
classification of normal retina and disease,
inner disease classification for diabetic Remote Grading
retinopathy and sickle cell retinopathy, and The obtained retinal image data were assessed
grading of each disease. remotely. It was possible to transfer the data via
Stevenson et al. [115] created a convolu- web and receive immediate online diagnosis
tional neural network-based algorithm for color [25, 29]. There was also a special teleretinal
fundus photo-based detection of DR, glaucoma screening software platform used [16, 22, 26, 27,
and AMD. The average sensitivity for each dis- 29, 30, 35, 37, 44, 47, 51, 54,
ease was 75% and average specificity was 89%. 56, 59, 65–67, 69, 70, 72, 75, 81, 97] or, alter-
Kanagasingam et al. [22] checked the use- natively, access to all information could be
fulness of AI in diabetic eye fundus photograph obtained through a special account on a medi-
grading in Australia. They created their own AI cal platform.
system for (1) the detection of DR versus no DR, The length of time the patients had to wait
(2) the detection of hard exudates and microa- for the result of the examination was also
neurysms, and (3) assessment of the severity of important. It could be obtained almost in real
DR, based on the International Clinical Diabetic time [77], or it could be minutes [25, 29], hours
Retinopathy Disease Severity Scale criteria. They or days [38]. If direct contact was not possible
achieved 92% specificity and a positive predic- through a teleretinal platform, the patients had
tive value of 12%. to wait for their examination results, and they
Saha et al. [24] proposed their own AI (deep were informed whether an additional appoint-
convolution neural network) system to ment in an ophthalmology center was needed.
instantly check the quality of eye retina pictures This could be done by phone, letter or email
meant for telescreening of DR, which saved sent to the patient and to their primary care
time waiting for the final diagnosis by elimi- physicians [38] or just to their primary care
nating the need for repeated examinations. The doctor; the results were then shown to the
results were obtained as ‘‘accept’’ or ‘‘reject’’ an patient during the next follow-up appointment.
image. The ‘‘reject’’ result meant a retake of the
fundus picture. The authors achieved 97% Limitations and Positive Aspects
agreement between the AI algorithm and con- of Telescreening of DR
trol human grading (accuracy of 100%).
Pedrosa et al. [57] presented a multidisci-
There were many limitations related to tele-
plinary collaborative platform based on
medicine highlighted by numerous researchers.
machine learning to create an algorithm to help
The authors of several publications reported
diagnose DR by evaluating image quality, dis-
some problems in obtaining retinal images of
carding healthy eyes and DR grading. The
good quality. Reasons for that included a small
results were presented as average time needed
Ophthalmol Ther (2021) 10:445–464 453

pupil of the examined eye and the need to take problems, distrust of the recommended hospi-
images after mydriasis [25, 82]. This was why tal, making the referral appointment, fear of
some authors suggested screening with a wide examination and treatment, no response to
pupil [14, 41]. Another cause of low-quality previous treatment and no sensation of being ill
photographs was poor transparency of optic [28, 29, 56, 116–118]. In an attempt to reduce
media caused by cataract and other opacities the negative effects of poor education, tele-
[13, 14, 21, 23, 25, 80, 82, 97–100]. screening was combined with diabetes educa-
There were also significant problems regard- tion, which improved diabetes control
ing telemedicine programs associated with a [24, 26, 39].
growing demand for trained/certified eye fun- There were also problems with telescreening
dus photographers [81] and image readers for DR, such as the need for expensive and
[18, 63]. As a solution to that problem, training complex systems to detect all stages of DR [44]
in picture-taking could be offered to nurses and or a long waiting time for the final diagnosis
technicians [12, 15, 18, 22, 23, 27, 31–34, after image acquisition [38, 99]. Here, AI algo-
38, 42–45, 47, 48, 50, 52–57, 60, 63, 64, 68, rithms could offer a solution as an effective
70–72, 74, 79–81, 98–100]. One promising alternative to human grading [14, 43, 60, 73].
solution was in-home testing, where patients Automated systems are cost-effective [40] and
themselves took pictures of their retinas [97]. can check the image quality instantaneously,
Grading could then be performed by certified thereby saving money and time otherwise nee-
technicians and nurses [71, 101, 102] or by AI, ded for follow-up appointments [14, 24, 61].
replacing retina specialists [14, 43, 60, 73, 100]. Those systems reduced waiting time for DR
The need for advanced and expensive screening, provided more regular screening
screening devices, such as fundus cameras and [14, 29, 30, 43, 78] and were highly effective
special software, was quite a serious problem (100,000 patients in 45 h) [14, 43]. Some
[14, 17, 46, 65, 73, 78, 82, 99]. Portable fundus authors noted the high quality of handheld
cameras [32, 48, 82] and complete mobile sets fundus cameras [32, 48, 83]. Virtual (driven)
could lower the total cost of telescreening by clinics (mobile set) offered the standard of eye
optimizing currently available resources care of a clinical setting and allowed optimiza-
[28, 52, 69, 70, 72]. However, some studies tion of currently used resources, and so reduced
noted limited sensitivity and specificity of the total cost of telescreening
smartphone examinations and lack of applica- [28, 52, 69, 70, 72]. Finally, some researchers
tions dedicated to screening. AI helped in the noted the lack of an integrated virtual platform
use of low-quality portable screening devices for which could bring all the necessary data toge-
high-quality assessments [117]. Smartphones ther [75].
needed a wide pupil for good-quality images Some authors found no positive influence of
[38, 63, 73] teleretinal diabetic retinopathy screening on
Some authors suggested that classic 45° sin- patients’ access to eye examination [15], and
gle-, two- or three-field images were insufficient others noted limited coverage of telemedicine
to detect all diabetic retinopathies [13, 14, 16, and the need for co-payment for some exami-
20, 27, 28, 35, 38, 45, 47, 49, 51–54, 62, nations [16]. Many studies reached quite oppo-
66, 70, 73–75, 80–82, 99, 100], and ultra-wide- site conclusions and mentioned high accuracy
field imaging (200° angle of view) was needed in diabetic retinopathy diagnostics, especially
for effective detection [41, 44, 46, 65, 66, 69, in treatable patients [21, 54, 79, 82]. Many
71, 72, 74]. authors pointed to the significant role of
A serious problem related to telescreening screening in the prevention of unnecessary
was a low percentage of follow-up of vision- referrals (reduction by 75%) [14, 19, 30, 73, 82].
threatening diabetic retinopathy (VTDR) due to Diabetic retinopathy screening covered close to
social and educational factors such as low edu- 70% of diabetic patients in rural regions
cation level, limited elderly mobility, transport [13, 14, 68, 72] and improved the quality of
difficulties, loneliness, depression, financial medical care, especially in remote areas
454 Ophthalmol Ther (2021) 10:445–464

[25, 33, 72, 78, 99]. Studies found that a higher DISCUSSION
percentage of diabetic patients received eye care
with telemedicine compared to traditional The predicted 50% rise in diabetes prevalence
surveillance [13, 14, 17, 59, 67, 79, 118, 119], by 2040 [1], particularly in developing coun-
along with a high satisfaction rate among tries, will lead to an increased demand for
screened patients [26, 47, 62] and among pro- ophthalmological testing in diabetic patients.
fessionals conducting the examinations This in turn will put a greater burden on the
[26, 47]. system of ophthalmic care that relies mainly on
Recommendations for further examinations direct examination, and will significantly chal-
(follow-up) were assessed [16, 24, 32, 35, 39, lenge its efficiency. Screening of diabetic
47, 54, 55, 71]. Better-educated patients pre- retinopathy is commonly known to be the best
ferred comprehensive eye examination to fun- method for preventing serious complications of
dus camera screening test only [28], and some diabetes. If screening is to cover the whole
authors concluded that telemedicine should be population, an efficient screening system is
targeted toward patients with poor access to needed, together with a well-functioning spe-
medical care [23, 120]. Non-ocular diseases and cialist care structure to treat more advanced
ocular diseases found incidentally by means of diabetes-related ophthalmic complications. Teo
DR telescreening and AI included AMD, glau- et al. found the prevalence of VTDR to be
coma, hypertensive retinopathy and disc pallor 7.26%, ranging from 14.3% in Africa to almost
[21, 28, 34, 38, 45, 49, 50, 55, 64, 75, 81]. 2.97% in Southeast Asia. The authors also cal-
In 2017, Xiao et al. presented a comprehen- culated an average of 7.16 ophthalmologists per
sive teleretinal eye treatment plan with SMS 1000 patients with VTDR globally, with signifi-
[short messaging service] patient information cant differences depending on the region. In
on the date of eye examination. To improve the rich Europe, the average number of ophthal-
quality of telescreening, internal self-checking mologists was 18.03, while in poor regions of
of image system quality was assessed [99], and a Africa the figure was as low as 0.91 ophthal-
development scheme was used to design a DR mologists for every 1000 VTDR patients [121].
telemedicine screening program [37]. To Screening may be performed by humans
improve access to DR screening, local women’s only (e.g. an English model [126]), it may rely
self-help groups were tasked with its imple- on remote image analysis using AI
mentation [74]. [103–105, 108], or it may be based on a mixed
According to some authors, the diagnosis of model, using both humans and AI [107].
diabetic macular edema (DME) should not be Screening tests may be conducted directly by an
made only on the basis of color fundus pho- ophthalmologist (a method characterized by
tographs, but should also include OCT scans high accuracy, but requiring considerable time
[76]. Screening performed with OCT provided and specialists, which now and in the future
diagnostic characteristics of OCT such as offers no possibility for screening the whole
epiretinal membrane (ERM), glaucoma, AMD diabetic population). A solution to this problem
and vitelliform degeneration [34]. On the other is an examination conducted with color fundus
hand, maintenance of fundus camera/OCTA images or OCT, further assessed by ophthal-
was time-consuming [73, 82], AI software was mologists, i.e. using telemedicine (the method
not integrated with OCT devices [34, 73], and AI recognized by WHO), which in this case is
for OCTA/retinal images needed a huge data- teleophthalmology. This is when a limited
base of 100,000–1,000,000 scans [14, 73]. number of specialists provide care for a much
The positive and negative aspects are pre- larger patient population. In developed coun-
sented in Tables 2 and 3. tries, screening programs receive substantial
funding, and consequently their efficiency is
higher; they mostly use stationary diagnostic
centers, offering care to a larger part of the
Table 2 Limitations of diabetic retinopathy screening
Limitation Reason Solution References

Poor quality of images Small pupil Mydriasis [9, 20, 36, 77]
Poor transparency of optic media Cataract extraction [8, 9, 16, 18, 20, 75, 77, 79–82]
Screening program organizational Need for trained photographers, graders and Training of technicians, nurses and [7, 9, 10, 12, 13, 17, 18, 20, 22, 26–29, 33,
problems retina specialists general practitioners 37–40, 42, 43, 45, 47–52, 55, 58,
Ophthalmol Ther (2021) 10:445–464

In-home testing with self-preliminary 59, 63, 65–68, 74–76, 76, 80–82]
images reading
AI grading
High cost of screening Expensive screening devices and software, Mobile screening sets [9, 12, 23, 27, 33, 41, 43, 47, 58, 60,
crew costs Cheap portable cameras 64, 65, 67, 68, 73, 77, 81, 86]

Smartphone screening
Telescreening
AI-assisted screening
Poor sensitivity of DR detection 1-, 2- or 3-field images- with too small Ultra-wide fundus cameras use [8, 9, 11, 15, 22, 23, 30, 33, 36, 39–42, 44, 46,
coverage of retina 47, 49, 51, 57, 60, 61, 64–70, 75–77, 81, 82]
Low percentage of follow-up Social and educational factors Basic diabetic education [19, 21, 23, 24, 34, 51, 86, 87, 89]
No positive results of telescreening Small widespread in population Diabetic education [10, 11]
Need for co-pay Better social insurance
The more advanced the diagnosis, the Expensive and complex screening schemes Common use of AI [39]
more expensive
Long waiting time for final diagnosis Insufficient screening system AI grading [9, 33, 38, 55, 68, 81]
Lack of an integrated virtual platform Lack of proper software Development of screening software [70]
for DR screening
455
456 Ophthalmol Ther (2021) 10:445–464

Table 3 Positive aspects of diabetic retinopathy screening

Positive aspect References
High accuracy in diabetic retinopathy diagnosis [16, 49, 74, 75]
Prevention of unnecessary referrals (reduction by as much as 75%) [9, 14, 25, 68, 77]
High percentage (70%) of screening of diabetics in rural regions [8, 28, 63, 67]
Improved medical care in remote regions [20, 28, 67, 73, 81]
More diabetics receiving eye care with telemedicine than with traditional [8, 9, 12, 54, 62, 74, 89, 91]
direct care scheme
High satisfaction rate of screened patients [21, 42, 57]
High satisfaction rate of medical staff [22, 42]
Recommendation for follow-up given to diabetics [11, 19, 27, 30, 33, 34, 39, 42, 47, 49, 50, 66]
Diagnosis of non-ocular problems or other incidental eye findings [16, 23, 29, 33, 40, 44, 45, 50, 59, 70, 76]
Computer-assisted screening more cost-effective [9, 19, 35, 56]
Highly effective AI-assisted process of image reading [9, 38]
Optimization of current resource use and lower total costs of telescreening [23, 47, 64, 65, 67]
than when virtual (driven) clinics were used

diabetic population (the UK, Denmark, the wide-field imaging and OCT use, supported by
Netherlands, Singapore, Sweden). In developing AI algorithms, may be used to detect not only
countries, however, where the costs of imple- DR but also co-occurring diseases such as AMD,
menting efficient screening models for DR are glaucoma, retinal degeneration of another eti-
relatively high, various forms of mobile diag- ology, choroidal nevi and intraocular tumors.
nostic units are used to reach patients [11]. The disadvantages of telemedicine include
A classic definition of telemedicine proposed its technical limitations (it is not possible to
in the 1970s by Strehle and Shabde should now investigate all aspects of the disease and verify
be modified to encompass the use of AI. Based them remotely); other limitations may include
on the assumption that the classic form of the initial costs of the diagnostic equipment,
teleophthalmology involves a remote software, training and staff-related costs.
patient–ophthalmologist relation, the concept Another problem may be the provision of
of telemedicine should be expanded to include screening services on a daily basis, such as
an indirect contact between a patient and the identifying patients and making their first and
doctor of another specialty, such as a general follow-up appointments, and patient post-
practitioner (GP), or even a trained technician, screening compliance.
and only later, if the situation requires, an A good solution for reducing the high costs
ophthalmologist. Similarly, screening sup- of DR screening is customizing intervals
ported by AI for the processing of eye fundus between routine check-up appointments.
images (computer-aided diagnostics) which are According to the standard recommendation, for
submitted to ophthalmologists only in justified a diabetic presenting no eye problem, a check-
cases can also be regarded as a form of tele- up should take place every year or every 2 years,
medicine, although mostly implemented but on evaluation of DR risk factors such as
locally, without sending patient’s data (online blood glucose levels, blood pressure, gender,
access for software). Telemedicine making use type of DM and duration of DM, the interval
of advanced imaging techniques such as ultra- may be extended even to 4 years [11, 122]. The
Ophthalmol Ther (2021) 10:445–464 457

RetinaRisk.com platform has successfully CONCLUSIONS

implemented this solution, providing efficient
service and safety for patients. It reduces The future of ophthalmic care seems to be
screening costs by extending the length remote eye examination done by a technician
between examinations to more than 1 year for and aided by AI, with in-home self-testing as a
mild stages of DR. This helps to reduce the preliminary test. Pre-specialist diagnostics will
burden of unnecessary ophthalmologist automatically identify healthy eyes, make a
appointments. diagnosis and recommend a referral only if
In the era of common use of smartphones necessary. This will make the system less
and special adapters with relatively cheap soft- dependent on highly specialized and expensive
ware, this technique facilitates self-screening of human resources and will indirectly lead to
DR (in-home testing) [86, 87, 97], at least as optimal use of current resources.
preliminary testing. Therefore, it would make
access to specialist eye care much easier for
middle- and low-income patients. In-home
testing, in the time of the COVID-19 pandemic,
ACKNOWLEDGEMENTS
could become a method of choice for DR
screening.
Funding. No funding or sponsorship was
Telemedicine platforms may fill many of the
received for this study or publication of this
gaps in access to DR screening, offering a cost-
article. The Journal’s Rapid Service Fee was
effective way to detect preventable disease in a
funded by the authors.
primary care setting [123–125].
However, improvements in retinal screening Authorship. All named authors meet the
do not consistently lead to the prevention of International Committee of Medical Journal
blindness from VTDR. In some settings, as many Editors (ICMJE) criteria for authorship for this
as 80% of patients found to have VTDR through article, take responsibility for the integrity of
retinal screening will not complete follow-up the work as a whole, and have given their
ophthalmological evaluation and treatment approval for this version to be published.
recommendations [11].
Kozak et al. [98] used telemedicine in their Authorship Contributions. Janusz Piec-
study not only to evaluate retinal images of zyński: PubMed search, analysis of articles,
diabetic patients; they went a step further and manuscript design and preparation, drafting of
planned laser therapy for patients with macular the manuscript. Patrycja Kuklo: PubMed search,
edema. In a referral center, grid patterns were analysis of articles, manuscript preparation.
planned on the pictures sent for each patient Andrzej Grzybowski: study concept, design and
and sent back to the examination center to supervision, drafting of the manuscript.
perform treatment. To facilitate the process of
sending data between the diagnostic centers, a Disclosures. Janusz Pieczyński, Patrycja
special web-based platform was used. This made Kuklo and Andrzej Grzybowski have nothing to
it possible to send, store and visualize pho- disclose.
tographs. Patient history and medical pictures
collected in the medical center were sent via the Compliance with Ethics Guidelines. This
web to an evaluator for diagnosis and referral article is based on previously conducted studies
[13]. Personal access was required to approach and does not contain any new studies with
the database [13]. human participants or animals performed by
any of the authors. The authors declare that
there is no conflict of interest regarding the
publication of this paper.
458 Ophthalmol Ther (2021) 10:445–464

Data Availability. The datasets generated 6. Rajalakshmi R, Subashini R, Anjana RM, Mohan V.
and/or analyzed during the current study are Automated diabetic retinopathy detection in
smartphone-based fundus photography using arti-
available from the corresponding author on ficial intelligence. Eye. 2018;32:1138.
reasonable request.
7. Gulshan V, Peng L, Coram M, Stumpe MC, Wu D,
Open Access. This article is licensed under a Narayanaswamy A, et al. Development and valida-
tion of a deep learning algorithm for detection of
Creative Commons Attribution-Non-
diabetic retinopathy in retinal fundus photographs.
Commercial 4.0 International License, which JAMA. 2016;316:2402–10.
permits any non-commercial use, sharing,
adaptation, distribution and reproduction in 8. Sreelatha OK, Ramesh SV. Teleophthalmology:
improving patient outcomes? Clin Ophthalmol.
any medium or format, as long as you give
2016;10:285–95.
appropriate credit to the original author(s) and
the source, provide a link to the Creative 9. Verdager J, Martı́nez F, Barrı́a, F. Actualización de la
Commons licence, and indicate if changes were Guı́a Clı́nica de Retinopatı́a Diabética. International
Council of Ophthalmology 2016. http://www.
made. The images or other third party material
icoph.org/dynamic/attachments/resources/
in this article are included in the article’s guiaclinicaretinopatiadiabetica2016.pdf.
Creative Commons licence, unless indicated
otherwise in a credit line to the material. If 10. Zimmer-Galler IE, Kimura AE, Gupta S. Diabetic
retinopathy screening and the use of telemedicine.
material is not included in the article’s Creative
Curr Opin Ophthalmol. 2015;26(3):167–72 (10
Commons licence and your intended use is not Ministry of Health Chile. Community health
permitted by statutory regulation or exceeds the records with a focus on social determinants, Bio-
permitted use, you will need to obtain permis- Bio Region, 2015).
sion directly from the copyright holder. To view
11. Pieczynski J, Grzybowski A. Review of diabetic
a copy of this licence, visit http:// retinopathy screening methods and programmes
creativecommons.org/licenses/by-nc/4.0/. adopted in different parts of the world. European
Ophthalmic Review. 2015;9:49–55.

12. Malerbi FK, Carneiro ABM, Katz M, et al. Retinal

exams requested at primary care unit: indications,
REFERENCES results and alternative strategies of evaluation.
Einstein (Sao Paulo). 2019;18:913. https://doi.org/
1. Ogustova K, da Rocha FJ, Huang Y, Linnenkamp U, 10.31744/einstein_journal/2020GS4913 (eCollec-
Guariguata L, Cho NH, et al. IDF Diabetes Atlas: tion 2020).
global estimates for the prevalence of diabetes for
2015 and 2040. Diabetes Res Clin Pract. 2017;128: 13. Avendaño-Veloso A, Parada-Hernández F, González-
40. Ramos R, Dougnac-Osses C, Carrasco-Sáez C, Scan-
lon PH. Teleophthalmology: a strategy for timely
2. Lee R, Wong TY, Sabanayagam C. Epidemiology of diagnosis of sight-threatening diabetic retinopathy
diabetic retinopathy, diabetic macular edema and in primary care, Concepción. Chile Int J Ophthal-
related vision loss. Eye Vis (Lond). 2015;2:17. mol. 2019;12(9):1474–8.

3. Cheung N, Mitchell P, Wong TY. Diabetic 14. Bhaskaranand M, Ramachandra C, Bhat S, Cuadros
retinopathy. Lancet. 2010;376(9735):124–36. J, Nittala MG, Sadda SR, Solanki K. The value of
automated diabetic retinopathy screening with the
4. Wilson J, Jungner G. The principles and practice of EyeArt system: a study of more than 100,000 con-
screening for disease, Public Health Papers. Geneva: secutive encounters from people with diabetes.
WHO; 1968. Diabetes Technol Ther. 2019;21(11):635–43.
https://doi.org/10.1089/dia.2019.0164 (Epub 2019
5. Tufail A, Kapetanakis VV, Salas-Vega S, Egan C, Aug 7).
Rudisill C, Owen CG, et al. An observational study
to assess if automated diabetic retinopathy image 15. Mamillapalli CK, Prentice JR, Garg AK, Hampsey SL,
assessment software can replace one or more steps Bhandari R. Implementation and challenges unique
of manual imaging grading and to determine their to teleretinal diabetic retinal screening (TDRS) in a
costs effectiveness. Health Technol Assess (Rockv). private practice setting in the United States. J Clin
2016;20:1–72 (Xxviii). Transl Endocrinol. 2019;19:100214. https://doi.org/
10.1016/j.jcte.2019.100214.
Ophthalmol Ther (2021) 10:445–464 459

16. Ramchandran RS, Yilmaz S, Greaux E, et al. Patient J Digit Imaging. 2018;31(6):869–78. https://doi.org/
perceived value of teleophthalmology in an urban, 10.1007/s10278-018-0084-9 (PMID: 29704086).
low income us population with diabetes. PLoS ONE.
2020;15(1):e0225300. https://doi.org/10.1371/ 25. Crossland L, Askew D, Ware R, Cranstoun P,
journal.pone.0225300 (eCollection 2020). Mitchell P, Bryett A, Jackson C. Diabetic retinopa-
thy screening and monitoring of early stage disease
17. Joseph S, Kim R, Ravindran RD, Fletcher AE, Ravilla in Australian general practice: tackling pre-
TD. Effectiveness of teleretinal imaging-based hos- ventable blindness within a chronic care model.
pital referral compared with universal referral in J Diabetes Res. 2016;16:8405395. https://doi.org/10.
identifying diabetic retinopathy: a cluster random- 1155/2016/8405395 (Published online 2015).
ized clinical trial. JAMA Ophthalmol. 2019;137(7):
786–92. 26. Serrano CI, Shah V, Abràmoff MD, et al. Use of
expectation disconfirmation theory to test patient
18. Liu Y, Rajamanickam VP, Parikh RS, Loomis SJ, satisfaction with asynchronous telemedicine for
Kloek CE, Kim LA, Hitchmoth DL, Song BJ, Xerras diabetic retinopathy detection. Int J Telemed Appl.
DC, Pasquale LR. Diabetic retinopathy assessment 2018. https://doi.org/10.1155/2018/7015272
variability among eye care providers in an urban (eCollection 2018).
teleophthalmology program. Telemed J E Health.
2019;25(4):301–8. https://doi.org/10.1089/tmj. 27. Wong IYH, Ni MY, Wong IOL, et al. Saving sight in
2018.0019 (Epub 2018 Jul 24). China and beyond: the lifeline express model. BMJ
Glob Health. 2018;3(4):e000766. https://doi.org/10.
19. Modjtahedi BS, Theophanous C, Chiu S, Luong TQ, 1136/bmjgh-2018-000766 (eCollection 2018).
Nguyen N, Fong DS. Two-year incidence of retinal
intervention in patients with minimal or no dia- 28. Dóra JE, Dóra JS, Greg R, Kirby P, Edit P, László N,
betic retinopathy on telemedicine screening. JAMA Facskó A, Morten CM, Petrovski BE. Diabetic
Ophthalmol. 2019;137(4):445–8 (Published online retinopathy screening using telemedicine tools:
2019 Feb 7). pilot study in Hungary. Res Art Open Access.
2016;4529824:9. https://doi.org/10.1155/2016/
20. Cunha LP, Figueiredo EA, Araújo HP, et al. Mydri- 4529824.
atic fundus retinography in screening for dia-
betic retinopathy: agreement between family 29. Sheba MG, Erin Moran H, Fish A, Daskivich LP,
physicians, general ophthalmologists, and a retinal Ogunyemi OI. Understanding the Knowledge Gap
specialist. Front Endocrinol (Lausanne). 2018;9:251. Experienced by U.S. Safety Net Patients in Telereti-
https://doi.org/10.3389/fendo.2018.00251. nal Screening. In: AMIA Annu Symp Proc.
2017;2016:590–9.
21. Safi S, Ahmadieh H, Katibeh M, Yaseri M, Nikkhah
H, Karimi S, Nourinia R, Tivay A, Zareinejad M, 30. Daskivich LP, Vasquez C, Martinez C Jr, et al.
Azarmina M, Ramezani A, Moradian S, et al. Implementation and evaluation of a large-scale
Modeling a telemedicine screening program for teleretinal diabetic retinopathy screening program
diabetic retinopathy in Iran and implementing a in the Los Angeles County Department of Health
pilot project in Tehran Suburb. Res Article Open Services. JAMA Intern Med. 2017;177(5):642–9.
Access. 2019. https://doi.org/10.1155/2019/ https://doi.org/10.1001/jamainternmed.2017.0204.
2073679.
31. Litvin TV, Weissenberg CR, Daskivich LP, Zhou Q,
22. Kanagasingam Y, Xiao D, Vignarajan J, et al. Eval- Bresnick GH, Cuadros JA. Improving accuracy of
uation of artificial intelligence-based grading of grading and referral of diabetic macular edema
diabetic retinopathy in primary care. JAMA Netw using location and extent of hard exudates in reti-
Open. 2018;1(5):e182665. https://doi.org/10.1001/ nal photography. J Diabetes Sci Technol.
jamanetworkopen.2018.2665. 2016;10(2):262–70.

23. Yaslam M, Al Adel F, Al-Rubeaan K, AlSalem RK, 32. Davila JR, Sengupta SS, Niziol LM, et al. Predictors
Alageel MA, Alsalhi A, AlNageeb D, Youssef AM. of photographic quality with a handheld nonmy-
Non-mydriatic fundus camera screening with diag- driatic fundus camera used for screening of vision-
nosis by telemedicine for diabetic retinopathy threatening diabetic retinopathy. Ophthalmolog-
patients with type 1 and type 2 diabetes: a hospital- ica. 2017;238(1–2):89–99. https://doi.org/10.1159/
based cross-sectional study. Ann Saudi Med. 000475773 (Epub 2017 Jul 5).
2019;39(5):328–36. https://doi.org/10.5144/0256-
4947.2019.328 (Epub 2019 Oct 3). 33. Hatef E, Alexander M, Vanderver BG, et al. Assess-
ment of annual diabetic eye examination using
24. Saha SK, Fernando B, Cuadros J, et al. Automated telemedicine technology among underserved
Quality assessment of colour fundus images for patients in primary care setting. Middle East Afr J
diabetic retinopathy screening in telemedicine.
460 Ophthalmol Ther (2021) 10:445–464

Ophthalmol. 2017;24(4):207–12. https://doi.org/ 42. Seoud L, Hurtut T, Chelbi J, Cheriet F, Langlois JMP.
10.4103/meajo.MEAJO_19_16. Red Lesion detection using dynamic shape features
for diabetic retinopathy screening. IEEE Trans Med
34. Invernizzi A, Bevilacqua MT, Cozzi M, Bianchi C, Imaging. 2016;35(4):1116–26. https://doi.org/10.
Pagani A, Cigada M, Staurenghi G. Diabetic 1109/TMI.2015.2509785 (Epub 2015 Dec 17).
retinopathy screening: the first telemedical
approach in an Italian Hospital. Observational 43. Silva PS, Cavallerano J, Haddad NMN, Tolls D,
Study. Eur J Ophthalmol. 2016;26(4):369–74. Thakore K, Patel B, Sehizadeh M, Tolson AM, Sun
https://doi.org/10.5301/ejo.5000719 (Epub 2015 JK, Aiello LP. Comparison of Nondiabetic Retinal
Dec 14). Findings Identified With Nonmydriatic Fundus
Photography vs Ultrawide Field Imaging in an
35. Jani PD, Forbes L, Choudhury A, et al. Evaluation of Ocular Telehealth Program. JAMA Ophthalmol.
diabetic retinal screening and factors for ophthal- 2016;134(3):330–4. https://doi.org/10.1001/
mology referral in a telemedicine network. JAMA jamaophthalmol.2015.5605.
Ophthalmol. 2017;135(7):706–14. https://doi.org/
10.1001/jamaophthalmol.2017.1150. 44. Fonda SJ, Bursell SE, Lewis DG, et al. The In-
dian Health Service primary care-based teleoph-
36. Maa AY, Patel S, Chasan JE, Delaune W, Lynch thalmology program for diabetic eye disease
MGG. Retrospective evaluation of a teleretinal surveillance and management. Telemed J E Health.
screening program in detecting multiple nondia- 2020. https://doi.org/10.1089/tmj.2019.0281 (On-
betic eye diseases. Telemed e-Health. 2017. https:// line ahead of print).
doi.org/10.1089/tmj.2016.0039 (Original
Research. Published Online:1 Jan 2017). 45. Gu D, Agron S, May LN, et al. Nonmydriatic reti-
nal diabetic screening in the primary care setting:
37. Safdari R, Langarizadeh M, Ramezani A, et al. assessing degree of retinopathy and incidence of
Development of a store-and-forward telescreening nondiabetic ocular diagnoses. Telemed J E Health.
system of diabetic retinopathy: lessons learned from 2020. https://doi.org/10.1089/tmj.2019 (Online
Iran. J Diabetes Metab Disord. 2018;17(1):31–6. ahead of print).
https://doi.org/10.1007/s40200-018-0335-9 (eCol-
lection 2018 Jun). 46. Toy BC, Aguinaldo T, Eliason J, Egbert J. Non-my-
driatic fundus camera screening for referral-war-
38. Murchison AP, Friedman DS, Gower EW, Haller JA, ranted diabetic retinopathy in a northern California
Lam BL, Lee DJ, McGwin G Jr, Owsley C, Saaddine J. safety-net setting. Ophthalmic Surg Lasers Imaging
Insight study group a multi-center diabetes eye Retina. 2016;47(7):636–42. https://doi.org/10.3928/
screening study in community settings: study 23258160-20160707-05.
design and methodology. Ophthalmic Epidemiol.
2016;23(2):109–15. https://doi.org/10.3109/ 47. Martin VY, Callirgos PGE, Martı́n MTM, et al. Sat-
09286586.2015.1099682 (Epub 2016 Mar 7). isfaction of patients and primary care profession-
als with a teleophthalmology-based screening
39. Aiello LP, Cavallerano J, Sun J, et al. Long-term ef- programme for diabetic retinopathy in a rural area
fect on HbA1c in poorly controlled diabetic pa- in Castilla y León, Spain. Rural Remote Health.
tients following nonmydriatic retinal image review 2020;20(1):5180. https://doi.org/10.22605/
at the time of endocrinology visit. Telemed J E RRH5180 (Epub 2020 Jan 16).
Health. 2020. https://doi.org/10.1089/tmj.2019.
0239. 48. Tsui I, Havunjian MA, Davis JA, Giaconi JA. Snap-
shot of teleretinal screening for diabetic retinopathy
40. Rico-Sergado L, Pérez-Canales JL, Pérez-Santonja JJ. at the West Los Angeles Medical Center. Telemed J E
Effect of visual impairment on teleretinal imaging Health. 2016;22(10):843–6. https://doi.org/10.
for diabetic retinopathy screening. Ophthalmic 1089/tmj.2015.0246 (Epub 2016 Mar 17).
Surg Lasers Imaging Retina. 2016;47(1):42–8.
https://doi.org/10.3928/23258160-20151214-06. 49. Mastropasqua L, Perilli R, D’Aloisio R, et al. Why
miss the chance? incidental findings while tele-
41. Boucher MC, Qian J, Brent MH, et al. Steering screening for diabetic retinopathy. Ohthalmic Epi-
Committee for Tele-Ophthalmology Screening, demiol. 2020. https://doi.org/10.1080/09286586.
Canadian Retina Research Network. Evidence- 2020.1715450 (Online ahead of print).
based Canadian guidelines for tele-retina screening
for diabetic retinopathy: recommendations from 50. Turnin M-C, Schirr-Bonnans S, Chauchard M-C,
the Canadian Retina Research Network Deglise P, Journot C, Lapeyre Y, Güell A, Lepage B,
(CR2N) Tele-Retina Steering Committee. Can J Hanaire H, Martini J. DIABSAT telemedicine itiner-
Ophthalmol. 2020;55(1S1):14–24. https://doi.org/ ant screening of chronic complications of diabetes
10.1016/j.jcjo.2020.01.001. using a satellite. Telemed J E Health. 2017;23(5):
Ophthalmol Ther (2021) 10:445–464 461

397–403. https://doi.org/10.1089/tmj.2016.0185 1622–7. https://doi.org/10.1136/bjophthalmol-

(Epub 2016 Dec 21). 2014-306536 (Epub 2015 May 20).

51. Strul S, Zheng Y, Gangaputra S, et al. Pediatric dia- 60. Walton OB 4th, Garoon RB, Weng CY, et al. Eval-
betic retinopathy telescreening. AAPOS. uation of automated teleretinal screening program
2020;24(1):10.e1-10.e5. https://doi.org/10.1016/j. for diabetic retinopathy. JAMA Ophthalmol.
jaapos.2019.10.010 (Epub 2020 Jan 12). 2016;134(2):204–9. https://doi.org/10.1001/
jamaophthalmol.2015.5083.
52. Vujosevic S, Pucci P, Casciano M, Daniele AR, Bini S,
Berton M, Cavarzeran F, Avogaro A, Lapolla A, 61. Bawankar P, Shanbhag N, Dhawan B, Palsule A,
Midena E. A decade-long telemedicine screening Kumar D, Chandel S, et al. Sensitivity and speci-
program for diabetic retinopathy in the North-East ficity of automated analysis of single-field non-
of Italy. J Diabetes Complications. 2017;31(8): mydriatic fundus photographs by Bosch DR Algo-
1348–53. https://doi.org/10.1016/j.jdiacomp.2017. rithm—Comparison with mydriatic fundus pho-
04.010 (Epub 2017 Apr 13). tography (ETDRS) for screening in undiagnosed
diabetic retinopathy. PLoS ONE. 2017;12:e0189854.
53. Zapata MA, Arcos G, Fonollosa A, Abraldes M, Ole-
ñik A, Gutierrez E, Garcia-Arumi J. Telemedicine for 62. Kanjee R, Dookeran RI, Mathen MK, Stockl FA,
a general screening of retinal disease using non- Leicht R. Six-year prevalence and incidence of dia-
mydriatic fundus cameras in optometry centers: betic retinopathy and cost-effectiveness of tele-
three-year results. Telemed J E Health. 2017;23(1): ophthalmology in Manitoba. Can J Ophthalmol.
30–6. https://doi.org/10.1089/tmj.2016.0020 (Epub 2017;52(Suppl 1):S15–8. https://doi.org/10.1016/j.
2016 May 26). jcjo.2017.09.022.

54. Clément M, et al. Evaluation of diabetic retinopa- 63. Horton MB, Brady CJ, Cavallerano J, et al. Practice
thy screening using fundus photography: evalua- guidelines for ocular telehealth-diabetic retinopa-
tion and epidemiologic factors at Nantes university thy, third edition. Telemed J E Health. 2020;26(4):
medical center. J Fr Ophtalmol. 2019;42(3):281–7. 495–543. https://doi.org/10.1089/tmj.2020.0006
(Epub 2020 Mar 25).
55. Porta M, Boscia F, Lanzetta P, Mannucci E, Men-
chini U, Simonelli F. Systematic screening of 64. Park DW, Mansberger SL. Eye disease in patients
retinopathy in diabetes (REaD Project): an Italian with diabetes screened with telemedicine. Telemed
Implementation Campaign. Eur J Ophthalmol. J E Health. 2017;23(2):113–8. https://doi.org/10.
2017;27(2):179–84. https://doi.org/10.5301/ejo. 1089/tmj.2016.0034 (Epub 2016 Jun 21).
5000912 (Epub 2016 Nov 21).
65. Silva PS, Horton M, Clary D, Lewis DG, Sun JK,
56. Date RC, Shen KL, Shah BM, et al. Accuracy of Cavallerano JD, Aiello LP. Identification of diabetic
detection and grading of diabetic retinopathy and retinopathy and ungradable image rate with ultra-
diabetic macular edema using teleretinal screening. wide field imaging in a national teleophthalmol-
Ophthalmol Retina. 2019;3(4):343–9. https://doi. ogy. Program Ophthalmol. 2016;123(6):1360–7.
org/10.1016/j.oret.2018.12.003 (Epub 2018 Dec https://doi.org/10.1016/j.ophtha.2016.01.043
24). (Epub 2016 Mar 2).

57. Pedrosa M, Silva JM, Silva JF, et al. SCREEN-DR: 66. Bursell SE, Fonda SJ, Lewis DG, et al. Prevalence of
Collaborative platform for diabetic retinopathy. Int diabetic retinopathy and diabetic macular edema in
J Med Inf. 2018;120:137–46. https://doi.org/10. a primary care-based teleophthalmology program
1016/j.ijmedinf.2018.10.005 (Epub 2018 Oct 18). for American Indians and Alaskan Natives. PLoS
ONE. 2018;13(6):e0198551. https://doi.org/10.
58. Sanborn GE, Wroblewski JJ. Evaluation of a combi- 1371/journal.pone.0198551 (eCollection 2018).
nation digital retinal camera with spectral-domain
optical coherence tomography (SD-OCT) that 67. Mansberger SL, Sheppler C, Barker G, et al. Long-
might be used for the screening of diabetic term comparative effectiveness of telemedicine in
retinopathy with telemedicine: A pilot study. J Dia- providing diabetic retinopathy screening examina-
betes Complic. 2018;32(11):1046–50. https://doi. tions: a randomized clinical trial. JAMA Ophthal-
org/10.1016/j.jdiacomp.2018.08.010 (Epub 2018 mol. 2015;133(5):518–25. https://doi.org/10.1001/
Aug 11). jamaophthalmol.2015.1.

59. Conlin PR, Asefzadeh B, Pasquale LR, et al. Accuracy 68. Ramakrishnan R, Khadar AMA, Srinivasan K, et al.
of a technology-assisted eye exam in evaluation of Diabetes mellitus in the Tamil Nadu State-non-
referable diabetic retinopathy and concomitant communicable diseases nurse model in diabetic
ocular diseases. Br J Ophthalmol. 2015;99(12): retinopathy screening. Indian J Ophthalmol.
462 Ophthalmol Ther (2021) 10:445–464

2020;68(Suppl 1):S78–82. https://doi.org/10.4103/ 78. Taraprasad Das and Rajeev Reddy Pappuru. Tele-
ijo.IJO_1987_19. medicine in diabetic retinopathy: access to rural
India. Indian J Ophthalmol. 2016;64(1):84–6.
69. Afshar AR, Oldenburg CE, Stewart JM. A novel
hybrid fixed and mobile ultra-widefield imaging 79. Matimba A, Woodward R, Tambo E, Ramsay M,
program for diabetic teleretinopathy screening. Gwanzura L, Guramatunhu S. Tele-ophthalmology:
Ophthalmol Retina. 2019;3(7):576–9. https://doi. opportunities for improving diabetes eye care in
org/10.1016/j.oret.2019.03.007 (Epub 2019 Mar resource- and specialist-limited Sub-Saharan African
20). Countries. J Telemed Telecare. 2016;22(5):311–6.
https://doi.org/10.1177/1357633X15604083 (Epub
70. Kern C, Kortuem K, Hamilton R, et al. Clinical 2015 Sep 24).
outcomes of a hospital-based teleophthalmology
service: what happens to patients in a virtual clinic? 80. Toy BC, Myung DJ, He L, Pan CK, Chang RT,
Ophthalmol Retina. 2019;3(5):422–8. https://doi. Polkinhorne A, Merrell D, Foster D, Blumenkranz
org/10.1016/j.oret.2019.01.011 (Epub 2019 Jan MS. Smartphone- based dilated fundus Photography
21). and near visual acuity testing as inexpensive
screening tools detect referral warranted diabetic
71. Hussain N, Edraki M, Tahhan R, et al. Telemedicine eye disease. Retina. 2016;36(5):1000–8. https://doi.
for diabetic retinopathy screening using an ultra- org/10.1097/IAE.0000000000000955.
widefield fundus camera. Clin Ophthalmol.
2017;14(11):1477–82. https://doi.org/10.2147/ 81. Martinez JA, Parikh PD, Wong RW, et al. Tele-
OPTH.S135287 (eCollection 2017). medicine for diabetic retinopathy screening in an
urban, insured population using fundus cameras in
72. Bartnik SE, Copeland SP, Aicken AJ, et al. Optome- a primary care office setting. Ophthalmic Surg
try-facilitated teleophthalmology: an audit of the Lasers Imaging Retina. 2019;50(11):e274–7. https://
first year in Western Australia. Clin Exp Optom. doi.org/10.3928/23258160-20191031-14.
2018;101(5):700–3. https://doi.org/10.1111/cxo.
12658 (Epub 2018 Feb 14). 82. Keshvardoost S, Bahaadinibeigy K, Shadman H,
et al. Design, development, and evaluation of a
73. Alam M, Le D, Lim JI, Chan RVP, Yao X. Supervised teleophthalmology system using a low-cost fundus
machine learning based multi-task artificial intelli- camera. Acta Inform Med. 2020;28(1):12–7. https://
gence classification of retinopathies. J Clin Med. doi.org/10.5455/aim.2019.28.12-17.
2019;8(6):872. https://doi.org/10.3390/jcm8060872
83. Zhang W, Nicholas P, Schuman SG, et al. Screening
74. Grau E, Horn F, Nixdorff U, Michelson G. OCT and for diabetic retinopathy using a portable, noncon-
IOP Findings in a healthy worker cohort: results tact, nonmydriatic handheld retinal camera. J Dia-
from a teleophthalmic study in occupational med- betes Sci Technol. 2017;11(1):128–34. https://doi.
icine. Graefes Arch Clin Exp Ophthalmol. org/10.1177/1932296816658902 (Epub 2016 Jul
2019;257(11):2571–8. https://doi.org/10.1007/ 11).
s00417-019-04457-1 (Epub 2019 Sep 5).
84. Bastawrous A. Smartphone fundoscopy. Ophthal-
75. Kortuem K, Fasler K, Charnley A, et al. Implemen- mology. 2012;119(2):432–3.
tation of medical retina virtual clinics in a tertiary
eye care referral centre. Br J Ophthalmol. 85. Micheletti M, Hendrick AM, Khan FN, Ziemer D,
2018;102(10):1391–5. https://doi.org/10.1136/ Pasquel FJ. Current and next generation
bjophthalmol-2017-311494 (Epub 2018 Jan 6). portable screening devices for diabetic retinopathy.
J Diabetes Sci Technol. 2016;10(2):295–300. https://
76. Wang YT, Tadarati M, Wolfson Y, et al. Comparison doi.org/10.1177/1932296816629158 (Published
of prevalence of diabetic macular edema based on online 2016 Feb 16).
monocular fundus photography vs optical coher-
ence tomography. JAMA Ophthalmol. 2016;134(2): 86. Choon HT, Bhone MK, Helen S, Colin ST, Lorainne
222–8. https://doi.org/10.1001/jamaophthalmol. TC. Use of smartphones to detect diabetic
2015.5332. retinopathy: scoping review and meta-analysis of
diagnostic test accuracy studies. J Med Internet Res.
77. Gadkari S, MRCOphth FRCS. Innovative model for 2020;22(5):e16658. https://doi.org/10.2196/16658
telemedicine-based screening for diabetic retinopa- (Published online 2020 May 15).
thy in the developing world. Can J Opthamol.
2016;51(3):e109–11. https://doi.org/10.1016/j.jcjo. 87. Kim TN, Meyers F, Reber C, et al. A smartphone-
2015.11.010 (50TH ANNIVERSARY ISSUE). based tool for rapid, portable, and automated wide-
field retinal imaging. Transl Vis Sci Technol.
2018;7(5):21.
Ophthalmol Ther (2021) 10:445–464 463

88. Bhat S, Bhaskaranand M, Ramachandra C, Margolis 97. Saeed E, Szymkowski M, Saeed K, et al. An approach
TP, Fletcher DA, Solanki K. Fully-automated dia- to automatic hard exudate detection in retina color
betic retinopathy screening using cellphone-based images by a telemedicine system based on the d-eye
cameras. Invest Ophthalmol Vis Sci. 2015;56(7): sensor and image processing algorithms. Sensors
1428–1428. (Basel). 2019;19(3):695. https://doi.org/10.3390/
s19030695.
89. Bolster NM, Giardini ME, Bastawrous A. The dia-
betic retinopathy screening workflow potential for 98. Kozak I, Payne JF, Schatz P, Al-Kahtani E, Winkler
smartphone imaging. J Diabetes Sci Technol. M. Teleophthalmology image-based navigated reti-
2016;10(2):318–24. https://doi.org/10.1177/ nal laser therapy for diabetic macular edema: a
1932296815617969 (Published online 2015 Nov concept of retinal telephotocoagulation. Graefes
23). Arch Clin Exp Ophthalmol. 2017;255(8):1509–13.
https://doi.org/10.1007/s00417-017-3674-1 (Epub
90. Sean C, David C, Geoffrey T, Karen H, David M, 2017 Apr 26).
Suman T. Utility and feasibility of teleophthalmol-
ogy using a smartphone-based ophthalmic camera 99. Nguyen H, Tan GSW, Tapp RJ, Mital S, Ting SWD,
in screening camps in Nepal. Asia Pac J Ophthalmol Wong HT, Tan C, Laude A, Tai ES, Chuan Tan N,
(Phila). 2020;9(1):54–8. https://doi.org/10.1097/01. Finkelstein E, Wong TY, Lamoureux EL. 10 cost-ef-
APO.0000617936.16124 (Published online 2020 fectiveness of a national telemedicine diabetic
Feb 3). retinopathy screening program in Singapore. Oph-
thalmology. 2016;123(12):2571–80. https://doi.org/
91. Bastawrous A, Giardini ME, Bolster NM, Peto T, 10.1016/j.ophtha.2016.08.021 (Epub 2016 Oct 7).
Shah N, Livingstone I, Weiss HA, Hu S, Rono H,
Kuper H, Burton M. Clinical validation of smart- 100. Abràmoff MD, Leng T, Ting DSW, Rhee K, Horton
phone based adapter: peek retina for optic disc MB, Brady CJ, Chiang MF. Automated and com-
imaging in Kenya. JAMA Ophthalmol. 2016;134(2): puter-assisted detection, classification, and diagno-
151–8. https://doi.org/10.1001/jamaophthalmol. sis of diabetic retinopathy. Telemed e-Health.
2015.4625 (Author manuscript; available in 2020;26(4):544–50. https://doi.org/10.1089/tmj.
PMC 2017 Feb 22. Published in final edited form 2020.0008.
as: JAMA Ophthalmol).
101. Rodrı́guez Villa S, Suárez Muñiz MT, Valle DDDR,
92. Patel TP, Kim NT, Yu G, Dedania VS, Lieu P, Qian et al. Enferm Retinopathy diabetic screening by
CX, Besirli CG, Demirci H, Margolis T, Fletcher DA, non-mydriatic retinography: concordance between
Paulus YM. Smartphone-based, rapid, wide-field primary care physicians, nurses and ophthalmolo-
fundus photography for diagnosis of pediatric reti- gists. Clin. 2018;28(1):44–8. https://doi.org/10.
nal diseases. Transl Vis Sci Technol. 2019;8(3):29. 1016/j.enfcli.2017.04.010 (Epub 2017 Jun 16).
https://doi.org/10.1167/tvst.8.3.29.
102. Boucher MC, Nguyen MTD, Qian J. Assessment of
93. Lekha T, Ramesh S, Sharma A, Abinaya G. MII training outcomes of nurse readers for diabetic
RetCam assisted smartphone based fundus imaging retinopathy telescreening: validation study. JMIR
for retinopathy of prematurity. Indian J Ophthal- Diabetes. 2020;5(2):e17309. https://doi.org/10.
mol. 2019;67(6):834–9. https://doi.org/10.4103/ijo. 2196/17309.
IJO_268_19.
103. Abràmoff MD, Lou Y, Erginay A, Clarida W, Amelon
94. Wintergerst MW, Jansen LG, Holz F, Finger RP. a R, Folk JC, et al. Improved automated detection of
novel device for smartphone-based fundus imaging diabetic retinopathy on a publicly available dataset
and documentation in clinical practice: compara- through integration of deep learning. Invest Oph-
tive image analysis study. JMIR Mhealth Uhealth. thalmol Vis Sci. 2016;57:5200–6.
2020;8(7):e17480. https://doi.org/10.2196/17480
(Published online 2020 Jul 29). 104. Abràmoff MD, Lavin PT, Birch M, Shah N, Folk JC.
Pivotal trial of an autonomous AI-based diagnostic
95. Khanamiri NH, Nakatsuka A, El-Annan J. Smart- system for detection of diabetic retinopathy in pri-
phone fundus photography. J Vis Exp. 2017;125: mary care offices. npj Digit Med. 2018;1:39.
55958. https://doi.org/10.3791/55958 (Published
online 2017 Jul 6). 105. Abràmoff MD, Folk JC, Han DP, Walker JD, Wil-
liams DF, Russell SR, et al. Automated analysis of
96. Kim Y, Chao DL. Comparison of smartphone oph- retinal images for detection of referable diabetic
thalmoscopy vs conventional direct ophthal- retinopathy. JAMA Ophthalmol. 2013;131:351–7.
moscopy as a teaching tool for medical students:
the COSMOS study. Clin Ophthalmol. 2019;13: 106. US Food and Drug Administration. FDA permits
391–401. https://doi.org/10.2147/OPTH.S190922 marketing of artificial intelligence-based device to
(Published online 2019 Feb 18). detect certain diabetes-related eye problems. News
464 Ophthalmol Ther (2021) 10:445–464

Release, US Food and Drug Administration; April 117. Zhu X, Xu Y, Lu L, Zou H, et al. Patients’ perspec-
2018. tives on the barriers to referral after telescreening
for diabetic retinopathy in communities. BMJ Open
107. Ribeiro L, Oliveira CM, Neves C, Ramos JD, Ferreira Diabetes Res Care. 2020;8(1):e000970. https://doi.
H, Cunha-Vaz J. Screening for diabetic retinopathy org/10.1136/bmjdrc-2019-000970.
in the central region of Portugal. Added value of
automated ‘disease/no disease’ grading. Ophthal- 118. Liu, Zupan NJ, Swearingen R, Jacobson N, Carlson
mologica. 2015;233:96–103. JN, Mahoney JE, Klein R, Bjelland TD, Smith MA.
Identification of barriers, facilitators and system-
108. Ribeiro ML, Nunes SG, Cunha-Vaz JG. Microa- based implementation strategies to increase
neurysm turnover at the macula predicts risk of teleophthalmology use for diabetic eye screening in
development of clinically significant macular a rural US primary care clinic: a qualitative study.
edema in persons with mild nonproliferative dia- http://orcid.org/0000-0002-0700-0148Yao.
betic retinopathy. Diabetes Care. 2012;36:1254–9.
119. Gibson MD. Estimates of the percentage of US
109. Ting DSW, Cheung CY-L, Lim G, Tan GSW, Quang adults with diabetes who could be screened for
ND, Gan A, et al. Development and validation of a diabetic retinopathy in primary care settings. JAMA
deep learning system for diabetic retinopathy and Ophthalmol. 2019;137(4):440–4. https://doi.org/
related eye diseases using retinal images from mul- 10.1001/jamaophthalmol.2018.6909 (Published
tiethnic populations with diabetes. JAMA. online 2019 Jan 31).
2017;318:2211–23.
120. Valikodath NG, Leveque TK, Wang SY, et al. Patient
110. Larsen N, Godt J, Grunkin M, Lund-Andersen H, attitudes toward telemedicine for diabetic
Larsen M. Automated detection of diabetic retinopathy. Telemed J E Health. 2017;23(3):
retinopathy in a fundus photographic screening 205–12. https://doi.org/10.1089/tmj.2016.0108
population. Invest Ophthalmol Vis Sci. 2003;44: (Epub 2016 Jun 23).
767–71.
121. Teo ZL, Tham YC, Yu M, et al. Do we have enough
111. Hansen AB, Hartvig NV, Jensen MS, Borch-Johnsen ophthalmologists to manage vision-threatening
K, Lund- Andersen H, Larsen M. Diabetic retinopa- diabetic retinopathy? A global perspective. Eye
thy screening using digital non-mydriatic fundus (Lond). 2020;34(7):1255–61. https://doi.org/10.
photography and automated image analysis. Acta 1038/s41433-020-0776-5 (Epub 2020 Jan 28).
Ophthalmol Scand. 2004;82:666–72.
122. Mehlsen J, Erlandsen M, Poulsen PL, Bek T. Indi-
112. Larsen M, Godt J, Larsen N, Lund-Andersen H, Sjølie vidualized optimization of the screening interval
AK, Agardh E, et al. Automated detection of fundus for diabetic retinopathy: a new model. Acta Oph-
photographic red lesions in diabetic retinopathy. thalmol. 2012;90:109–14.
Invest Ophthalmol Vis Sci. 2003;44:761–6.
123. Kirkizlar E, Serban N, Sisson JA, Swann JL, Barnes
113. Gargeya R, Leng T. Automated identification of CS, Williams MD. Evaluation of telemedicine for
diabetic retinopathy using deep learning. Ophthal- screening of diabetic retinopathy in the veterans
mology. 2017;124:962–9. health administration. Ophthalmology.
2013;120(12):2604–10.
114. Li Z, Keel S, Liu C, He Y, Meng W, Scheetz J, et al. An
automated grading system for detection of vision- 124. Taylor CR, Merin LM, Salunga AM, Hepworth JT,
threatening referable diabetic retinopathy on the Crutcher TD, O’Day DM, Pilon BA. Improving dia-
basis of color fundus photographs. Diabetes Care. betic retinopathy screening ratios using tele-
2018;41:2509–16. medicine-based digital retinal imaging technology:
the Vine Hill study. Diabetes Care. 2007;30(3):
115. Stevenson CH, Hong SC, Ogbuehi KC. Develop- 574–8.
ment of an artificial intelligence system to classify
pathology and clinical features on retinal fundus 125. Cuadros J, Bresnick G. EyePACS: an adapt-
images. Clin Exp Ophthalmol. 2019;47(4):484–9. able telemedicine system fordiabetic retinopathy
https://doi.org/10.1111/ceo.13433 (Epub 2018 Nov screening. J Diabetes Sci Technol. 2009;3(3):509–16.
15).
126. Scanlon PH. The English National Screening Pro-
116. Das V, Dandapat S, Bora PK. A novel diagnostic gramme for diabetic retinopathy 2003–2016. Acta
information based framework for super-resolution Diabetol. 2017;54:515–25.
of retinal fundus images. Comput Med Imaging
Graph. 2019;72:22–33. https://doi.org/10.1016/j.
compmedimag.2019.01.002 (Epub 2019 Feb 1).