key: cord-323273-q53wf6au authors: Olivia Li, Ji-Peng; Liu, Hanruo; Ting, Darren S.J.; Jeon, Sohee; Chan, R.V.Paul; Kim, Judy E.; Sim, Dawn A.; Thomas, Peter B.M.; Lin, Haotian; Chen, Youxin; Sakomoto, Taiji; Loewenstein, Anat; Lam, Dennis S.C.; Pasquale, Louis R.; Wong, Tien Y.; Lam, Linda A.; Ting, Daniel S.W. title: Digital technology, tele-medicine and artificial intelligence in ophthalmology: A global perspective date: 2020-09-06 journal: Prog Retin Eye Res DOI: 10.1016/j.preteyeres.2020.100900 sha: doc_id: 323273 cord_uid: q53wf6au The simultaneous maturation of multiple digital and telecommunications technologies in 2020 has created an unprecedented opportunity for ophthalmology to adapt to new models of care using tele-health supported by digital innovations. These digital innovations include artificial intelligence (AI), 5th generation (5G) telecommunication networks and the Internet of Things (IoT), creating an inter-dependent ecosystem offering opportunities to develop new models of eye care addressing the challenges of COVID-19 and beyond. Ophthalmology has thrived in some of these areas partly due to its many image-based investigations. Tele-health and AI provide synchronous solutions to challenges facing ophthalmologists and healthcare providers worldwide. This article reviews how countries across the world have utilised these digital innovations to tackle diabetic retinopathy, retinopathy of prematurity, age-related macular degeneration, glaucoma, refractive error correction, cataract and other anterior segment disorders. The review summarises the digital strategies that countries are developing and discusses technologies that may increasingly enter the clinical workflow and processes of ophthalmologists. Furthermore as countries around the world have initiated a series of escalating containment and mitigation measures during the COVID-19 pandemic, the delivery of eye care services globally has been significantly impacted. As ophthalmic services adapt and form a “new normal”, the rapid adoption of some of telehealth and digital innovation during the pandemic is also discussed. Finally, challenges for validation and clinical implementation are considered, as well as recommendations on future directions. consolidation of tele-health, the development of 5 th generation wireless networks 152 (5G), artificial intelligence (AI) approaches such as machine learning (ML) and deep 153 learning (DL), and the Internet of Things (IoT), as well as digital security capabilities 154 such as blockchain, have created an extraordinary ecosystem for new opportunities 155 in healthcare and other industries ). These developments could 156 potentially address some of the most urgent challenges facing health service In 2019, WHO started developing a framework for the adoption of digital innovations 189 and technology in healthcare. The WHO recommendations on digital interventions in 190 healthcare promotes assessment on the basis of 'benefits, harms, acceptability, 191 feasibility, resource use and equity considerations', and views these tools as still 192 very much that -tools -in the journey to achieving universal health coverage and 193 sustainability (World Health Organisation 2019). There are several digital interventions that have been prioritised for review by the 196 WHO. Of relevance to this discussion are: the use of client-to-provider telemedicine 197 to complement health service delivery; the use of provider-to-provider telemedicine; 198 targeted customised health information transmission; health worker decision making 199 support; digitised health information tracking; and education. In all these scenarios, 200 the review highlights the need for monitoring of patient safety, privacy, traceability, 201 accountability and security, with plans in place to address any breaches. Processes 202 for these have been innate within the pharmaceutical and other medical devices 203 industries, and new technological entrants to this traditional sector should consider 204 these during development of the services. There will also be ethical conundrums that 205 have yet to be articulated and debated. The engaged clinician should seek to be 206 involved in the development of these new advances to closely align any innovations 207 to solve unmet clinical needs. Simultaneously, clinicians should examine if any 208 innovation complies with quality, ethical, and sustainable healthcare, as legislation 209 invariably lags behind such momentous leaps in innovation. Telemedicine enable clinicians to evaluate their patients remotely. This can be 216 desirable for several reasons. First, telemedicine can facilitate more efficient and 217 equitable distribution of limited healthcare resources. This allows delivery of care to 218 distant areas where there is a shortage of doctors and other professionals, reduces 219 travel and the associated carbon footprints, and connects patients with rare diseases 220 to speciality care and address the transport challenges some patients face. Waiting 221 times could be reduced through increased capacity and access to care for both 222 chronic and acute disease patient. In the acute setting, patients could receive 223 immediate specialist input even if one is not available locally. Second, amid the COVID-19 pandemic and in mitigating infection risk in the 226 healthcare setting, real-time telemedicine has been rapidly incorporated into routine 227 care delivery. The patient population telemedicine aims to serve is no-longer focused 228 on targeting remote regions. Instead it is rapidly becoming a new standard of care. It 229 enables triaging prior to patients' arrival into hospital to avoid unnecessary visits and 230 exposure risks and has been adopted by multiple centres across the world 231 J o u r n a l P r e -p r o o f (Hollander and Third, video-consultations in combination with innovative service design already exist 235 that further limits patient journeys and clinic visits whilst maximising the quality of the 236 telemedicine consultation. In Scotland, optometric practices have been set up 237 strategically across some regions to provide primary eye care services (NHS 238 Scotland 2020). Smart phones attached to slit-lamps enable ocular biomicroscopic 239 videography, empowering ophthalmologists to view the patient's examination 240 features in real-time without the patient attending. Also, simplification of image 241 sharing of data such as OCT scans can be achieved by screen sharing, which has 242 long been a challenge both within ophthalmology and in radiology due to the variety 243 of available formats and software. devices such as tonometers may be prohibitively expensive. Effective tele-screening programmes require multiple components. First, there 251 should be a reliable, cost-effective and operator-friendly data gathering system. A 252 preferred goal is to achieve longitudinal consistency of data format to facilitate 253 comparisons. The device itself should be simple, with mechanisms in place to 254 facilitate data transmission to the IoT. Ideal designs should involve networks where 255 multiple, simpler devices can communicate with a central station. System updates 256 would involve the central stations to enable streamlined logistics and cost efficiency, 257 particularly if the network has widely dispersed simpler devices. Second, the data must be processed and enabled to identify the disease of interest. 260 The most frequently adopted model at present is the use of trained persons to read 261 the collected images, as in diabetes tele-retinal screening programmes. Whilst larger 262 numbers can be screened this way in comparison to direct clinician reviews, it 263 remains a costly and resource intensive process involving highly trained graders. 264 While DL is starting to be incorporated to this process, the potential benefits from this 265 adaptation are unknown. Regulatory bodies recognise the potential of AI in 266 healthcare, and the FDA has approved the use of an AI algorithm for the diagnosis of 267 DR in the primary care setting (Abramoff et al. 2018 ). Finally, the outcome must be conveyed in a timely manner to the patient and the 270 healthcare provider to facilitate appropriate medical management. This 271 communication again could involve a clinician consultation, but most normal 272 outcomes may be communicated in an automated manner such as via a smart 273 phone app or text message. 274 275 J o u r n a l P r e -p r o o f Beyond simply replicating current services albeit remotely, the collection, storage 276 and transmission of offer the potential of combining telemedicine with AI. When used 277 prospectively with longitudinal data, vast swathes of new knowledge such as disease 278 progression and real-world, real-time incidence calculation could be harnessed. If 279 well adopted, the data collected would enter the realms of big data, and far exceed 280 the capabilities of data capture that most individual studies are able to achieve. 281 Moreover, this could grow into a consistent source of longitudinal data which would 282 be valuable in the development of disease progression forecasting capabilities, 283 incorporating AI. 284 285 2.2 5 th Generation (5G) telecommunications 286 287 5G wireless communications was designed to meet the challenges of serving large-288 scale complex network connections. These networks have extremely low latency, 289 higher capacity, and improve the speed of data transmission through the use of 290 higher frequency millimetre waves compared to existing networks (Simko and 291 Mattsson 2019). Latency in 5G transmission can be less than 1 millisecond of delay 292 compared to about 70milliseconds on the 4G network, and give significant 293 improvement to the users' perception of the service (Samsung 2015) . Download 294 speeds on 5G networks can be increased 20 fold from the current 1 gigabit per 295 second on 4G (Nordrum, Clark, and staff. 2017) . And all this magnitude increase in 296 function whilst simultaneously reducing energy consumption by the connected 297 devices (Agiwal, Roy, and Saxena 2016). 5G networks will deliver an end-to-end 298 latency of less than 5 milli-seconds and over-the-air latency of less than one 299 millisecond -which is one-tenth of the 4G network latency (Samsung 2015) . 300 301 5G utilises small cells, which are miniature base stations that have low power 302 requirements. However, because 5G transmits at higher frequencies, signal 303 attenuation becomes a greater challenge, and these base stations need to be placed 304 closer than 4G base stations (every 250 meters or so) (National Academies of 305 Sciences et al. 2019). To ensure consistent signal transmission, base stations will 306 need to be densely populated. Despite the base stations being smaller in size, the 307 increased infrastructure needs of a 5G network with these cells will not be practical 308 in sparsely populated rural regions. Thus whilst telemedicine has been traditionally 309 regarded as being able to contribute to healthcare delivery to these areas in a 310 meaningful way, it may in fact continue to exclude those who already struggle to 311 access physical care. 312 313 In addition to being able to support increasing bandwidth demands from users and 314 patients, 5G enables Ultra-High-Definition (UHD) multimedia streaming with 315 enhanced user experience. The high-resolution images can be more easily 316 transferred. Better quality and reliable video-consultations with improved patient 317 experience may contribute to forging better physician-patient relationship. Real-time 318 slitlamp examinations streamed in high-definition has the potential to become 319 J o u r n a l P r e -p r o o f common place. With imperceptible latency, the clinician could control a slit-lamp 320 remotely whilst looking at a mobile device displaying the eye being examined 321 remotely. The immersive experience promised by 5G can also be used to augment 322 the learning experience, particularly the visually-based tasks such as surgery. Despite these great expectations, 5G will not be the panacea for all connectivity 325 challenges. The reported speeds assume that every network is using 5G, but not 326 surprisingly the implementation of 5G will be gradual as new cells are built and 327 installed. This incremental adoption of expensive infrastructure means that the 328 network will need to remain compatible with legacy networks, and with other 329 operators who may be implementing at a different speed (Rashid 2020) . 330 331 In being compatible, and with the networks essentially being a patchwork of wireless 332 connections incorporating various generations, the same vulnerabilities found in 333 older generation networks will remain. Well-knowns flaws of the data packet 334 transmission protocol that is used across the different generations of networks, the 335 General Packet Radio Service (GPRS) Tunneling Protocol (GTP), include not 336 validating users' physical location permitting attackers to spoof locations and 337 allowing attackers to impersonate other users or use false credentials, so the 338 impersonated subscriber is charged for costs incurred. Attackers can block all 339 connections stemming from a single node so legitimate subscribers cannot access a 340 connection in the given geographical region, in a denial-of-service attack (Rashid 341 2020). The most basic requirements of connectivity in healthcare are security and 342 reliability, and despite the impressive numbers 5G promises, it may be still some 343 time before these two basic tenets are consistently achieved. 344 345 2.1.1 5G and the COVID-19 pandemic 346 347 The lockdown orders across the world has brought a sudden strain on existing 348 cellular networks. As countries responded, work, education, healthcare, and most 349 other human interactions were suddenly pushed onto the virtual arena. The 350 pandemic has shown that telemedicine is not only reserved for the remote and 351 underserved. In fact, telemedicine can routinely serve the wider population if it can 352 be shown to be safe, efficient, and inclusive, with measures to ensure security, 353 robustness and capacity, particularly in densely populated regions with massive 354 competing demands for bandwidth. Though few examples currently exist, 5G telemedicine has already been 357 implemented. In China, the successful utilisation of a 5G telemedicine network was 358 reported in Sichuan province (Hong et al. 2020 process is time-consuming and costly, but also makes ophthalmology one of the 454 specialities particularly well-suited to DL techniques and its real-world application. 455 The application of DL to ophthalmic images, such as digital fundus photographs and 456 visual fields, has been reported to achieve the automated screening and diagnosis of 457 common vision-threatening diseases, including diabetic retinopathy (DR) ( healthcare is notably slower. There is a real risk that high hopes for the new 568 technologies described elsewhere in this paper will flounder upon the reality of 569 healthcare systems that remain digitally immature. Some barriers to innovation in 570 healthcare are perfectly legitimate, for example the real risk that sub-optimal 571 deployment of a digital technology could lead to patient harm. Other barriers are 572 entirely artificial, and foremost among these are the perverse incentives created by 573 billing and tariff systems. In the UK, for example, there has only recently been a 574 move to correct the imbalance between poorly reimbursed remote consultations and 575 well reimbursed face-to-face consultations (Brennan, Serle, and Clover 2018). 576 577 When a technology has successfully navigated the ethical, financial, regulatory, and 578 safety barriers to implementation in healthcare, the rate of attrition remains high. In 579 order to be scalable beyond local pilots, the technology must either fit in seamlessly 580 with existing clinical practice, or it must be sufficiently compelling to cause clinical 581 practice to change (as we have seen with OCT platforms in ophthalmology). The 582 failure of the UK's National Programme for IT is a case study for this 583 phenomenon(Robertson, Bates, and Sheikh 2011). Where local adoption has been 584 successful, innovations can be slow to spread through a fragmented system, with 585 funding for spread of innovation often a small fraction of the research and 586 development budget (Collins 2018) . A partial solution to these challenges has been the creation of innovation units 589 embedded in hospitals and academic medical centres (e.g. Cleveland Clinic 590 Innovations and the Digital Clinical Lab at Moorfields Eye Hospital). These units can 591 help to develop digital technologies that improve healthcare delivery in the real world, 592 rather than developing solutions that can't easily be incorporated into routine practice. 593 While innovation units can earmark resources, a major enabler is their ability to bring 594 together multi-disciplinary teams that allow the development of useful solutions. 595 These include, among others, engineers, developers, behavioural scientists, 596 intellectual property specialists, and clinicians. The development of local capabilities 597 to drive digital innovation mirrors the acceptance that national initiatives, such as 598 EMR deployment, can be more successful when driven from "bottom up" process 599 whereby local solutions are integrated in a modular fashion (Aanestad and Jensen 600 2011). A key enabler to this modular approach to innovation is the adoption of shared 603 interoperability standards. Without these standards, we run the risk of creating a 604 complex ecosystem of technologies that are incapable of communicating with each 605 other. Ophthalmology is particularly retrograde on this, with most devices using 606 vendor-specific file formats. Vendor-neutral approaches will improve the ability of AI 607 algorithms, for example, to work on a common data substrate. These standards have 608 long been suggested, but we are now beginning to see concerted effort towards their 609 adoption, for example SMART-on-FHIR, a standards-based interoperable apps 610 platform for EHR (Mandel et screening programmes, whilst also utilising the data generated during the screening 755 process to aide in the further development of existing and new algorithms. Figure 4 756 demonstrates the electronic systems that are already in place to streamline the 757 management of a patient's journey, with virtual integration of each step of their 758 journey from registration to EHR to management of images. Myriad DL programmes 759 are being developed for DR diagnosis, with several models evolving into clinical 760 adoption. training datasets and diagnostic performance for optic disc pathology using OCT. 1128 The widespread availability of such an algorithm could extend the utility of fundus 1129 images acquired in non-ophthalmic centres. Compared to optic disc images, VF data are characterized by low dimensionality and 1132 high noise, and such datasets could be refined using unsupervised ML algorithms. The two most reported unsupervised algorithms are clustering and component 1134 analysis ( The most intractable problem of treating AMD is the frequent and time-consuming 1207 appointments requiring review, evaluation and possible subsequent intravitreal 1208 injection. Since AMD treatment is determined mainly from the VA and OCT findings, 1209 telemedicine could be as useful as face-to-face office consultation. A meta-analysis 1210 in 2018 suggested that teleophthalmology for AMD is as effective as face-to-face 1211 examination, and potentially increases patient participation in screening ( In 2015, the first prospective randomized study to assess the efficacy of telemedicine 1217 for both in the initial screening and recurrence monitoring of neovascular AMD was 1218 reported in Canada (Li et al. 2015) . Best corrected visual acuity, IOP, color fundus 1219 photography, and macula OCT were incorporated in a "store and forward" 1220 telemedicine model. Those in the telemedicine arm attended a local ophthalmologist 1221 who performed the screening, and the data was stored on a database, which was 1222 then reviewed electronically by a retina specialist. In those referred for initial 1223 screening of neovascular AMD, there was no statistically significant difference in 1224 patient waiting times to further diagnostic tests and to treatment. There was also no 1225 significant difference in patient satisfaction except for parking issues. In those 1226 monitored for recurrence, there was no significant difference in the visual outcome 1227 between groups (20/184.8 vs. 20/180.7, p=0.99). This "store and forward" model still utilizes an ophthalmologist as the initial screener. While a technician can be for initial data acquisition used for screening, telemedicine 1231 can be applied further so that initial screening and subsequent monitoring can be 1232 remote, out of the clinical setting and into the home. Home monitoring and self-care have taken centre stage in modern medicine. 1237 Remote in-home monitoring is currently practiced to monitor acute and chronic 1238 diseases such as body temperature to assess a upper respiratory infection, blood 1239 pressure for hypertension (Noah et The Alleye TM application ( Figure 5) , which similarly tests hyperacuity, but examines 1311 a larger area of the macula (12 degrees compared to 3 degrees of field) has 1312 demonstrated its ability to detect neovascular AMD and discriminately classify 1313 between dry and wet disease ( prove to be important during pandemic as well as in the future to limit in-person visit 1325 only when needed. 1326 1327 The use of telemedicine for AMD in the United States has centered on AMD 1330 screening and remote-monitoring systems with some utilising artificial intelligence 1331 applications but as yet there are no large-scale programs for either screening or 1332 monitoring of AMD (Brady and Garg 2020). There are unique challenges to the 1333 screening and monitoring of AMD with lack of consensus on the suitability of the 1334 disease for population screening, and the need for OCTs rather than simple fundus 1335 photographs as used in DR screening and AI algorithms (Brady and Garg 2020). The 1336 Mayo the virtual clinics accounted for approximately 40% of AMD service appointments. With the introduction of the virtual clinics, patients were followed up with a mean of 1359 5.3 weeks compared to 6.9 weeks in the period of conventional clinics. Refractive error is a key public health concern with more than 650 million people 1388 suffering from insufficient or no refractive correction globally (Global Burden of 1389 Disease Study 2015) with the incidence of myopia increasing and poised to escalate 1390 further with urbanization and higher literacy rates (Pan, Ramamurthy, and Saw 2012). Adding to this, the optometrist to population ratio is 1:10,000 in high-income 1395 countries and 1:600,000 in low and middle-income countries (Di Stefano 2001). To evaluate refractive error, traditional visual acuity examination is time-consuming, 1398 and requires the availability of equipment, and examiners skilled in the art of 1399 prescribing spectacles. The procedure is also challenging for people with difficulty in 1400 expressing themselves, such as young children, the elderly, and patients with verbal costly investments for its equipment as well as the hiring of experienced examiners. 1406 Consequently, economic implications due to incorrect dispensing remain high even 1407 in developed countries (Vitale et al. 2006) . Providing good quality refraction services 1408 acceptable to the general population is greatly needed. While myopia alone increases the risk posterior segment complications, these risks 1411 are notably increased in pathologic myopia (PM) when potentially blinding posterior 1412 segment pathological changes appear as a result of the globe elongation 1413 (Grossniklaus and Green 1992 The focus of tele-myopia has been on to prediction of refractive error from easily 1426 obtainable and consistent methods proven in other disease; namely, using the 1427 acquisition of fundus photographs. To be able to accurate define refractive error to 1428 enable a prescription that is acceptable to the patient would be a significant leap 1429 forward in solving the burden of refractive error. Several advanced techniques that assess refractive error accurately have been 1432 developed, and Patients were found to be sufficiently motivated to report their symptoms at least 1598 once a month with a good correlation between the two dry eye questionnaires 1599 (r=0.67), underscoring the potential utility of a tele-health approach for monitoring telemedicine presents different challenges in comparison to screening. Screening is 1620 repetitive and elective, and the process can be planned with clarity for the input, 1621 processing and outputs. For diagnosis, on the other hand, a telemedicine diagnostic 1622 service must consider a much wider variety of conditions and include more abnormal 1623 conditions. In addition, it is more challenging to streamline and process input data in 1624 manner that achieves high diagnostic accuracy. Achieving such accuracy requires 1625 highly trained personnel. clinical agreement between the clinical and e-diagnosis, high (97%) patient 1635 satisfaction, and 37% reduction of unnecessary referral to the hospital eye services. 1636 Moreover, the referrals (with digital images if necessary) were processed within 24 1637 hours, enabling a timely triage and management of any urgent and sight-threatening 1638 diseases. When this programme went live throughout southeast Scotland, the 1639 referral-to-consultation waiting time was reduced from 14 weeks to 4 weeks. The 1640 foundation of this integration project enabled the safe delivery of eye care services 1641 during the COVID-19 pandemic with many primary and urgent eye care services 1642 enabling non-hospital patient care (NHS Scotland 2020). A cloud-based referral system in the UK has demonstrated that more than half of 1645 referrals for possible retinal pathologies to hospital eye services from optometrists 1646 could be avoided with a consultant ophthalmologist reviewing fundus photographs of 1647 the referred patients (Kern et al. 2019 ), similar to the pathway shown in Figure 6 . 1648 Although there are still many factors to be addressed such as safety, economic 1649 benefit, patient satisfaction, and outcomes for those patients who were not referred, 1650 there are notable advantages such as timely patient triage, enhanced provider 1651 correspondence and education. This system enabled the referring doctor to be able 1652 to receive the patient outcome via the platform, allowing each case to be an 1653 educational opportunity. The safety of remote triage in emergency ophthalmology still needs to be 1656 demonstrated. One early study showed that of 500 patients who were triaged 1657 remotely in an emergency unit, 1% had delayed treatment due to misdiagnosis 1658 (Bourdon et al. 2020 ). Prior to widespread adoption of tele-triage, the potential for 1659 harm needs to be more accurately characterised as well as mechanisms put in place 1660 to mitigate the shortcomings of remote reviews. Since in the absence the visual isolation of cases in an effort to stop transmission. With the mitigation approach, the 1677 study found that 8 of 10 people may still be affected, resulting in 510,000 deaths in 1678 the UK and 2.2 million deaths in the US by the end of the pandemic. The study 1679 suggested infected cases could be significantly decreased with a suppression 1680 strategy ("lockdown"), which involved closing schools/universities, case isolation, 1681 household quarantine and social distancing. As country after country began imposing "lockdown" measures, including 1684 quarantines and travel bans in an unprecedented scale (Parmet and Sinha 2020). other specialties, telemedicine was employed to follow-up routine patients, and to 1701 triage and manage new patients presenting to ophthalmology departments. Telephone consultations alone could suffice for some patients, but the addition of 1703 video features allows the clinician additional information to more appropriately triage 1704 a patient. Live video information can be particularly useful in specialties such as 1705 oculoplastics (Kang et al. 2020 ) and strabismus, but also in external eye diseases 1706 where corneal infiltrates may be observed. Furthermore, telemedicine allows for non-1707 verbal communication and aids in fostering physician-patient engagement. Effective 1708 triage not only keeps many patients out of the hospital but can also shorten the 1709 patient's journey once they arrive in hospital. A patient with classic symptoms of a 1710 retinal detachment may bypass the emergency department and be referred directly 1711 to a vitreoretinal surgeon. The rapid introduction of telemedicine and teleophthalmology during the pandemic 1714 has moved beyond the traditional model of connecting specialists with patients from 1715 remote and underserved regions. Instead it has the potential to become the new 1716 standard of care, in particular for triaging patients prior to their hospital attendance. 1717 The new telemedicine systems replacing routine care needs evaluation to ensure 1718 patient safety. Governments such as the China and the US have taken steps to facilitate the rapid 1721 upscaling of these services, with the Chinese national health insurance agency 1722 covering virtual consultation fees, and the US Centres for Medicare and Medicaid 1723 Services (CMS) implementing temporary waivers to enable flexibility within the 1724 healthcare system (Webster 2020). The manifold surge in uptake reported by CMS is 1725 staggering: nearly 1.7 million beneficiaries receiving telehealth services in the last 1726 week of April, 2020 compared to around 13,000 beneficiaries a week prior to the 1727 pandemic (Verma 2020). Of the 9 million beneficiaries who used a telehealth service 1728 three months from mid March 2020, 30% were conducted over the telephone 1729 suggesting there is still significant work to be done in terms of telecommunications 1730 network, healthcare facilities and clinicians adopting new applications, and 1731 consideration of patient factors. As countries consider the model of eyecare in the post-COVID-19 "new normal", 1734 there are several key considerations (Table 6) . First, services must allow for 1735 sustainable social distancing measures for protection of patients, staff and the public. 1736 Second, those at high risk of serious morbidity and mortality with COVID-19 should 1737 be facilitated to isolate wherever possible with access to services at home. Third, 1738 plans must be in place for the management of patients who develop eye conditions 1739 concurrently with COVID-19. Fourth, contingency to manage the 'surge' of patients 1740 who have had deferred appointments or presented late as a result of "lockdown". 1741 Fifth, services should have the agility to expand and shut down to essential 1742 provisions responsively in preparation for future peaks of COVID-19, and indeed 1743 other future pandemics. Finally, there should be measures in place to continually 1744 assess the outcomes of these services to ensure quality of care. The COVID-19 pandemic has come at a time when many technologies and the 1747 necessary infrastructure are mature and already established. Much can be achieved 1748 with simple and universally available technologies such as telephones, messaging, 1749 and video-calling, albeit via safer and secure applications. Subsequently, more 1750 sophisticated eye examinations via telemedicine can occur. This pandemic has 1751 significantly altered the landscape of health care delivery and may have permanent 1752 implications. Time is still needed to establish the safety telemedicine on a massive 1753 scale, but the paradigm shift in acceptability to both patients and doctors will be 1754 profound. Aside from the technical and infrastructural challenges, there are concerns 1755 over how patients will respond to such a shift in healthcare delivery, and if the loss of 1756 rapport gained from physical interaction will cause harm. Clinicians are also 1757 discovering that face-to-face healthcare delivery in the post-COVID era has also 1758 changed. Face masks and social distancing result in loss some of the non-verbal 1759 communication, impede the delivery of empathy. Though there is physical distancing 1760 over a video-consultation, patients are able to see their doctor, and both are able to 1761 see the facial expressions of the other. Acceptance in both patients and physicians is 1762 on the increase (Pappot, Taarnhoj, and Pappot 2020; Hao 2020). Even when 1763 teleophthalmology services have been rapidly adopted during the pandemic, 1764 feedback from a prospective study of 66 patients in an oculoplastics service reported 1765 62% preferred the video consultations to face-to-face, and in this group ranging from 1766 18 years to 88 years (mean 50.7 years), 92% would recommend video consultation 1767 to others (Kang et al. 2020 It would not be possible to provide care at pre-COVID-19 levels whilst practicing 1776 social distancing and maintaining a safe environment for patients and staff alike. 1777 New models of care are being and need to continue to be rapidly upscaled to enable 1778 safe delivery of care until an effective vaccine or treatment is found for COVID-19. The overriding principle of safe care in the COVID-19 in ophthalmic practice is 1781 minimizing exposure: mainly by reducing the number and duration of in-person clinic 1782 visits. Assessments, tests, consultations and even pharmacy and interventions need 1783 to be minimised to those essential for safe care. The integration of 1784 teleophthalmology will be fundamental and can be utilised at multiple points of a 1785 patient's eye care journey. Telemedicine can be and already is being adopted for strabismus. Figure 7 provides an example of semi-automated remote triage 1797 workflow for emergency ophthalmology. Non-ophthalmologist health care workers including optometrists, nurses and 1800 technicians should be trained in multiple skills if possible so that a single person may 1801 perform several tasks such as assessment of visual acuity and intraocular pressure, 1802 instead of patients moving through a number of different clinical staff each 1803 performing a specific task. This improves efficiency and limits exposure risk. 1804 Furthermore, integrating second opinion services to primary care and optometry 1805 practices may enable more appropriate referral into specialized eye units. These measures protect patients and health care workers and contribute to the 1808 larger public health measures. Telemedicine also enable ophthalmologists in 1809 isolation to continue to contribute in clinical work and lessen the impact of key staff 1810 shortages. 1811 1812 This current climate provides the perfect ecosystem to reassess care delivery and to 1815 adopt the synergistic and complementary digital technologies discussed above, 1816 incorporating teleophthalmology and AI utilising and facilitated by 5G networks, IoT 1817 and Big Data analysis. There is widespread media interest and raising of public 1818 awareness of the role telemedicine has already started to play in risk mitigation 1819 during the pandemic. 1820 1821 The emergency department may be a good candidate for widespread introduction of 1822 virtual triage prior to attending in person. The patient benefits as they may discover 1823 they do not need to attend in person, and can be treated with medicines prescribed 1824 remotely. If they do need to attend, their in hospital journal may be much more 1825 efficiently managed, being seen directly by the specialists if appropriate. Additionally, 1826 with the maturation of chatbots, much of the patient counselling can be done 1827 seamlessly from the video consultation. The healthcare providers too reap the benefits of reduced in person attendance, 1830 costs associated with additional time and space utilisation, as well as use of personal 1831 protective equipment at a time where sustainability must also always be considered. 1832 Staff who are able to work from home can contribute, facilitating efficient use of 1833 human resources. Reduced attendances also reduces the general workforce risk of 1834 COVID-19, avoiding the highly undesirable scenario of transmission between 1835 clinicians and patients. Safety of such systems, the remote triaging and automated counselling need to be 1838 evaluated, and until then, clinicians need to oversee each consultation as is standard 1839 process prior to the pandemic. 1840 1841 The figure below demonstrates how a virtual video-based triaging system, with semi-1842 automated features such as registration and counselling, might work. When patients 1843 register, there can be early algorithmic assessment of their presenting complaint. 1844 Symptoms such as flashing lights and floaters, new binocular double vision and new 1845 anisocoria will invariably require in-person examination, and as such can be directed 1846 early to a physical appointment without the patient waiting for a full virtual 1847 assessment first. Patients who do not necessarily require clinician input, such as 1848 mild dry eyes or chalazia, or followup patients who have seen resolution of their 1849 symptoms, for example treated pre-septal cellulitis or contact-lens related keratitis, 1850 can be directed to a chatbot or video for discussion. The remaining patients will be 1851 connected to a clinician when can proceed with a full history and basic examination 1852 which may involve visual acuity assessment using web-based tools. For conditions 1853 that may be managed remotely, such as early pre-septal cellulitis, mild recurrent 1854 anterior uveitis or indeed early non-vision involving contact lens associated keratitis, 1855 medication can be prescribed and sent to the patient via a dedicated delivery service 1856 or local phamarcy. If necessary, plans can be made for the patient to attend in 1857 person for review. Digital transformation through the adoption of teleophthalmology and AI is more than 1869 simply buying new software and hardware, and the next section explores some of 1870 the key challenges to be overcome. Real-world validation has proven to be challenging. The size and heterogenous 1877 nature of the digital health sector with its constant and rapid evolution has created a 1878 complex environment for physicians, healthcare providers, patients and regulatory 1879 bodies in assess these tools to address unmet clinical needs (Mathews et al. 2019) . 1880 There is a need for a rigorous and transparent validation framework, which has some 1881 flexibility in being applied to a broad range of technological innovations. One 1882 proposed framework suggests evaluation based on technical and clinical 1883 considerations, usability, and cost (Mathews et al. 2019 ). Technical evaluation is the most obvious, and is the first step to validation. This is 1886 the fundamental aspect of the technology, and should address if the technology 1887 performs its purported function, its accuracy and robustness. For example, does a 1888 video consultation platform enable patients to register to a virtual waiting room and 1889 be connected to the appropriate clinicians in a safe and effective manner, with due 1890 consideration for data protection. Clinical validation approaches should reflect those that are well established in clinical 1893 research, but can be tailored for digital technologies. Such studies are still 1894 uncommon and may be at least in part due to the lack of clinical experts 1895 simultaneously engaged with technological advances (Hatef, Sharfstein, and 1896 Labrique 2018) The cost of prospective clinical trials as a comparison to existing gold 1897 standards may be off-putting for some in the technology sector who seek rapid 1898 product cycles and returns. Usability, and also accessibility, and the intended user of the technology must be 1901 assessed. Clinicians may need new skills in order to effectively use the tools. The 1902 effectiveness of their use by patients unsupervised should be assessed, as well as 1903 consideration of those who face barriers in adopting the technologies. Cost, and cost effectiveness, as well as the longer term costs should be estimated. 1906 Costs may be obvious, such as purchasing the rights to an algorithm, or hidden, 1907 such as increased referrals seen through telemedicine screening services. 1908 Implications for all stakeholders needs to be considered, from the patient to clinician, 1909 to funding bodies as well as the state. Regulatory bodies attempt to provide guidance for users and payers. workflow disruption and security and privacy concerns (Ajami and Bagheri-Tadi 1986 2013). Some of these issues might be potentially overcome with education and 1987 training of the end-users and provision of financial incentives by the government for 1988 meaningful use of EHR system (Patel et al. 2013 (Patel et al. ). 1989 (Patel et al. 1990 After validating the technological and clinical performance, cost-effectiveness 1991 represents the next hurdle to be overcome before the implementation of a specific 1992 tele-health programme. A notable example was reported in the UK where a large 1993 randomised controlled trial in England evaluating the cost-effectiveness of tele-health 1994 intervention for long-term conditions (including heart failure, chronic obstructive 1995 pulmonary disease, and diabetes) demonstrated no additional benefit when 1996 compared to standard care (Henderson et al. 2013) . That said, tele-ophthalmology 1997 intervention, particularly for DR screening, has proven to be a cost-effective 1998 approach and is already being implemented in many countries, including the US, UK, 1999 and Singapore, at nationwide levels (Kirkizlar et Orbis International uses a free online ophthalmic telemedicine program partnering 2025 doctors in developing countries with expert mentors internationally (Prakalapakorn, 2026 Smallwood, and Helveston 2012). In a survey of this offering, they reported e-2027 DL algorithm uses the "black box" approach where clinical features that confirm a 2071 diagnosis are not apparent. To underscore the reasons prompting a specific 2072 diagnosis by algorithms would be highly beneficial as it allows for clinicians to 2073 understand assess if the correct features were identified, and to offer new insight into 2074 diseases not previously known. This lack of explainability is a hurdle both for clinician 2075 and patient trust. It is challenging when there is disagreement between the algorithm 2076 and the patient and root cause analysis stops short. It is not possible to know if there 2077 is an inherent error in the algorithm that might be corrected. Processes need to be in 2078 place such disagreements, such as an independent third party of a multi-disciplinary 2079 team meeting as would occur where there is clinical uncertainty. There needs to be recognition though, that AI may be proven to be more accurate 2082 than a physician, and detect features humans cannot, as demonstrated by an 2083 algorithm being able to identify sex from fundus photographs (Poplin et al. 2018 ). Thus it becomes harder to adjudicate between the clinician and AI, when the 2085 adjudicator will invariably be another clinician, in particular if the AI decision making 2086 process is unexplainable. In these cases, it may become unethical not to use AI, 2087 even though we do not fully understand how they work. It is unlikely though, that an 2088 individual algorithm will be able to replace the holistic role of a physician, and 2089 increasingly the role of the physician could evolve the use of AI for specific tasks, 2090 and digest the various outputs to collectively to manage the patient. Education on the use and appraisal of AI systems should be incorporated into 2093 medical school programs, and clinicians already in practice will need training to 2094 facilitate its adoption when the technology reaches maturation for clinical practice. Technically able staff who would not form part of existing human resources will need 2096 to be recruited, and work with clinicians to champion adoption. In cases of poor 2097 image quality, automated processes may be able to enhance those images and 2098 enable their reading by the algorithm. However, those with residual artefacts will 2099 remain ungradable and require referral to a clinician. 2100 2101 Early AI algorithms were tested on images collected in the clinical trials setting with 2102 strict inclusion and exclusion criteria (Burlina et al. 2017 With the rapid advancement in digital technology, including EHR, smartphone and 2219 4G/5G technologies, tele-health is likely to pave the way for assessment and 2220 management in the field of ophthalmology. In order for a comprehensive and robust 2221 teleophthalmology platform to thrive, a well-planned eye care delivery system must 2222 exist that considers the resources that are available in specific regions. In 2018, the AAO Telemedicine Task Force published an information statement 2225 regarding the development and implementation of teleophthalmology, including 2226 validation of a teleophthalmology programme against a reference standard, 2227 requirement and standards of data acquisition and communication devices, 2228 competency and qualification of involved personnel, quality assurance, and data 2229 protection (American Academcy Ophthalmology (AAO) Telemedicine Task Force 2230 2018). In principle, it is recommended that a tele-health programme should be 2231 implemented and integrated with evidence-based clinical practices where traditional 2232 process of care is already established (American Academcy Ophthalmology (AAO) 2233 Telemedicine Task performing the specific given task. The replicability of these frameworks may also 2259 vary from country to country due to cultural differences. Understanding the prevalence of the common ocular diseases at a national public 2262 health level, country-specific, is paramount as it helps policymakers and relevant 2263 stakeholders to maximise the cost-effectiveness of the tele-medicine programmes by 2264 targeting highly prevalent diseases. In addition, common diseases that are 2265 dependent on image-based diagnosis with universally agreed-upon, evidence-based 2266 classifications (e.g. DR, AMD, glaucoma and cataract) should also be prioritised in 2267 the set-up of teleophthalmology programmes. The data derived from tele-health may 2268 also be harnessed to generate big data research and to offer more diverse 2269 information such as patient journey education and disease progression forecasting 2270 (McCall 2020). Aspiring to health equality and protection of vulnerable groups should 2271 be a key consideration in every stage of digital innovation and implementation. 2272 2273 The existing digital technologies are predominantly focussed on diagnosis. AI of the 2274 future can increasingly play a role in the guidance of treatment, such as prediction of 2275 how likely patients are to respond to treatments such as intra-vitreal injections in wet 2276 AMD or DMO. Increasing use of AI in the prediction of refractive outcomes following 2277 cataract surgery can help refine lens selection. For children requiring patching or 2278 those requiring accommodation exercises, digital solutions may be able to help 2279 adherence to treatments, with gamification and introduction of incentives for 2280 compliance, although debate will exist around if such use of technology is desirable 2281 for children. Recently, ML associating perimetric cone sensitivities to local OCT in patients with 2284 retinitis pigmentosa was applied to predict visual function in Lebers congenital 2285 amaurosis (LCA) (Sumaroka et al. 2019 ). Though the training dataset was small, 2286 cone vision improvement potential in some LCA was shown to be predictable. This 2287 may permit individual prediction of likely response to treatments and influence 2288 selection to clinical trials so that those with maximal potential gains are selected. Increasingly, isolated algorithms will integrate data from across modalities, and 2291 across disciplines. The utilisation of multi-modal imaging is important for specific 2292 diagnosis (for e.g., determination of the neovascular AMD subtype, diagnosis of 2293 glaucoma and etc). Multi-modal machine learning can be used to evaluate whether 2294 the predictive or diagnostic power of the AI algorithms will increase with the addition 2295 of more imaging modalities. Additionally, data from history, and other metrics such 2296 as blood pressure HbA1c can be used to increase the predictive power of the 2297 algorithms, and data collected from other specialities such as endocrinology and 2298 rheumatology could contribute. 2299 2300 Multi-modal inputs may be help improve the diagnostic and predictive power of AI 2301 systems, and move closer to simulating the decision-making process of a clinician, 2302 but deployment of such multi-modal algorithms in the real-world setting can be 2303 difficult. If the AI has been trained using the ground truth generated by a multi-modal 2304 imaging and additional biomarkers but during clinical use only a limited data is 2305 collected, then that algorithm may not be applicable. Therefore a balance needs to 2306 be achieved between what is practical for routine clinical use versus a complex 2307 algorithm that incorporates multiple inputs. 2308 2309 AI may also play a role in interpreting genetic diseases, such as those with variable 2310 expressivity and phenotypes. DL has been applied in genomics but still remains in its 2311 infancy. There have been studies that have shown some success with various - challenges exist, such as the lack of explainable AI, balanced datasets representing 2318 both disease and healthy states, and integration of heterogeneous data, which is 2319 akin to some of the challenges presented by multi-modal algorithms discussed 2320 above (Koumakis 2020). Medical schools and medical training programmes also need to adapt and 2323 incorporate understanding of digital innovations into training. Clinicians should learn 2324 to interpret studies on areas such as AI or DL algorithms (Ting, Lee, and Wong 2019) 2325 to know if and when such technologies would be suitable for their practice. Medical 2326 students should also learn to conduct remote consultations, be that video or 2327 telephone based only. Without the patient being physically present, the focus of 2328 consultations changes somewhat with the importance of excluding pathologies that 2329 require in person assessment rather than simply managing the presenting complaint. 2330 Nuanced changes to communication strategies need also to be developed adapted 2331 for virtual consultations, and clinicians need to develop at least some basic 2332 understanding of the technical aspect of each platform to enable simple trouble-2333 shooting for new users. Finally patient attitudes need to be studied whilst recognising 2334 these will evolve, as any reaction to something novel. Education driven by evidence 2335 and not politics or other motivations, communicated effectively to reach a wide 2336 audience will be crucial in influencing patients to make their own considered 2337 decisions. Conclusions 2340 2341 Myriad innovations have created a milieu ripe for telemedicine in ophthalmology to 2342 thrive and COVID-19 has hastened the development and embracement of these 2343 digital technologies. The growing AI and telecommunications technologies can 2344 potentially transform the delivery of the data-rich and image-dependent specialty of 2345 ophthalmology globally. 5G, IoT and AI are starting to be introduced into 2346 ophthalmology, but the potential for reliably linked machines such as OCTs and 2347 fundus cameras and algorithms changing ophthalmic service delivery is significant, 2348 and is likely to become more prevalent as the 5G network coverages grows, 2349 enabling a more mature IoT. These technologies may be able to make key 2350 contributions towards the provision of quality, sustainable eye care to all patients, 2351 and experiences from the pandemic has revealed the utility of telemedicine even in 2352 well-resourced and densely populated. Challenges associated with implementation 2353 of these technologies remain, including validation, patient acceptance, and education 2354 and training of end-users on these technologies. Physicians must continue to adapt 2355 to the changing models of care delivery, and collaborate with broader teams 2356 involving technology experts and data scientists to achieve universal quality and 2357 sustainable ophthalmic services. The dash box refers to automated pathway, which could proceed without an 3540 ophthalmologist reviewing the case and images. Example of 'simple' case: dry AMD diagnosed and recorded but no clinical action 3543 required and clinician oversight not required. Example of 'complex case: macular hole potentially suitable for surgery, with 3546 clinician alerted and further clinical decision to be made. Table 1 . Countries, their national screening strategies and the adoption of tele-screening and artificial intelligence in diabetic retinopathy screening. were well received by users, with 96% of 107 users wishing to continue 2028 its use, and 94% of the 78 not using the system wishing to do so. Whilst success in 2029 terms of patient and physician satisfaction has been demonstrated with this 'store-2030 and-forward From a medicolegal perspective, physician-patient interaction in tele-health is 2040 currently considered the same as face-to-face consultation. Though physicians are 2041 concerned about missing a diagnosis or finding (due to inadequate medical 2042 information or suboptimal image quality), the digital images used could serve as a 2043 powerful objective evidence of the consultation. Another noteworthy aspect is that 2044 laws governing physician-patient interactions are disparate across states and 2045 countries. Having an overarching regulation of telemedicine would expedite the 2046 introduction and implementation of telemedicine in routine healthcare service Regulation of telemedicine is also evolving. The Centres of Medicare and Medicaid CMS) broadened provision of telehealth services as part of the emergency 2055 response to the COVID-19 pandemic to enable provision of care whilst limited 2056 community spread of the virus 5.2 Challenges in clinical deployment of AI 2063 2064 AI has remained largely constrained to the research domain with few examples of 2065 real-world adoption in ophthalmology and healthcare more generally. There are 2066 many contributing factors for this. Whilst there is enormous interest and increasingly 2067 robust evidence for the role of AI in DR screening Building Nation-Wide Information Infrastructures in 2363 Healthcare through Modular Implementation Strategies Using telemedicine to screen for 2366 retinopathy of prematurity Pivotal trial of an 2368 autonomous AI-based diagnostic system for detection of diabetic retinopathy in 2369 primary care offices Automated and Computer-Assisted Detection, Classification, and 2372 Diagnosis of Diabetic Retinopathy Improved Automated Detection of Diabetic Retinopathy on a 2375 Publicly Available Dataset Through Integration of Deep Learning Pivotal trial of an autonomous AI-based diagnostic system for detection of diabetic 2379 retinopathy in primary care offices Home Monitoring of Retinal 2382 Sensitivity on a Tablet Device in Intermediate Age-Related Macular Degeneration Next generation 5G wireless networks: a 2385 comprehensive survey Vision Network Validation Study: pilot image stabilization phase Barriers for Adopting Electronic Health Records 2390 (EHRs) by Physicians' Novel Use of Telemedicine for Corneal Tissue Evaluation in Eye 2393 Banking: Establishing a Standardized Approach for the Remote Evaluation of Donor 2394 Corneas for Transplantation Results of the National Program for the 2398 Prevention of Blindness in Childhood by Retinopathy of Prematurity in Argentina 2399 Optic Cup Segmentation Methodologies for Glaucoma Image Detection: A Survey Telemedicine 2404 for Ophthalmology Information Statement Augmented intelligence in health care A new low-cost, compact, auto-phoropter for refractive assessment in 2408 developing countries Web-based longitudinal remote assessment of dry eye 2410 symptoms L'Examen qualitatif de la fonction maculaire Can 2414 Home Monitoring Allow Earlier Detection of Rapid Visual Field Progression in 2415 Glaucoma? Integrating holistic 2417 and local deep features for glaucoma classification Improved Access and Cycle Time with an 2420 ''In-House'' Patient-Centered Teleglaucoma Program Versus Traditional In-Person 2421 Assessment Detecting Preperimetric Glaucoma 2423 with Standard Automated Perimetry Using a Deep Learning Classifier Integrating diabetic retinopathy screening 2427 within diabetes education services in Australia's diabetes and indigenous primary care 2428 clinics' The 2430 accuracy of accredited glaucoma optometrists in the diagnosis and treatment 2431 recommendation for glaucoma Promising Artificial Intelligence-Machine Learning-Deep 2433 Learning Algorithms in Ophthalmology Validation of near eye tool for refractive 2435 assessment (NETRA) -pilot study Development and Validation of a Smartphone-Based Visual Acuity 2438 Test (Peek Acuity) for Clinical Practice and Community-Based Fieldwork Artificial Intelligence Screening for 2442 Diabetic Retinopathy: the Real-World Emerging Application Representation Learning: A Review and 2444 New Perspectives', Ieee Transactions on Pattern Analysis and Machine Intelligence WHO warns governments 'this is not a drill' as coronavirus 2447 infections near 100,000 worldwide Diagnostic Accuracy of Ophthalmoscopy vs Telemedicine in 2450 Examinations for Retinopathy of Prematurity Recent Developments in Clinical 2452 Terminologies -SNOMED CT, LOINC, and RxNorm' Machine Learning of the Progression of Intermediate 2456 Age-Related Macular Degeneration Based on OCT Imaging Teleconsultation in primary ophthalmic emergencies during the COVID-19 lockdown 2460 in Paris: Experience with 500 patients in Causes of vision loss worldwide, 1990-2010: a systematic 2464 analysis Predicting Glaucomatous 2467 Progression in Glaucoma Suspect Eyes Using Relevance Vector Machine Classifiers 2468 for Combined Structural and Functional Measurements Telemedicine for Retinopathy of 2471 Prematurity Telemedicine for Age-Related Macular Degeneration Improving accuracy for intraocular lens selection in 2475 cataract surgery Tariff set for sweeping change from next year Neural networks to identify glaucoma with 2480 structural and functional measurements Ranibizumab versus verteporfin for 2483 neovascular age-related macular degeneration Automated Diagnosis of 2487 Plus Disease in Retinopathy of Prematurity Using Deep Convolutional Neural 2488 Networks Failing to 2490 plan and planning to fail. Can we predict the future growth of demand on UK Eye 2491 Care Services? Optical coherence tomography machine learning 2494 classifiers for glaucoma detection: a preliminary study Automated Grading of Age-Related Macular Degeneration From Color Fundus 2498 Images Using Deep Convolutional Neural Networks Expert Diagnosis of Plus Disease in Retinopathy of Prematurity From 2505 Computer-Based Image Analysis A telemedicine program for diabetic retinopathy in a Veterans 2509 Joslin Vision Network Eye Health Care Model MEDICARE TELEMEDICINE HEALTH 2512 CARE PROVIDER FACT SHEET Automated Detection of Glaucoma From Topographic Features of 2516 the Optic Nerve Head in Color Fundus Photographs Comparison of machine learning and traditional classifiers in glaucoma 2520 diagnosis Healthy China 2030: moving from blueprint to action 2522 with a new focus on public health Characteristics of severe retinopathy of prematurity patients in 2524 China: a repeat of the first epidemic? Gene expression inference 2526 with deep learning 2528 'Peripapillary atrophy detection by sparse biologically inspired feature manifold Artificial 2531 Intelligence in Diabetic Eye Disease Screening EFFECTIVENESS OF DIFFERENT MONITORING MODALITIES IN THE 2535 DETECTION OF NEOVASCULAR AGE-RELATED MACULAR 2536 DEGENERATION: The Home Study Interexpert agreement 2538 of plus disease diagnosis in retinopathy of prematurity Accuracy and reliability of remote 2542 retinopathy of prematurity diagnosis Detection of clinically significant retinopathy of prematurity using 2545 wide-angle digital retinal photography: a report by the American Academy of 2546 Ophthalmology Deep Learning Approaches 2549 Predict Glaucomatous Visual Field Damage from OCT Optic Nerve Head En Face 2550 Images and Retinal Nerve Fiber Layer Thickness Maps Evaluation of the SVOne: A Handheld Virtual clinics in 2554 glaucoma care: face-to-face versus remote decision-making' Diabetic retinopathy in Tanzania: prevalence and risk factors at entry into a 2558 regional screening programme Adoption and spread of innovation in the NHS Virtual glaucoma clinics: patient acceptance and 2562 quality of patient education compared to standard clinics Neighborhood 2570 consortium Multitrait analysis of glaucoma identifies new risk loci and 2574 enables polygenic prediction of disease susceptibility and progression Cryotherapy for Retinopathy of Prematurity: ophthalmological outcomes at 10 years Medicare costs for 2580 neovascular age-related macular degeneration Clinically applicable deep learning for diagnosis and referral 2585 in retinal disease', Nature medicine Low Vision Enhancement with Head-mounted Video 2588 Display Systems: Are We There Yet? Lessons Learned from a Community-Academic Project Using Telemedicine for Eye 2591 Screening Among Urban Latinos Coherence Tomography Images of the Optic Nerve Head Wide-field digital 2597 retinal imaging versus binocular indirect ophthalmoscopy for retinopathy of 2598 prematurity screening: a two-observer prospective, randomised comparison VISION 2020: the right to sight. A global initiative for the 2601 elimination of avoidable blindness Progress in examining cost-effectiveness of AI in diabetic retinopathy 2603 screening', The Lancet Digital Health Can We Stop the Current 2605 Epidemic of Blindness From Retinopathy of Prematurity? Revised 2608 indications for the treatment of retinopathy of prematurity: results of the early 2609 treatment for retinopathy of prematurity randomized trial HEADS-UP SURGERY FOR VITREORETINAL 2612 PROCEDURES: An Experimental and Clinical Study 2614 'Dermatologist-level classification of skin cancer with deep neural networks Impact of non-pharmaceutical interventions (NPIs) to reduce COVID19 2618 mortality and healthcare demand Stanford University Network for Diagnosis of Retinopathy 2623 of Prematurity (SUNDROP): five years of screening with telemedicine Concerns about exercise are related 2627 to walk test results in pulmonary rehabilitation for patients with COPD Global causes of blindness and distance vision impairment 1990-2020: a 2633 systematic review and meta-analysis Screening for primary 2635 open-angle, glaucoma in the primary care setting: An update for the US Preventive 2636 Services Task Force Surveillance of sight loss due to delay in ophthalmic 2638 treatment or review: frequency, cause and outcome Global cost of correcting vision impairment from uncorrected 2641 refractive error Prevalence of age-related macular degeneration in the United States Quebec population and telehealth: a 2647 survey on knowledge and perceptions Comparison of a Novel Cell Phone-Based Refraction 2650 with Subjective Refraction', Investigative Ophthalmology & 2651 Visual Science Telemedicine for 2653 Glaucoma: Guidelines and Recommendations Automated identification of diabetic retinopathy 2655 using deep learning Potential Biases in 2657 Machine Learning Algorithms Using Electronic Health Record Data Diabetic retinopathy and age-related macular degeneration in the Self-management education and regular 2663 practitioner review for adults with asthma Retinopathy of prematurity: a global perspective of the epidemics, population 2667 of babies at risk and implications for control Global, regional, and national 2669 incidence, prevalence, and years lived with disability for 301 acute and chronic 2670 diseases and injuries in 188 countries, 1990-2013: a systematic analysis for the Global 2671 Burden of Disease Study Disease Detection in Diabetic Retinopathy Screening Using Two-Field Photography Comparing machine learning classifiers for diagnosing glaucoma from standard 2678 automated perimetry Interpretation of automated perimetry for glaucoma by neural 2681 network Smartphone guided wide-field imaging for retinopathy of prematurity 2684 in neonatal intensive care unit -a Smart ROP (SROP) initiative' A Deep Learning Algorithm 2688 for Prediction of Age-Related Eye Disease Study Severity Scale for Age Macular Degeneration from Color Fundus Photography Video consultations for covid-2692 19 Pathologic findings in pathologic myopia 2697 'Randomized trial of a home monitoring system for early detection of choroidal 2698 neovascularization home monitoring of the Eye (HOME) study Development and Validation of a Deep Learning Algorithm for 2703 Detection of Diabetic Retinopathy in Retinal Fundus Photographs Performance of a Deep-Learning Algorithm vs Manual Grading for Detecting 2708 Diabetic Retinopathy in India Acceptability and use of glaucoma virtual clinics in the UK: a national survey of 2711 clinical leads Beneficial Effects of Spatial Remapping for Reading 2714 With Simulated Central Field Loss Practice Patterns in Retinopathy of 2717 Prematurity Treatment for Disease Milder Than Recommended by Guidelines 2720 'A treat and extend regimen using ranibizumab for neovascular age-related macular 2721 degeneration clinical and economic impact Diabetes mellitus medication use and 2723 catastrophic healthcare expenditure among adults aged 50+ years in China and India: 2724 results from the WHO study on global AGEing and adult health (SAGE) Automatic extraction of retinal 2727 features from colour retinal images for glaucoma diagnosis: A review Doctors are using AI to triage covid-19 paitents. The tools may be here to 2730 stay Reducing Blindness from Retinopathy of 2736 ROP) in Argentina Through Collaboration Telemedicine Glaucoma Detection and Follow-up Study: Methods and Screening 2743 Results Improving Access to Eye Care among Persons at High-Risk of Glaucoma 2747 in Philadelphia -Design and Methodology: The Philadelphia Glaucoma Detection and 2748 Treatment Project Innovation and Entrepreneurship: 2750 Harnessing the Public Health Skill Set in a New Era of Health Reforms and 2751 Investment 2754 'Long-term pattern of progression of myopic maculopathy: a natural history study Retinopathy of prematurity Cost effectiveness of telehealth for patients with 2761 long term conditions (Whole Systems Demonstrator telehealth questionnaire study): 2762 nested economic evaluation in a pragmatic Classifying visual field data Shedding Light on Telemedicine & Online Prescribing: The Need to 2766 Balance Access to Health Care and Quality of Care Smartphone use in ophthalmology: 2768 What is their place in clinical practice? Virtually Perfect? Telemedicine for Covid-19', N Engl 2770 2020. 2772 'Telemedicine During the COVID-19 Pandemic: Experiences From Western China Practice Guidelines for Ocular Telehealth-2777 Diabetic Retinopathy, Third Edition' Lawn, and Group Born Too Soon Preterm 2779 Birth Action Rule extraction for glaucoma detection with 2782 summary data from StratusOCT Diode laser photocoagulation for threshold 2784 retinopathy of prematurity. A randomized study Characteristics and Risk Factors Associated With 2788 Diagnosed and Undiagnosed Symptomatic Dry Eye Using a Smartphone Application International Committee for the Classification of Retinopathy of, Prematurity International Council of Ophthalmology Cost analysis of remote telemedicine 2795 screening for retinopathy of prematurity An adaptive threshold based image 2797 processing technique for improved glaucoma detection and classification Systematic Review of Current Devices for 24-h Intraocular Pressure Monitoring Standards of medical care for 2805 type 2 diabetes in China Telemedicine screening of retinal 2807 diseases with a handheld portable non-mydriatic fundus camera A Pilot Study to 2810 Improve Access to Eye Care Services for Patients in Rural India by Implementing 2811 Real-time teleophthalmology in 2814 rural Western Australia Diabetic retinopathy in China: population-2818 based studies and clinical and experimental investigations Feasibility of a novel remote daily monitoring system for age-related macular 2821 degeneration using mobile handheld devices: results of a pilot study 2824 'Metamorphopsia Score and Central Visual Field Outcomes in Diabetic Cystoid 2825 Consortium Informatics in Retinopathy of Prematurity Research. 2016. 2830 'Plus Disease in Retinopathy of Prematurity: Improving Diagnosis by Ranking Disease Severity and Using Quantitative Image Analysis Uddin. 2020. 2834 'Oculoplastic video-based telemedicine consultations: Covid-19 and beyond Self-management in patients with 2839 COPD: theoretical context, content, outcomes, and integration into clinical care Collaborative care 2843 and teleglaucoma: a novel approach to delivering glaucoma services in Northern 2844 Tele-Ophthalmology for Age-Related Macular 2847 Degeneration and Diabetic Retinopathy Screening: A Systematic Review and Meta-2848 Analysis Development and validation of a 2851 deep-learning algorithm for the detection of neovascular age-related macular 2852 degeneration from colour fundus photographs Covid-19 and Health Care's Digital 2854 Revolution Retinopathy of prematurity care: 2856 patterns of care and workforce analysis 2862 'Identifying Medical Diagnoses and Treatable Diseases by Image-Based Deep 2863 Learning Implementation of a cloud-based referral platform 2866 in ophthalmology: making telemedicine services a reality in eye care Improving patient access to prevent sight 2869 loss: ophthalmic electronic referrals and communication (Scotland)', Public Health Preventing 2872 diabetes blindness: cost effectiveness of a screening programme using digital non mydriatic fundus photography for diabetic retinopathy in a primary health care setting 2874 in South Africa Diabetic 2876 retinopathy--results of a two year screening programme in two medical units in 2877 Singapore 2879 'Evaluation of telemedicine for screening of diabetic retinopathy in the Veterans 2880 Health Administration The efficacy of a nurse-led 2882 preoperative cataract assessment and postoperative care clinic 2885 'Fifteen-year cumulative incidence of age-related macular degeneration: the Beaver 2886 Dam Eye Study Experiences with developing 2888 and implementing a virtual clinic for glaucoma care in an NHS setting A technician-delivered 'virtual clinic' for 2891 triaging low-risk glaucoma referrals Deep learning models in genomics; are we there yet? Grader variability and the importance 2896 of reference standards for evaluating machine learning models for diabetic 2897 retinopathy Glaucoma screening: analysis of conventional and telemedicine-2900 friendly devices Telemedicine-friendly, portable tonometers: an 2903 evaluation for intraocular pressure screening Tele-ophthalmology and conventional ophthalmology using in remote 2909 Greece a mobile medical unit A systematic review of 2911 teleophthalmological studies in Europe Deep Learning at Chest Radiography: Automated 2913 Classification of Pulmonary Tuberculosis by Using Convolutional Neural Networks Laser therapy for retinopathy of prematurity. Laser ROP 2916 Study Group Deep learning' Deep learning is effective for 2919 classifying normal versus age-related macular degeneration OCT images A multicenter, retrospective pilot study of resource use and costs 2925 associated with severity of disease in glaucoma Media Health Literacy, eHealth Literacy, and the 2928 Role of the Social Environment in Context Prevalence of diabetic retinopathy and visual 2931 impairment in patients with diabetes mellitus in Zambia through the implementation 2932 of a mobile diabetic retinopathy screening project in the Copperbelt province: a cross-2933 sectional study Prospective evaluation of 2935 teleophthalmology in screening and recurrence monitoring of neovascular age-related 2936 macular degeneration: a randomized clinical trial Automatic differentiation of 2939 Glaucoma visual field from non-glaucoma visual filed using deep convolutional 2940 neural network' Development and Evaluation of a Deep Learning System 2942 for Screening Retinal Hemorrhage Based on Ultra-Widefield Fundus Images', 2943 Translational vision science and technology Deep learning for detecting retinal detachment and discerning macular status 2947 using ultra-widefield fundus images A deep learning 2950 system for identifying lattice degeneration and retinal breaks using ultra-widefield 2951 fundus images Efficacy of a Deep Learning 2953 System for Detecting Glaucomatous Optic Neuropathy Based on Color Fundus 2954 Photographs An Automated Grading System for Detection 2957 of Vision-Threatening Referable Diabetic Retinopathy on the Basis of Color Fundus 2958 Photographs Prevalence of primary 2961 open angle glaucoma in a rural adult Chinese population: the Handan eye study Association 2964 Between a Centrally Reimbursed Fee Schedule Policy and Access to Cataract Surgery 2965 in the Universal Coverage Scheme in Thailand Neural networks to identify 2967 glaucomatous visual field progression 2973 'Prediction of myopia development among Chinese school-aged children using 2974 refraction data from electronic medical records: A retrospective, multicentre machine 2975 learning study Glaucomatous Optic Neuropathy Using Fundus Photographs Replacing the Amsler grid: a new method for 2985 monitoring patients with age-related macular degeneration A 2988 novel smartphone ophthalmic imaging adapter: User feasibility studies in Hyderabad A Novel Tele-2991 Eye Protocol for Ocular Disease Detection and Access to Eye Care Services', 2992 Telemedicine and E-Health 2994 'Novel telemedicine device for diagnosis of corneal abrasions and ulcers in resource-2995 poor settings SMART on 2997 FHIR: a standards-based, interoperable apps platform for electronic health records Deep-3000 learning Classifier With an Ultrawide-field Scanning Laser Ophthalmoscope Detects 3001 Glaucoma Visual Field Severity Deep-learning classifier with an ultrawide-field scanning 3004 laser ophthalmoscope detects glaucoma visual field severity 15 ways Silicon Valley is harnessing Big Data for health A proposal for the Dartmouth 3011 summer research project on artificial intelligence Laser 3014 photocoagulation for stage 3+ retinopathy of prematurity From Machine to Machine: An 3016 OCT-Trained Deep Learning Algorithm for Objective Glaucomatous Damage in Fundus Photographs 3019 'Current and Next Generation Portable Screening Devices for Diabetic Retinopathy Glaucoma Screening 3023 in Nepal: Cup-to-Disc Estimate With Standard Mydriatic Fundus Camera Compared 3024 to Portable Nonmydriatic Camera Wilt, and 3028 US Preventive Serv Task Force Cardiovascular Disease Risk Assessment With the Ankle-Brachial Index in Adults: 3030 U.S. Preventive Services Task Force Recommendation Statement Coordination of diabetic retinopathy screening in the 3033 Kimberley region of Western Australia' Hybrid Deep Learning on Single Wide-field Optical 3036 Coherence tomography Scans Accurately Classifies Glaucoma Suspects 3039 'Stanford University Network for Diagnosis of Retinopathy of Prematurity 3040 (SUNDROP): 24-month experience with telemedicine screening Discrimination By Artificial 3043 Intelligence In A Commercial Electronic Health Record-A Case Study ULTRA-WIDEFIELD 3046 FUNDUS IMAGING: A Review of Clinical Applications and Future Trends The National 3050 Academies Collection: Reports funded by National Institutes of Health The Transformational Impact of 5G: Proceedings of a Workshop-in 3052 Brief Copyright 2019 by the National Academy of Sciences. All rights reserved Cost-effectiveness of a National Telemedicine Diabetic Retinopathy Screening 3058 Program in Singapore Impact of remote patient 3063 monitoring on clinical outcomes: an updated meta-analysis of randomized controlled 3064 trials 5G Bytes: Small Cells Explained', 3066 IEEE Spectrum Associations 3069 between metamorphopsia and foveal microstructure in patients with epiretinal 3070 membrane Turning the tide of corneal blindness A national 3074 telemedicine network for retinopathy of prematurity screening Eye 3077 Care Quality and Accessibility Improvement in the Community (EQUALITY) for 3078 adults at risk for glaucoma: study rationale and design Clinical Study on the Initial Experiences of French Vitreoretinal 3081 Surgeons with Heads-up Surgery Worldwide prevalence and risk factors 3083 for myopia Telemedicine and e-Health Solutions for 3085 COVID-19: Patients' Perspective' Covid-19 -The Law and Limits of Quarantine Ophthalmologists Are 3089 More Than Eye Doctors-In Memoriam Li Wenliang Global estimates of visual impairment: 2010' Smartphone-based fundus 3094 photography for screening of plus-disease retinopathy of prematurity Variation in 3097 electronic health record adoption and readiness for meaningful use The efficacy of automated "disease/no 3101 disease" grading for diabetic retinopathy in a systematic screening programme The Past, Present, and Future of Virtual And 3104 Augmented Reality Researcb: A network and cluster analysis of the literature Costs of a community-based glaucoma detection 3108 programme: analysis of the Philadelphia Glaucoma Detection and Treatment Project Prediction of cardiovascular risk factors from retinal fundus 3112 photographs via deep learning ORBIS telemedicine 3114 users NHS Diabetic Eye Screening Programme. Summary statistics: 3116 England DanQ: a hybrid convolutional and recurrent deep neural 3118 network for quantifying the function of DNA sequences Validation of Smartphone Based Retinal Photography for Diabetic 3121 Retinopathy Screening Ensuring 3123 Fairness in Machine Learning to Advance Health Equity Artificial intelligence deep learning algorithm for 3127 discriminating ungradable optical coherence tomography three-dimensional 3128 volumetric optic disc scans Integrated 3130 model of primary and secondary eye care for underserved rural areas: the L V Prasad 3131 Eye Institute experience 5G Networks Will Inherit Their Predecessors' Security Issues The Current State of 3136 Teleophthalmology in the United States Deep learning 3143 versus human graders for classifying diabetic retinopathy severity in a nationwide 3144 screening program Evaluation of a deep 3148 learning image assessment system for detecting severe retinopathy of prematurity Randomized, double-masked, sham-controlled trial of ranibizumab for 3152 neovascular age-related macular degeneration: PIER Study year 1' Balaji Anitha, Raj Deepa, Rajendra Pradeepa Prevalence of Diabetic Retinopathy in Urban India: The 3156 Chennai Urban Rural Epidemiology Study (CURES) Eye Study, I', Investigative 3157 Ophthalmology & Visual Science Glaucoma 3159 Patient Knowledge, Perceptions, and Predispositions for Telemedicine 3162 'Speed of telemedicine vs ophthalmoscopy for retinopathy of prematurity diagnosis The rise and fall of England's National 3165 Programme for IT Ranibizumab for neovascular age-related macular 3168 degeneration 3170 'Screening for retinopathy of prematurity employing the retcam 120: sensitivity and 3171 specificity Deep learning versus human graders for classifying diabetic retinopathy 3174 severity in a nationwide screening program The Relationship 3177 between Variability and Sensitivity in Large-Scale Longitudinal Visual Field Data Global and regional diabetes prevalence 3182 estimates for A Deep Learning 3189 Framework for Predicting Response to Therapy in Cancer Virtual Ophthalmology: 3191 Telemedicine in a Covid-19 Era The Role of Teleophthalmology in the Management 3193 of Diabetic Retinopathy Unsupervised machine learning with independent component analysis to 3197 identify areas of progression in glaucomatous visual fields The Glaucoma Book: a Practical, Evidence-Based 3200 Approach to Patient Care 6G The next Hyper-Connected Experience for All Some studies in machine learning using the game of checkers 3204 (Reprinted from The English national screening programme for sight-threatening 3207 diabetic retinopathy Fully 3212 Automated Detection and Quantification of Macular Fluid in OCT Using Deep 3213 Learning Accuracy of a Self-3215 monitoring Test for Identification and Monitoring of Age-related Macular 3216 Degeneration: A Diagnostic Case-control Study Reliability and diagnostic performance of a novel mobile app for hyperacuity 3219 self-monitoring in patients with age-related macular degeneration Deep learning in neural networks: An overview Machine Learning to Analyze the Prognostic 3225 Value of Current Imaging Biomarkers in Neovascular Age-Related Macular 3226 Degeneration Artificial intelligence in retina Telemedical evaluation 3230 and management of retinopathy of prematurity using a fiberoptic digital fundus 3231 camera Costs and consequences of automated 3235 algorithms versus manual grading for the detection of referable diabetic retinopathy Telemedicine for ROP Development of a deep residual learning algorithm 3241 to screen for glaucoma from fundus photography Why 'Exponential Growth' Is So Scary For The COVID-19 Coronavirus Tackling COVID-19 with Telemedicine 5G Wireless Communication and Health Effects-A 3247 Pragmatic Review Based on Available Studies Regarding 6 to 100 GHz Image processing 3250 based automatic diagnosis of glaucoma using wavelet features of segmented optic 3251 disc from fundus image Evaluation 3254 of telemedicine for slit lamp examination of the eye following cataract surgery Potential lost 3257 productivity resulting from the global burden of uncorrected refractive error Prevalence, risk factors 3260 and burden of diabetic retinopathy in China: a systematic review and meta-analysis 3263 'Prevalence of glaucoma in a rural northern china adult population: a population-3264 based survey in kailu county, inner mongolia Teleophthalmology: improving patient outcomes? Virtually 3268 controlled computerised visual acuity screening in a multilingual Indian population', 3269 Rural and Remote Health Telemedicine in the Management of Exudative Age-Related Macular Degeneration 3275 within an Integrated health care System Global prevalence of vision impairment and blindness: magnitude 3279 and temporal trends Treatment Potential for Macular Cone 3282 Vision in Leber Congenital Amaurosis Due to CEP290 or NPHP5 Mutations: 3283 Predictions From Artificial Intelligence Automatic detection of pathological myopia using variational level 3286 set Artificial intelligence using a deep learning 3290 system with transfer learning to predict refractive error and myopic macular 3291 degeneration from color fundus photographs Strategies to 3296 improve early diagnosis in glaucoma Strategies for improving early 3298 detection of glaucoma: the combined structure-function index Global 3301 Prevalence of Glaucoma and Projections of Glaucoma Burden through 2040 A 3302 Systematic Review and Meta-Analysis The Royal College of Ophthalmologists Telemedical evaluation of 3305 ocular adnexa and anterior segment Diabetic retinopathy: global prevalence, 3307 major risk factors, screening practices and public health challenges: a review Artificial intelligence-assisted 3310 telemedicine platform for cataract screening and management: a potential model of 3311 care for global eye health Artificial intelligence for anterior segment diseases: 3314 Emerging applications in ophthalmology Psychosocial impact of COVID-3316 19 pandemic lockdown on people living with eye diseases in the UK Effect of high-vacuum 3318 setting on phacoemulsification efficiency Artificial 3320 intelligence, the internet of things, and virtual clinics: ophthalmology at the digital 3321 translation forefront Digital technology and COVID-3324 19 Development and 3330 Validation of a Deep Learning System for Diabetic Retinopathy and Related Eye 3331 Diseases Using Retinal Images From Multiethnic Populations With Diabetes Deep learning in estimating 3336 prevalence and systemic risk factors for diabetic retinopathy: a multi-ethnic study An Ophthalmologist's Guide to 3339 Deciphering Studies in Artificial Intelligence Telemedicine and Diabetic 3341 Retinopathy: Review of Published Screening Programs Automated detection of 3343 exudative age-related macular degeneration in spectral domain optical coherence 3344 tomography using deep learning', Graefe's Archive for What is the real gold standard for ROP screening? The concept 3348 of virtual clinics in monitoring patients with age-related macular degeneration An observational study to assess if automated diabetic retinopathy 3353 image assessment software can replace one or more steps of manual imaging grading 3354 and to determine their cost-effectiveness Automated Diabetic Retinopathy Image Assessment 3358 Software: Diagnostic Accuracy and Cost-Effectiveness Compared with Human 3359 Graders The persistent 3361 dilemma of microbial keratitis: Global burden, diagnosis, and antimicrobial 3362 resistance Patient Attitudes Toward Telemedicine for 3365 Diabetic Retinopathy Evaluation 3368 of birth weight and gestational age in infants with treatment requiring retinopathy of 3369 prematurity in ROPE-SOS trial Deep Learning for 3372 Predicting Refractive Error From Retinal Fundus Images Effectiveness of self-management programmes in diabetes management: A 3376 systematic review Early impact of CMS expansion of Medicare telehealth during COVID-19 The changing 3380 scenario of retinopathy of prematurity in middle and low income countries: Unique 3381 solutions for unique problems Costs of refractive correction of 3383 distance vision impairment in the United States The Philadelphia Glaucoma Detection 3387 and Treatment Project: Detection Rates and Initial Management Agreement among 3390 pediatric ophthalmologists in diagnosing plus and pre-plus disease in retinopathy of 3391 prematurity The end of cordon sanitaire in 3393 Wuhan: the role of non-pharmaceutical interventions Prevalence and Ethnic Pattern of 3396 Diabetes and Prediabetes in China in 2013 An Artificial Intelligence Approach to Detect Visual 3400 Field Progression in Glaucoma Based on Spatial Pattern Analysis Virtual health care in the era of COVID-19 The pathophysiology and treatment of 3404 glaucoma: a review The impact of COVID policies on acute ophthalmology 3407 services-experiences from Moorfields Eye Hospital NHS Foundation Trust Aging barriers influencing mobile health 3410 usability for older adults: A literature based framework (MOLD-US)' Non-contact smartphone-based fundus imaging compared to 3414 conventional fundus imaging: a low-cost alternative for retinopathy of prematurity 3415 screening and documentation Validation of an Independent Web-Based Tool for Measuring Visual Acuity 3418 and Refractive Error (the Manifest versus Online Refractive Evaluation Trial): 3419 Prospective Open-Label Noninferiority Clinical Trial Economic Evaluation of a Home-Based Age-Related Macular Degeneration 3422 Monitoring System Saving sight in 3424 China and beyond: the Lifeline Express model The War on Diabetic Retinopathy: Where Are We 3426 Now? Teleophthalmic Approach for Detection of Corneal Diseases: 3429 Accuracy and Reliability Programme for the Prevention of Blindness and Deafness: 3431 Global initiative for the elimination of avoidable blindness WHO Director-General's opening remarks at the media 3435 briefing on COVID-19 -11 Service innovation in glaucoma management: using 3437 a web-based electronic patient record to facilitate virtual specialist supervision of a 3438 shared care glaucoma programme Universal artificial 3443 intelligence platform for collaborative management of cataracts Impact of Different Visual Field Testing Paradigms on 3446 Sample Size Requirements for Glaucoma Clinical Trials 3448 'Content Design and System Implementation of a Disease Diagnosis and Treatment and Its Preliminary Practice in Guangdong Artificial intelligence for teleophthalmology-3452 based diabetic retinopathy screening in a national program: a modelled economic 3453 analysis study Metamorphopsia and vision-related quality 3455 of life among patients with age-related macular degeneration Beijing Eye Public Health Care 3459 Project A new computer-based pediatric vision-screening test Diabetes China 3465 National, and Group Metabolic Disorders Study Global prevalence 3473 and major risk factors of diabetic retinopathy The 3475 optimum time to employ telephotoscreening to detect retinopathy of prematurity Unsupervised Gaussian 3479 Mixture-Model With Expectation Maximization for Detecting Glaucomatous 3480 Progression in Standard Automated Perimetry Visual Fields Detecting glaucomatous change in visual fields: Analysis with an optimization 3485 framework Detection of Longitudinal Visual Field Progression in Glaucoma 3488 Using Machine Learning Perioperative cataract OP management by means of teleconsultation Automatic diagnosis of pathological myopia from heterogeneous biomedical data Optimizing Glaucoma 3497 Screening in High-Risk Population: Design and 1-Year Findings of the Screening to 3498 Prevent (SToP) Glaucoma Study' 2020. 3500 'Detecting glaucoma based on spectral domain optical coherence tomography imaging 3501 of peripapillary retinal nerve fiber layer: a comparison study between hand-crafted 3502 features and deep learning model