Diabetic Retinopathy Screening Using

Smartphone-Based Fundus Imaging in India

Maximilian W.M. Wintergerst, MD,1 Divyansh K. Mishra, DNB,2 Laura Hartmann,1 Payal Shah, DNB,2
Vinaya K. Konana, DNB,2 Pradeep Sagar, MD,2 Moritz Berger, PhD,3 Kaushik Murali, DNB,2
Frank G. Holz, MD,1 Mahesh P. Shanmugam, FRCSEd,2,* Robert P. Finger, PhD1,*

Purpose: Early detection and treatment can prevent irreversible blindness from diabetic retinopathy (DR),
which is the leading cause of visual impairment among working-aged adults worldwide. Some 80% of affected
persons live in low- and middle-income countries, yet lack of resources has largely prevented DR screening
implementation in these world regions. Smartphone-based fundus imaging (SBFI) allows for low-cost mobile
fundus examination using an adapter on a smartphone; however, key aspects such as image quality, diagnostic
accuracy, and comparability of different approaches have not been systematically assessed to date.
Design: Evaluation of diagnostic technology.
Participants: A total of 381 eyes of 193 patients with diabetes were recruited at outreach eye clinics in South
India.
Methods: We compared 4 technically different approaches of SBFI (3 approaches based on direct and 1
approach based on indirect ophthalmoscopy) in terms of image quality and diagnostic accuracy for DR screening.
Main Outcome Measures: Image quality (sharpness/focus, reflex artifacts, contrast, and illumination), field-
of-view, examination time, and diagnostic accuracy for DR screening were analyzed against conventional fundus
photography and clinical examination.
Results: Smartphone-based fundus imaging based on indirect ophthalmoscopy yielded the best image
quality (P < 0.01), the largest field-of-view, and the longest examination time (111 vs. 68e86 seconds, P <
0.0001). Agreement with the reference standard (Cohen’s kappa 0.868) and sensitivity/specificity to detect DR
were highest for the indirect SBFI approach (0.79/0.99 for any DR and 1.0/1.0 for severe DR, 0.79/1.0 for diabetic
maculopathy).
Conclusions: Smartphone-based fundus imaging can meet DR screening requirements in an outreach
setting; however, not all devices are suitable in terms of image quality and diagnostic accuracy. Smartphone-
based fundus imaging might aid in alleviating the burden of DR screening in low- and middle-income coun-
tries, and these results will allow for a better selection of SBFI devices in field trials for DR
screening. Ophthalmology 2020;127:1529-1538 ª 2020 by the American Academy of Ophthalmology

Supplemental material available at www.aaojournal.org.

Diabetic retinopathy (DR) is the leading cause of visual effectiveness of screening for DR.9 However many low- and
impairment among working-aged adults around the world middle-income countries do not have any DR screening
and is present in one-third of patients with diabetes.1-3 Its programs because of lack of resources.4,6,7,10-14
prevalence continues to increase alarmingly, most markedly Thus, the WHO and the International Diabetes Federa-
in middle-income countries,3-5 where cost of care and tion have prioritized low-cost screening technologies such
limited access to eye specialists are the major barriers to as digital retinal imaging by nonphysicians, remote grading,
sufficient eye care.6,7 With approximately 80% of all and mobile healthcare services.7,14,15 Tele-ophthalmologic
persons with diabetes and associated DR living in low- DR screening has proven to be a practical approach
and middle-income countries,3,7,8 the burden is felt most whereby rural areas, otherwise hard to reach, can be pro-
in the developing world, where many healthcare systems vided with DR screening.10,13,14,16-18 However, there is a
are ill equipped to properly identify and manage the growing need for novel, more cost-effective DR screening
disease.4,6,7 solutions because financial factors remain a barrier to
Against this background, DR has been defined as a pri- widespread implementation.13,19
ority eye disease by the World Health Organization (WHO). With the advent of smartphone-based fundus imaging
Blindness and visual impairment due to DR can be pre- (SBFI), an inexpensive option for mobile fundus examina-
vented in the majority of cases with early diagnosis and tion and documentation has become available, which has the
timely treatment. Several studies have reported on the cost- potential to revolutionize eye care in terms of availability

ª 2020 by the American Academy of Ophthalmology https://doi.org/10.1016/j.ophtha.2020.05.025 1529

Published by Elsevier Inc. ISSN 0161-6420/20
 Ophthalmology Volume 127, Number 11, November 2020

and cost-effectiveness.20-23 Furthermore, tele- SBFI device with an iPod Touch (6th generation, Apple Inc.,
ophthalmology would be a natural extension of SBFI as Cupertino, CA), as required by the Paxos Scope app. The indirect
the hardware for mobile connectivity comes with the device. SBFI device was equipped with a Pan Retinal 2.2 lens (Volk
Thus, SBFI could be a feasible low-cost alternative to Optical, Inc., Mentor, OH) for indirect ophthalmoscopy. Video
resolution for SBFI examination was 1280720 with 15 frames/
conventional digital retinal imaging in the context of tele-
seconds (Samsung Galaxy S4) and 19201080 with 30 frames/
ophthalmologic DR screening; however, image quality, second (iPod Touch) rendering a typical fundus field-of-view res-
comparability of devices, and diagnostic accuracy remain olution of approximately 400400 up to 600600 pixels
unclear to date. Some SBFI approaches have already been (dependent on the varying distance to the eye) for the direct SBFI
applied to DR screening, although with considerable devices and approximately 550550 pixels for the indirect SBFI
inconsistency in imaging protocols and resulting sensitiv- devices, respectively.
ities to detect DR ranging from 50% to >90%.24-29
Furthermore, most of the reported studies were Training
manufacturer-initiated studies, took place in tertiary hospital Six optometrists at the Sankara Eye Hospital Bangalore were
settings in highly developed countries, and have not intensively trained for 3 days in SBFI use with the different devices
compared the different devices available. on voluntary subjects. Training comprised acquisition of high-
We compared 4 different SBFI devices in terms of quality fundus examination videos, achieving full compliance
various image quality criteria, field-of-view, examination with the imaging protocol (see “Analysis Including Image Quality
time, and diagnostic accuracy to detect DR in outreach eye Scales”) and recognizing signs of DR and diabetic maculopathy.
clinics in South India.
Clinical Examination and Data Acquisition
Methods Eyes were dilated with 0.8% tropicamide and 5% phenylephrine
after a gross anterior segment evaluation to rule out a shallow
Setting and Participants anterior chamber using torchlight and slit-lamp, and imaged with
the SBFI devices and conventional 7-field color fundus photog-
Participants were recruited at 13 DR outreach eye clinics in and raphy. All SBFI imaging was performed by trained optometrists or
around Bangalore, India, between April and May 2017. Eye clinics ophthalmologists. Examiners were masked to the results of the
were organized in collaboration with local healthcare providers and clinical evaluation. Smartphone-based fundus imaging was per-
nongovernmental organizations, for example, HelpAge India. Pa- formed in video mode, following a preset imaging protocol starting
tients were offered free of charge refraction, visual acuity mea- at the optic nerve head, followed by imaging of the temporal
surement, plasma glucose measurement (via fingertip lancet inferior arcade, the macula, the temporal superior arcade, the nasal
puncture), and dilated fundus examination by an ophthalmologist, inferior arcade, and the nasal superior arcade (Fig S1, available at
and glasses were provided for 250 Indian rupee (w $3.60 USD) for www.aaojournal.org). All image acquisition was performed in
monofocal and 350 Indian rupee (w $5.10 USD) for bifocal darkened rooms. Clinical examination was done using
glasses if required. Ethical approval was obtained from the ethics torchlights, slit-lamp, and indirect binocular ophthalmoscopy.
committee at the Sankara Eye Hospital Bangalore, India (ethics Conventional 7-field color fundus photography and clinical ex-
committee registration number ECR/705/Inst/KA/2015) and the amination by indirect ophthalmoscopy were used as the reference
University of Bonn, Germany (ethics committee ID 010/18), and standard. Diabetic retinopathy was classified in accordance to the
informed consent was obtained from all study participants before International Clinical Diabetic Retinopathy Severity Scale.32 If
inclusion. The study adhered to the tenets of the Declaration of only retinal exudates were present, the eye was classified as DR
Helsinki. Inclusion criteria were an age of at least 18 years, known stage 1, and if macular exudates were present, the eye was
or suspected diabetes mellitus, or plasma glucose
200 mg/dl classified as diabetic maculopathy.
(
11.1 mmol/l).30 Exclusion criteria were known allergies to the
dilating eye drops or a shallow anterior chamber. Analysis Including Image Quality Scales
Devices The SBFI image material was analyzed for image quality,
compliance with the imaging protocol, examination time, field-of-
The following devices were used for imaging: view, and diagnostic accuracy to detect DR. To compare image
quality between the different devices, we developed semi-
 Peek Retina (referred to as “direct SBFI device 1,” Peek quantitative scales with exemplary images for the assessment of
Vision Ltd., London, UK, adapter version from 2017, Fig sharpness/focus, reflex artifacts, contrast, and illumination (Fig S2,
1A); available at www.aaojournal.org). To reduce bias due to contrast
 D-EYE (“direct SBFI device 2,” D-EYE S.r.l., Padova, Italy, and illumination, sharpness/focus was evaluated on the basis of
adapter version from 2016, Fig 1B);
vessels at the optic nerve head, which provided a well-
 a do-it-yourself solution developed by the Sankara Eye illuminated area with sufficient contrast on all videos. Image
Foundation31 (“direct SBFI device 3,” Fig 1C); quality of a sharpness/focus grade of 1 or less, a reflex artifact
 Paxos Scope adapter (“indirect SBFI device,” Digisight grade 0 or a contrast and illumination grade 1 were considered
Technologies Inc., now Verana Health Inc., San Francisco, insufficient. Compliance with the imaging protocol was assessed
CA, adapter version from 2017, Fig 1D); and according to a stepwise scale for imaging of the macular and the 4
 conventional 7-field color fundus photography (3nethra arcades (Fig S3, available at www.aaojournal.org). Presence of DR
royal, Forus Health Pvt Ltd., Bangalore, India). and diabetic maculopathy were assessed according to the
The first 3 SBFI devices were based on direct ophthalmoscopy, International Clinical Diabetic Retinopathy Severity Scale.32
and the last device was based on indirect ophthalmoscopy. The 3 Field-of-view was compared on exemplary single images. No
direct SBFI devices were used with Samsung Galaxy S4 smart- postprocessing was applied to any of the SBFI image material
phones (Samsung Electronics, Seoul, South Korea) and the indirect included in this article. Analyses of image quality, compliance with

1530
 Wintergerst et al 
Smartphone-Based Fundus Imaging in DR

Figure 1. Different approaches for smartphone-based fundus imaging (SBFI) compared in this study. Direct SBFI approach 1 (A), direct SBFI approach 2
(B), direct SBFI approach 3 (C), and indirect SBFI approach (D) fully equipped and ready for examination.

the imaging protocol, examination time, and presence of DR were stage 2/3/4 according to the International Council of Ophthal-
done by a masked study assistant (L.H.) experienced in image quality mology and the American Diabetes Association.33 Because many
assessment and DR grading on a 27-inch monitor. Where necessary, low- and middle-income countries implement cutoffs for referrals
single frames were reviewed in detail. All videos from eyes with any only at a later stage,34 we also included the diagnostic accuracy
positive finding for DR in any of the imaging modalities or the measures for detection of DR stages 3/4. The KruskaleWallis test
reference standard were additionally analyzed for image quality, was used for multiple comparison between groups for nonpara-
compliance with the imaging protocol, and presence of DR by an metric data and analysis of variance for parametric data. Correlation
ophthalmologist (M.W.M.W.). In case of discrepancies, the grading of the DR grade with the reference standard, weighted Cohen’s
of the more experienced reader (M.W.M.W.) was used for further kappa, and sensitivity and specificity were calculated to assess
analysis. A random subset of eyes was analyzed a second time (at agreement with the reference standard. For assessment of inter- and
least 3 months apart from the first analysis) by the first reader (L.H.) intra-reader reliability, weighted Cohen’s kappa was calculated.
to calculate intra-reader reliability. Statistical analyses were performed with R (R: A Language and
Environment for Statistical Computing, R Core Team, R Founda-
Statistical Analyses tion for Statistical Computing, Vienna, Austria, 2016).
Image quality, examination time, agreement of the DR grading with Role of the Funding Source
the reference standard, and sensitivity/specificity to detect any DR,
referral-warranted DR, and diabetic maculopathy were compared The funders had no role in study design, data collection, data
between the devices. Referral-warranted DR was defined as DR analysis, data interpretation, or writing of the report. The

1531
 Ophthalmology Volume 127, Number 11, November 2020

corresponding author had full access to all data in the study and Table 1. Demographics
had final responsibility for the decision to submit for publication.
Mean ± SD or
No. of Eyes (%)
Results
Age*
Mean (yrs) 56.6410.85
Demographics and Clinical Characteristics Range (yrs) 23e76
A total of 381 eyes of 193 patients with diabetes were included for Sex*
whom 1.445 videos (35 hours of total video material) and 1.800 Male 80 (41%)
single images were acquired. Pupil size (mean  standard devia- Female 113 (59%)
tion) after dilation was 7.31.1 mm. Three eyes were excluded Visual acuity (logMAR þ
rounded Snellen equivalent)
because no imaging was possible for any of the modalities because
Mean 0.430.50 (20/50)
of severe media opacities (Fig S4, available at Range 0e3.0 (20/20eno
www.aaojournal.org). No videos were recorded because of light perception)
software or other technical problems in 10 (2.6%), 24 (6.3%), 27 Lens status
(7.1%), and 13 (3.4%) eyes with the direct SBFI devices 1, 2, 3, Phakic, clear lens 142 (38%)
and indirect SBFI device, respectively. Table 1 shows the Cataract 191 (51%)
characteristics of the sample. Pseudophakic 39 (10%)
A total of 63 patients (102 eyes) were referred for management Aphakic 1 (0.3%)
of DR (45%), cataract (42%), and for other reasons (17%, in total 9 DR
patients: 3 for branch retinal vein occlusion, 2 for suspected neo- No DR 250 (66%)
vascular age-related macular degeneration, 2 for posterior capsule Stage 1 34 (9%)
opacification, 1 for suspected glaucoma, and 1 for macular hole). Stage 2 81 (22%)
Stage 3 7 (1.9%)
Evaluation of Image Quality and Field of View Stage 4 4 (1.1%)
Diabetic maculopathy
All 4 SBFI approaches yielded sharp images for most of the ex- Diabetic maculopathy present 54 (14%)
aminations (Fig 2A), whereas there was greater heterogeneity No diabetic maculopathy 319 (86%)
regarding reflex artifacts (Fig 2B) and image contrast and Duration of DM*
illumination (Fig 2C). Some 68%, 17%, 16%, and 8.2% of the mean (mo) 83.5779.08
videos were classified as insufficient image quality for the direct range (mo) 1e384
SBFI devices 1, 2, and 3 and the indirect SBFI device, Unknown 16 (8.3%)
Antidiabetic medication*
respectively. Sharpness/focus, reflex artifacts, and contrast and
None 19 (9.8%)
illumination were significantly different between SBFI devices (P
Oral 157 (81%)
values for KruskaleWallis tests were 0.0031, <0.0001, and Insulin 8 (4.1%)
<0.0001, respectively). Exemplary images for comparison of im- Insulin þ oral 9 (4.7%)
age quality and field-of-view between devices are provided in Fig DM complications, selfereported*
3. Overall, the best image quality and largest field of view were None 182 (94%)
achieved with the indirect SBFI device. The indirect SBFI device Kidney disease 4 (2.1%)
allowed for examination of a large fundus area (Fig S5, available Heart disease 5 (2.6%)
at www.aaojournal.org). Stroke 1 (0.5%)
Diabetic foot ulcer 1 (0.5%)
Comparison of Examination Time
Mean examination time ranged between 68 and 111 seconds and DM ¼ diabetes mellitus; DR ¼ diabetic retinopathy; logMAR ¼ logarithm
of the minimum angle of resolution; SD ¼ standard deviation.
was shortest for the direct SBFI device 1 (6828 seconds), fol-
*Reported patient-wise; no * reported eye-wise.
lowed by the direct SBFI device 3 (7940), direct SBFI device 2
(8646), and indirect SBFI device (11161 seconds; P < 0.0001).

Compliance with the Imaging Protocol specificity for detection of DR stages 3 and 4 were 0.55 (0.23-
0.83) / 1.0 (0.99-1.0), 0.64 (0.31-0.89) / 1.0 (0.99-1.0), 0.57
The overall average score for compliance with the imaging pro- (0.18-0.90) / 1.0 (0.98-1.0), and 1.0 (0.72-1.0) / 1.0 (0.98/1.0)
tocol was 10.23.6 of the maximum of 13 points. The imaging for direct SBFI approaches 1, 2, and 3 and the indirect approach,
protocol was more easily completed with the indirect SBFI device respectively. Exemplary images for detectability of DR and
compared with the direct SBFI devices (P < 0.0001; Fig S6, diabetic maculopathy between devices are provided in Figure 3.
available at www.aaojournal.org). Assessment of diabetic maculopathy was sometimes impaired on
direct SBFI because there was less spatial orientation compared
Diagnostic Accuracy of Smartphone-Based with indirect SBFI, that is, it was not always clear where the
Fundus Imaging current field-of-view was located on the fundus.
Agreement of DR grading on SBFI with the reference standard Analysis of Inter- and Intra-Reader Reliability
ranged from moderate to excellence. The approach for indirect
SBFI achieved the strongest correlation and agreement with the A total of 520 of the 1445 videos (36%; all videos from eyes with
clinical grading by indirect ophthalmoscopy, as well as highest any positive finding for DR in any of the imaging modalities or the
sensitivities and specificities for detection of any DR, referral- reference standard) were graded by 2 readers (M.W.M.W. and
warranted DR, and diabetic maculopathy (Table 2). Sensitivity/ L.H.). Agreement ranged from substantial to excellent for the

1532
 Wintergerst et al 
Smartphone-Based Fundus Imaging in DR

grading of image quality and was substantial for grading of

compliance with the imaging protocol, DR, and diabetic macul-
opathy (Table S1, available at www.aaojournal.org). Videos from
78 eyes were analyzed 2 times by L.H. (at least 3 months apart)
for analysis of intra-reader reliability. Agreement was excellent
for grading of image quality, substantial for grading of compliance
with the imaging protocol, and excellent for grading of DR and
diabetic maculopathy (Table S1, available at www.aaojournal.org).

Discussion

This is the first study directly comparing different ap-

proaches for SBFI in terms of image quality, field-of-view,
examination time, and applicability to DR screening, and it
took place in a real-world outreach setting in India. We
found significant differences in image quality between SBFI
devices, which were also reflected in varying sensitivity and
specificity to detect DR. Overall, the indirect approach for
SBFI yielded the best image quality, largest field of view,
and best diagnostic accuracy in detecting DR.
Our study supports existing data indicating that SBFI can
yield acceptable results when used for DR screening pur-
poses.35,36 With the majority of persons with diabetes living
in low- and middle-resources settings3,7,8 without any
established DR screening programs due to resource
constraints,4,6,7,10-14 SBFI may offer an inexpensive, mo-
bile screening solution for many low- and middle-income
countries using virtually ubiquitous smartphones.
Russo et al25 evaluated a device, which was also included
in our study (direct SBFI device 2), and found substantial
agreement with clinical evaluation (kappa 0.78) and a
sensitivity/specificity to detect DR of between 0.55/0.99
(stage 3) and 0.89/1.0 (stage 4) and of 0.89/1.0 to detect
diabetic maculopathy.25 Toy et al27 evaluated the indirect
SBFI device, which was also included in our study and
found excellent agreement with clinical evaluation (kappa
0.91) and a sensitivity/specificity of 0.91/0.99 to detect
moderate nonproliferative or worse DR. Both studies
found a better performance of the 2 evaluated devices
than we did; however, comparability is limited, because
their studies took place in a tertiary hospital setting,
whereas our study took place in an off-site outreach eye
clinic setting. Furthermore, SBFI was performed by oph-
thalmologists only in the mentioned studies, whereas it was
also performed by trained optometrists in our study. Our
study tested applicability to DR screening under more
realistic conditions, that is, in a field-readiness scenario,
potentially yielding worse image quality and protocol
compliance for those reasons.
Systematic comparison of different imaging devices re-
quires standardized assessment of image parameters
including quality. To date, no such standards exist for SBFI
imaging.37,38 Standardized scales for the assessment of
image quality proposed for smartphone-based medical im-
Figure 2. Image quality of the different approaches for smartphone-based
fundus imaging (SBFI). The frequencies of semiquantitative image qual-
aging in other disciplines are not applicable to SBFI imag-
ity grades for sharpness/focus (A), reflex artifacts (B), and contrast and ing.39-41 Therefore, we have developed semiquantitative
illumination (C) are displayed for the 4 approaches for SBFI compared in image quality scales for sharpness/focus, reflex artifacts,
this study. Higher grades correspond to a better image quality (Fig S2 contrast, and illumination with substantial to excellent intra-
[available at www.aaojournal.org] shows the respective semiquantitative and inter-reader reliability. Furthermore, we have developed
image quality scales). a scale to evaluate compliance with the imaging protocol,

1533
 Ophthalmology Volume 127, Number 11, November 2020

Figure 3. Exemplary comparison of image quality between different approaches for smartphone-based fundus imaging (SBFI). Images in each row were
acquired from the same eye using the 4 different approaches for SBFI. Two topmost rows: healthy phakic eyes; third row: healthy pseudophakic eye; fourth
row: phakic eye with diabetic maculopathy; bottom row: phakic eye with proliferative diabetic retinopathy (DR).

1534
 Wintergerst et al 
Smartphone-Based Fundus Imaging in DR

0.97 (0.94e0.98)

0.98 (0.96e0.99)

0.98 (0.95e0.99)
providing means for quality control in future SBFI studies.

1.0 (0.98e1.0)
On the basis of our results, image quality and field of view

Specificity
are directly related to both sensitivity and specificity, and

Diabetic Maculopathy
thus should be assessed and reported for all used SBFI de-
vices. Against this background, image quality is crucial

(95% CI)
when it comes to applicability, and both referral rates and
cost-effectiveness are likely strongly affected. This might
0.60 (0.45e0.73)

0.58 (0.44e0.72)

0.64 (0.49e0.77)

0.79 (0.65e0.89)
especially be the case in regions with low cataract surgical
rates and coverage (e.g., sub-Saharan Africa) where a high
Sensitivity

prevalence of media opacity could further aggravate the

observed differences in image quality and diagnostic accu-
racy.42 This emphasizes the need for a standardized
assessment of SBFI image quality as proposed by our study.
As tele-ophthalmologic DR screening has already proven
Table 2. Diagnostic Accuracy of Smartphone-Based Fundus Imaging versus the Reference Standard

a feasible approach for areas hard to reach,10,13,14,16-18 and

0.98 (0.96e0.99)
1.0 (0.98e1.0)

0.99 (0.97e1.0)

0.99 (0.97e1.0)

as the WHO and the International Diabetes Federation have

Specificity

prioritized low-cost mobile screening technologies by non-

Referral-Warranted DR*

physicians,7,14,15 SBFI by paramedical staff would be

another natural application for tele-ophthalmologic DR
(95% CI)

screening. Training requirements for SBFI are an important

aspect when assessing applicability in different settings. In
0.21 (0.13e0.31)

0.41 (0.30e0.52)

0.57 (0.46e0.68)

0.76 (0.66e0.85)

this study, we trained optometrists; thus, we cannot

comment on the applicability by paramedical staff/techni-
Sensitivity

cians without any prior experience in ophthalmoscopy, but

research on the SBFI learning-curve dynamics of para-
medical staff/technicians is warranted. Being less expensive
compared with conventional retinal imaging systems, easily
integrated into web-based tele-ophthalmology applications,
0.96 (0.93e0.98)

0.96 (0.93e0.98)

0.94 (0.90e0.97)

and delegable to non-ophthalmologist staff, SBFI could

0.99 (0.97e1.0)

significantly lower the financial burden for tele-

Specificity

ophthalmologic DR screening. Artificial intelligence solu-

CI ¼ confidence interval; DR ¼ diabetic retinopathy; SBFI ¼ smartphone-based fundus imaging.

tions are another natural extension of SBFI-based DR

screening. In fact, there has been substantial progress
(95% CI)
Any DR

recently in the development of algorithms for automated DR

detection on retinal color images43,44 also using SBFI.45,46
0.52 (0.42e0.61)

0.59 (0.50e0.68)

0.73 (0.64e0.81)

0.79 (0.71e0.86)
Sensitivity

Study Strengths and Limitations

The strengths of our study are the application of SBFI in a
real-world, low-resources DR screening setting, and a
comprehensive direct comparison of 4 approaches for SBFI
evaluating 3 parameters for image quality, compliance with
Squared Cohen’s

the imaging protocol, and examination time. Most impor-

P < 0.0001

P < 0.0001

P < 0.0001

P < 0.0001

tant, we assessed applicability to DR screening and diag-

kappa

nostic accuracy against an established reference standard.

0.549

0.621

0.728

0.868

Additional strengths are the standardization of image quality

DR Stage

assessment using newly developed scales as well as a

comprehensive evaluation of intra- and inter-reader reli-
ability. All direct SBFI adapters were used with the same
P < 0.0001

P < 0.0001

P < 0.0001

P < 0.0001
correlation r
Spearman’s

smartphone; thus, results can be directly compared. To

0.592

0.627

0.715

0.853

achieve a comparable image resolution of the examined

field-of-view, the indirect SBFI adapter was used with an
*Defined as DR stage 2/3/4.

iPod Touch, which yields comparable image quality. In fact,

another study found no statistically significant differences
Direct SBFI device 1

Direct SBFI device 2

Direct SBFI device 3

Indirect SBFI device

for image quality of Apple and Samsung smartphone cam-

eras in the context of smartphone-based flexible fiberoptic
cystoscopy.40 Limitations of our study include the relatively
low prevalence of more severe stages of DR in our cohort.
However, this reflects a true outreach screening setting
where most participants would have no or very early signs

1535
 Ophthalmology Volume 127, Number 11, November 2020

of DR. The image quality scales are of a semiquantitative 8. Sabanayagam C, Banu R, Chee ML, et al. Incidence and
nature, whereas fully quantitative analyses would be progression of diabetic retinopathy: a systematic review.
preferable, but more difficult to realize. Using Lancet Diabetes Endocrinol. 2019;7:140e149.
semiquantitative scales has been a standard in 9. Singer DE, Nathan DM, Fogel HA, Schachat AP. Screening
ophthalmology for many decades, however, and is well for diabetic retinopathy. Ann Intern Med. 1992;116:660e671.
10. Ramasamy K, Raman R, Tandon M. Current state of care for
established.47-49 Another limitation is that no full masking diabetic retinopathy in India. Curr Diab Rep. 2013;13:
for image modality was possible because of typical image 460e468.
artifacts, for example, the very bright, well-circumscribed 11. Sasongko MB, Widyaputri F, Agni AN, et al. Prevalence of
crescent reflex on the direct SBFI device 3 and the 2 cen- diabetic retinopathy and blindness in Indonesian adults with
tral round small reflexes on the indirect SBFI device. type 2 diabetes. Am J Ophthalmol. 2017;181:79e87.
However, this is analogous to most comparisons of image 12. Song P, Yu J, Chan KY, et al. Prevalence, risk factors and
modalities (e.g., varying hue and color ranges on different burden of diabetic retinopathy in China: a systematic review
conventional color fundus photography devices and varying and meta-analysis. J Glob Health. 2018;8:010803.
noise and general brightness levels on OCT angiography 13. Murthy KR, Murthy PR, Kapur A, Owens DR. Mobile dia-
devices50), and if postprocessing would be applied, for betes eye care: experience in developing countries. Diabetes
Res Clin Pract. 2012;97:343e349.
example, normalization of brightness and contrast, some 14. Sabanayagam C, Yip W, Ting DS, et al. Ten emerging trends
bias would remain due to these characteristic artifacts. in the epidemiology of diabetic retinopathy. Ophthalmic Epi-
Furthermore, we did not stratify into different diabetic demiol. 2016;23:209e222.
maculopathy severity grades and cannot assess the 15. World Health Organization. Global Initiative for the Elimi-
diagnostic accuracy of SBFI for the assessment of diabetic nation of Avoidable Blindness Action Plan 2006e2011.
maculopathy severity. Geneva, Switzerland: WHO; 2007.
Our study demonstrates that SBFI may be a feasible 16. Jones S, Edwards RT. Diabetic retinopathy screening: a sys-
approach for screening of treatment-requiring DR in low- tematic review of the economic evidence. Diabet Med.
resources settings. However, there are marked differences 2010;27:249e256.
among the SBFI devices used here regarding image quality 17. Rachapelle S, Legood R, Alavi Y, et al. The cost-utility of
telemedicine to screen for diabetic retinopathy in India.
and diagnostic accuracy. Smartphone-based fundus imaging Ophthalmology. 2013;120:566e573.
might have the potential to alleviate the burden of DR 18. Shi L, Wu H, Dong J, et al. Telemedicine for detecting diabetic
screening in low- and middle-income countries. However, retinopathy: a systematic review and meta-analysis. Br J
agreement on standard SBFI imaging protocols, image Ophthalmol. 2015;99:823e831.
quality scales, and more studies on its integration into ser- 19. Zimmer-Galler IE, Kimura AE, Gupta S. Diabetic retinopathy
vice provision appear prudent. screening and the use of telemedicine. Curr Opin Ophthalmol.
2015;26:167e172.
Acknowledgment 20. Lord RK, Shah VA, Filippo ANS, Krishna R. Novel uses of
smartphones in ophthalmology. Ophthalmology. 2010;117:
The authors thank the optometrists Bhagyashree Rajesh Rawal, 1274e1274.e3.
Kala N., Anugna S., Y. Syed Noor Mohammed, and Nagaraju 21. Bastawrous A. Smartphone fundoscopy. Ophthalmology.
Pamarthi for their diligent support; and HelpAge India for their 2012;119:432e433.e2.
support in organizing outreach eye clinics. 22. Bolster NM, Giardini ME, Livingstone IA, Bastawrous A.
How the smartphone is driving the eye-health imaging revo-
lution. Exp Rev Ophthalmol. 2014;9:475e485.
References 23. Jani PD, Forbes L, Choudhury A, et al. Evaluation of diabetic
retinal screening and factors for ophthalmology referral in a
telemedicine network. JAMA Ophthalmol. 2017;135:
1. Cheung N, Mitchell P, Wong TY. Diabetic retinopathy. Lan- 706e714.
cet. 2010;376:124e136. 24. Ryan ME, Rajalakshmi R, Prathiba V, et al. Comparison
2. Yau JW, Rogers SL, Kawasaki R, et al. Global prevalence and among methods of retinopathy assessment (CAMRA) study
major risk factors of diabetic retinopathy. Diabetes Care. smartphone, nonmydriatic, and mydriatic photography.
2012;35:556e564. Ophthalmology. 2015;122:2038e2043.
3. International Diabetes Federation. IDF Diabetes Atlas. 8th ed. 25. Russo A, Morescalch F, Costagliola C, et al. Comparison of
Brussels, Belgium: IDF; 2017. smartphone ophthalmoscopy with slit-lamp biomicroscopy for
4. World Health Organization. Global Report on Diabetes. grading diabetic retinopathy. Am J Ophthalmol. 2015;159:
Geneva, Switzerland: WHO; 2016. 360e364.
5. Flaxman SR, Bourne RRA, Resnikoff S, et al. Global causes of 26. Rajalakshmi R, Arulmalar S, Usha M, et al. Validation of
blindness and distance vision impairment 1990-2020: a sys- smartphone based retinal photography for diabetic retinopathy
tematic review and meta-analysis. Lancet Glob Health. screening. PLoS One. 2015;10.
2017;5:e1221ee1234. 27. Toy BC, Myung DJ, He L, et al. Smartphone-based dilated
6. Ruta LM, Magliano DJ, Lemesurier R, et al. Prevalence of fundus photography and near visual acuity testing as inex-
diabetic retinopathy in type 2 diabetes in developing and pensive screening tools to detect referral warranted diabetic
developed countries. Diabet Med. 2013;30:387e398. eye disease. Retina. 2016;36:1000e1008.
7. International Diabetes Federation. The Diabetic Retinopathy 28. Kim TN, Myers F, Reber C, et al. A smartphone-based tool for
Barometer Report: Global Findings. Brussels, Belgium: IDF; rapid, portable, and automated wide-field retinal imaging.
2017. Transl Vis Sci Technol. 2018;7:21.

1536
 Wintergerst et al 
Smartphone-Based Fundus Imaging in DR

29. Bilong Y, Katte JC, Koki G, et al. Validation of smartphone- 41. Lu S, Cottone CM, Yoon R, et al. Endoscope: a disrup-
based retinal photography for diabetic retinopathy screening. tive endoscopic technology. J Endourol. 2019;33:
Ophthalmic Surg Lasers Imaging Retina. 2019;50:S18eS22. 960e965.
30. Petersmann A, Nauck M, Muller-Wieland D, et al. Definition, 42. Rao GN, Khanna R, Payal A. The global burden of cataract.
classification and diagnosis of diabetes mellitus. Exp Clin Curr Opin Ophthalmol. 2011;22:4e9.
Endocrinol Diabetes. 2018;126:406e410. 43. Ting D, Cheung C, Lim G, et al. Development and validation
31. Shanmugam MP, Mishra DKC, Madhukumar R, et al. Fundus of a deep learning system for diabetic retinopathy and related
imaging with a mobile phone: a review of techniques. Indian J eye diseases using retinal images from multiethnic populations
Ophthalmol. 2014;62:960e962. with diabetes. JAMA. 2017;318:2211e2223.
32. Wilkinson CP, Ferris 3rd FL, Klein RE, et al. Proposed in- 44. van der Heijden AA, Abramoff MD, Verbraak F, et al. Vali-
ternational clinical diabetic retinopathy and diabetic macular dation of automated screening for referable diabetic retinop-
edema disease severity scales. Ophthalmology. 2003;110: athy with the IDx-DR device in the Hoorn Diabetes Care
1677e1682. System. Acta Ophthalmol. 2018;96:63e68.
33. Vujosevic S, Aldington SJ, Silva P, et al. Screening for dia- 45. Rajalakshmi R, Subashini R, Anjana RM, Mohan V. Auto-
betic retinopathy: new perspectives and challenges. Lancet mated diabetic retinopathy detection in smartphone-based
Diabetes Endocrinol. 2020;8:337e347. fundus photography using artificial intelligence. Eye (Lond).
34. Wang LZ, Cheung CY, Tapp RJ, et al. Availability and vari- 2018;32:1138e1144.
ability in guidelines on diabetic retinopathy screening in Asian 46. Natarajan S, Jain A, Krishnan R, et al. Diagnostic accuracy of
countries. Br J Ophthalmol. 2017;101:1352e1360. community-based diabetic retinopathy screening with an off-
35. Taylor R, Broadbent DM, Greenwood R, et al. Mobile retinal line artificial intelligence system on a smartphone. JAMA
screening in Britain. Diabet Med. 1998;15:344e347. Ophthalmol. 2019;137(10):1182e1188.
36. Papavasileiou E, Dereklis D, Oikonomidis P, et al. An effec- 47. Nussenblatt RB, Palestine AG, Chan CC, Roberge F.
tive programme to systematic diabetic retinopathy screening in Standardization of vitreal inflammatory activity in inter-
order to reduce diabetic retinopathy blindness. Hell J Nucl mediate and posterior uveitis. Ophthalmology. 1985;92:
Med. 2014;17(Suppl 1):30e34. 467e471.
37. DeBuc DC. The role of retinal imaging and portable screening 48. Chylack Jr LT, Wolfe JK, Singer DM, et al. The Lens
devices in tele-ophthalmology applications for diabetic reti- Opacities Classification System III. The Longitudinal Study
nopathy management. Curr Diabet Rep. 2016;16:132. of Cataract Study Group. Arch Ophthalmol. 1993;111:
38. Mohammadpour M, Heidari Z, Mirghorbani M, Hashemi H. 831e836.
Smartphones, tele-ophthalmology, and VISION 2020. Int J 49. Jabs DA, Nussenblatt RB, Rosenbaum JT. Standardization of
Ophthalmol. 2017;10:1909e1918. Uveitis Nomenclature (SUN) Working Group. Standardization
39. Wyatt KD, Willaert BN, Pallagi PJ, et al. PhotoExam: adop- of Uveitis Nomenclature for Reporting Clinical Data. Results
tion of an iOS-based clinical image capture application at of the First International Workshop. Am J Ophthalmol.
Mayo Clinic. Int J Dermatol. 2017;56:1359e1365. 2005;140:509e516.
40. Tse C, Patel RM, Yoon R, et al. The endoscope using next 50. Munk MR, Giannakaki-Zimmermann H, Berger L, et al.
generation smartphones: “a global opportunity”. J Endourol. OCT-angiography: a qualitative and quantitative comparison
2018;32:765e770. of 4 OCT-A devices. PLoS One. 2017;12:e0177059.

Footnotes and Financial Disclosures

Originally received: January 21, 2020. Vision, Kanghong, LinBioscience, Oxurion, Stealth BioTherapeutics; Per-
Accepted: May 11, 2020. sonal fees e Boehringer-Ingelheim, Geuder; Nonfinancial support e Cen-
Available online: May 14, 2020. Manuscript no. D-20-00141 terVue, Ellex, NightStarX, Optos, outside the submitted work.
1
Department of Ophthalmology, University Hospital Bonn, Bonn, R.F.: Grants e Else Kroener-Fresenius Foundation and German Scholars Or-
Germany. ganization (EKFS/GSO 16); Grants, Personal fees, Nonfinancial support e
2
Sankara Academy of Vision, Sankara Eye Hospital Bangalore, Bangalore, Novartis; Personal fees e Bayer, Roche/Genentech, Alimera, Santhera, Ellex,
India. Opthea, Inositec; Nonfinancial support e CentreVue, Heidelberg Engineering,
3 Optos, Zeiss Meditec, outside the submitted work. Supported by the German
Department of Medical Biometry, Informatics and Epidemiology, Uni-
Ophthalmic Society (Tropenophthalmologie research grant), the German
versity Hospital Bonn, Bonn, Germany.
Retina Society (travel grant) and the BONFOR GEROK Program, Faculty of
*M.P.S. and R.P.F. contributed equally. Medicine, University of Bonn, (Grant No. O-137.0028) [to M.W.M.W.]) and
Financial Disclosure(s): by the Else Kroener-Fresenius Foundation/German Scholars Organization
The author(s) have made the following disclosure(s): M.W.: Grants e (EKFS/GSO 16 [to R.P.F.]).
German Ophthalmic Society, German Retina Society, BONFOR GEROK HUMAN SUBJECTS: Human subjects were included in this study. Ethical
Program, Faculty of Medicine, University of Bonn (Grant No. O- approval was obtained from the ethics committee at the Sankara Eye
137.0028), German Federal Ministry for Economic Cooperation and Hospital Bangalore, India (ethics committee registration number ECR/705/
Development / Else Kroener-Fresenius Foundation (Health Partnerships - Inst/KA/2015) and the University of Bonn, Germany (ethics committee ID
Global Program); Grants, Personal Fees, Nonfinancial support e Heine 010/18), and informed consent was obtained from all study participants
Optotechnik; Personal Fees, Nonfinancial support e Askin & Co; Grants before study initiation. All research adhered to the tenets of the Declaration
and Nonfinancial support e Berlin-Chemie AG, outside the submitted of Helsinki.
work.
No animal subjects were used in this study.
F.H.: Grants, Personal fees, Nonfinancial support e Acucela, Bayer, Roche/
Author Contributions:
Genentech, Heidelberg Engineering, Novartis, Pixium Vision; Grants,
Nonfinancial support e Allergan, Zeiss, outside the submitted work; Per- Conception and design: Wintergerst, Mishra, Murali, Holz, Shanmugam,
sonal fees, Nonfinancial support e Appelis, Bioeq/Formycon, Grayburg Finger

1537
 Ophthalmology Volume 127, Number 11, November 2020
Data collection: Wintergerst, Mishra, Shah, Konana, Sagar Abbreviations and Acronyms:
Analysis and interpretation: Wintergerst, Mishra, Hartmann, Berger, DR ¼ diabetic retinopathy; SBFI ¼ smartphone-based fundus imaging;
Shanmugam, Finger WHO ¼ World Health Organization.
Obtained funding: Wintergerst, Finger Correspondence:
Overall responsibility: Wintergerst, Mishra, Hartmann, Shah, Konana, Robert P. Finger, PhD, Department of Ophthalmology, University Hospital
Sagar, Berger, Murali, Holz, Shanmugam, Finger Bonn, Ernst-Abbe-Straße 2, 53127 Bonn, Germany. E-mail: robert.finger@
ukbonn.de.

Pictures & Perspectives

Scanning Electron Microscopic of Intraocular Lens Pits after Nd:YAG Capsulotomy

A 69-year-old unstable and excessively mobile patient on the Nd:YAG laser, who inadvertently had multiple pitting of the intraocular
lens (IOL) during several attempts of capsulotomy elsewhere, presented with reduced visual acuity of 20/200, halo and glare. Figure A
shows his hydrophilic acrylic IOL (Akreos AO, Bausch & Lomb, Rochester, NY) with extensive pitting. He underwent IOL exchange for
visual restoration. Figures B and C demonstrate the magnified view (30 and 100, respectively) of the IOL surface on scanning electron
microscopy. Figure D shows the petaloid feature of Nd:YAG effect on this IOL, which is magnified 1140. (Magnified version of Fig 1A-F
is available online at www.aaojournal.org).
MAYANK A. NANAVATY, FRCOPHTH, PHD1,2
JONATHAN SALVAGE, BSC, PHD3
ZAHRA ASHENA, FRCOPHTH1
1
Sussex Eye Hospital, Brighton & Sussex University Hospitals NHS Trust, Eastern Road, Brighton, United Kingdom; 2Brighton & Sussex Medical
School, University of Sussex, Falmer, Brighton. United Kingdom; 3Image and Analysis Unit, University of Brighton, Brighton, United Kingdom

1538