You are here

Article Text

Article menu

Download PDFPDF

Ophthalmology
Protocol
Diagnostic performance of deep learning in infectious keratitis: a systematic review and meta-analysis protocol
  1. Zun Zheng Ong1,
  2. Youssef Sadek1,
  3. Xiaoxuan Liu2,
  4. Riaz Qureshi3,
  5. Su-Hsun Liu3,
  6. Tianjing Li3,
  7. Viknesh Sounderajah4,5,
  8. Hutan Ashrafian4,5,
  9. Daniel Shu Wei Ting6,7,
  10. Dalia G Said1,8,9,
  11. Jodhbir S Mehta6,7,
  12. Matthew J Burton10,11,
  13. Harminder Singh Dua1,8,
  14. Darren Shu Jeng Ting2,8,12
  1. 1Department of Ophthalmology, Queen's Medical Centre, Nottingham, UK
  2. 2Academic Unit of Ophthalmology, Institute of Inflammation and Ageing, University of Birmingham, Birmingham, UK
  3. 3Department of Ophthalmology, School of Medicine, University of Colorado Anschutz Medical Campus, Aurora, Colorado, USA
  4. 4Institute of Global Health Innovation, Imperial College London, London, UK
  5. 5Department of Surgery & Cancer, Imperial College London, London, UK
  6. 6Duke-NUS Medical School, National University of Singapore, Singapore
  7. 7Singapore National Eye Centre, Singapore Eye Research Institute, Singapore
  8. 8Academic Ophthalmology, School of Medicine, University of Nottingham, Nottingham, UK
  9. 9Research Institute of Ophthalmology, Cairo, Egypt
  10. 10International Centre for Eye Health, London School of Hygiene and Tropical Medicine, London, UK
  11. 11National Institute for Health Research (NIHR) Biomedical Research Centre (BRC) for Ophthalmology, Moorfields Eye Hospital NHS Foundation Trust and UCL Institute of Ophthalmology, London, UK
  12. 12Birmingham and Midland Eye Centre, Birmingham, UK
  1. Correspondence to Dr Darren Shu Jeng Ting; ting.darren{at}gmail.com

Abstract

Introduction Infectious keratitis (IK) represents the fifth-leading cause of blindness worldwide. A delay in diagnosis is often a major factor in progression to irreversible visual impairment and/or blindness from IK. The diagnostic challenge is further compounded by low microbiological culture yield, long turnaround time, poorly differentiated clinical features and polymicrobial infections. In recent years, deep learning (DL), a subfield of artificial intelligence, has rapidly emerged as a promising tool in assisting automated medical diagnosis, clinical triage and decision-making, and improving workflow efficiency in healthcare services. Recent studies have demonstrated the potential of using DL in assisting the diagnosis of IK, though the accuracy remains to be elucidated. This systematic review and meta-analysis aims to critically examine and compare the performance of various DL models with clinical experts and/or microbiological results (the current ‘gold standard’) in diagnosing IK, with an aim to inform practice on the clinical applicability and deployment of DL-assisted diagnostic models.

Methods and analysis This review will consider studies that included application of any DL models to diagnose patients with suspected IK, encompassing bacterial, fungal, protozoal and/or viral origins. We will search various electronic databases, including EMBASE and MEDLINE, and trial registries. There will be no restriction to the language and publication date. Two independent reviewers will assess the titles, abstracts and full-text articles. Extracted data will include details of each primary studies, including title, year of publication, authors, types of DL models used, populations, sample size, decision threshold and diagnostic performance. We will perform meta-analyses for the included primary studies when there are sufficient similarities in outcome reporting.

Ethics and dissemination No ethical approval is required for this systematic review. We plan to disseminate our findings via presentation/publication in a peer-reviewed journal.

PROSPERO registration number CRD42022348596.

  • Artificial intelligence
  • Corneal infection
  • Diagnosis
  • Keratitis
  • Ophthalmology
https://creativecommons.org/licenses/by/4.0/

This is an open access article distributed in accordance with the Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits others to copy, redistribute, remix, transform and build upon this work for any purpose, provided the original work is properly cited, a link to the licence is given, and indication of whether changes were made. See: https://creativecommons.org/licenses/by/4.0/.

https://doi-org.ezproxy.u-pec.fr/10.1136/bmjopen-2022-065537

Statistics from Altmetric.com

    Request Permissions

    If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

    STRENGTHS AND LIMITATIONS OF THIS STUDY

    • This systematic review and meta-analysis will include all relevant articles related to the diagnostic performance of deep learning in infectious keratitis through a comprehensive search of all literature.

    • This review will examine the diagnostic performance of deep learning in identifying and differentiating the subtypes of infectious keratitis, including bacterial, fungal, viral and parasitic keratitis.

    • This review will examine the diagnostic accuracy of deep learning in distinguishing infectious keratitis from healthy eyes and other ocular surface diseases.

    • Meta-analysis will be performed to determine the pooled diagnostic sensitivity and specificity of deep learning and will be compared with the performance of healthcare professionals (if data are available).

    • The quality of the study will depend on the quality of the available published literature related to this topic.

    Introduction

    Due to worldwide population ageing and urbanisation, it is expected that close to 900 million people will suffer from distance vision impairment, of whom 61 million people will be blind by 2050.1 Infectious keratitis (IK), also commonly known as corneal infection, currently represents the fifth-leading cause of blindness worldwide.2 3 It can be caused by a wide variety of pathogens such as bacteria, fungi, protozoa and viruses.3 4 Once considered a ‘silent epidemic’ in low-income and middle-income countries (LMICs), IK has so far caused ~5 million cases of blindness around the world and is estimated to cause ~2 million monocular blindness each year, placing significant burden on global human health.3 5 A recent meta-analysis conducted by Brown et al6 estimated that the global incidence of fungal keratitis alone (without accounting for other types of IK) is likely >1 million cases per year, primarily affecting the populations in Africa and Asia. Previous studies have also consistently reported a disproportionately higher incidence of IK in the LMICs (113–799 per 100 000 populations-year) than in high-income countries (HICs; 2.5–40.3 per 100 000 populations-year),3 7 8 which was likely attributable to increased risk of trauma from agricultural and other occupational activities, environmental factors, the use of traditional eye medicine (which may contain pathogens) and the limited access to primary and secondary eye care.3 9–11

    Patients affected by IK are often debilitated by severe ocular pain and sight loss, and some are at risk of losing the eye due to intractable infection.6 11–14 The outcome of IK is critically dependent on a timely and accurate diagnosis, followed by appropriate medical and/or surgical interventions. In current clinical practice, IK is usually diagnosed on clinical grounds with support from additional tests, including microbiological investigations (eg, smear microscopy, culture and sensitivity testing, and PCR) and/or corneal imaging (eg, in vivo confocal microscopy (IVCM)).15–17 However, these approaches have multiple challenges, including the need for clinical expertise and equipment, low microbiological culture yield, long turnaround time, poorly differentiated clinical features and polymicrobial infections.7 15 18–20 Moreover, access to such microbiological and imaging investigations is not available in many ophthalmic units in LMICs, leading to a reliance of empirical treatment. This can lead to a misdiagnosis when based on clinical features alone and the use of incorrect antimicrobial therapy (eg, fungal keratitis being treated only with antibacterial agents). This can result in delays in the initiation of effective treatment, with consequent poorer clinical outcomes and higher risk of ocular complications.

    In recent years, the interest of integrating artificial intelligence (AI) into clinical medicine with the hope of improving the quality of healthcare services has been reignited,21 primarily owing to the advancement in deep learning (DL) techniques, improvement in computing power and increased availability of big data.22–26 DL, a subfield of AI, has demonstrated promise in assisting automated medical diagnosis, clinical triage and decision-making, as well as improving the workflow efficiency in healthcare services in both developed and developing countries.23–25 27–30 Within the realm of ophthalmology, DL research previously focused mainly on various posterior segment diseases (eg, age-related macular degeneration, diabetic retinopathy and glaucoma) and demonstrated comparable, if not better, diagnostic accuracy compared with healthcare professionals.23 24 31–33 Although several recent studies have demonstrated the potential of DL in assisting the diagnosis of IK and distinguishing IK from other ocular diseases,34–39 the diagnostic accuracy of these DL models remains to be elucidated.

    To the best of our knowledge, there is no published systematic review and/or meta-analysis specifically evaluating the diagnostic performance of DL in IK. In view of the current diagnostic challenges of IK and the potential of DL in addressing the highlighted limitations, this systematic review and meta-analysis aims to critically examine and compare the performance of various DL models with clinical experts (the current ‘gold standard’) in diagnosing IK, which can help inform the clinical practice on the potential clinical applicability and deployment of this DL model.

    Review questions/objectives

    The proposed systematic review aims to answer the following main questions:

    1. What is the diagnostic accuracy of DL models in detecting and differentiating IK from healthy eyes?

    2. What is the diagnostic performance of DL models in differentiating IK from other types of corneal or ocular diseases?

    3. What is the diagnostic accuracy of DL models in differentiating the types of IK (eg, bacterial keratitis vs fungal keratitis)?

    Methods

    This protocol was produced based on the Preferred Reporting Items for Systematic Reviews and Meta-Analyses Protocols (PRISMA).40 This systematic review will be conducted in accordance with the recommendations of the Cochrane Handbook for Systematic Reviews of Diagnostic Test Accuracy (DTA). We will write the resulting paper following the PRISMA-DTA41 and the CHecklist for critical Appraisal and data extraction for systematic Reviews of prediction Modelling Studies.42

    Eligibility criteria

    This diagnostic accuracy systematic review will consider all relevant clinical studies, including prospective and retrospective comparative cohort studies, case–control studies and cross-sectional studies, which examined the accuracy of DL in diagnosing any types of IK, encompassing bacterial, fungal, Acanthamoeba and/or viral keratitis. We will exclude case reports and reviews. We will only include studies that employed corneal imaging tests, which may include slit-lamp photography, IVCM, anterior segment optical coherence tomography (AS-OCT) and/or corneal topography/tomography. We will exclude AI studies that contained only data without any imaging or those that focused on image segmentation instead of disease classification. There will be no restriction on patients’ age, gender, ethnicity and geographical location. There will be no restriction on the number and proportion of images used for each stage of the DL models, including training, validation and testing stages.

    Information sources and search strategy

    We will search various bibliographic databases, including EMBASE (OVID), MEDLINE (OVID), CINAHL and DANS EASY Archive, and trial registries, including the Cochrane Central Register of Controlled Trials (CENTRAL), ISRCTN registry (www.isrctn.com/), US National Institutes of Health Ongoing Trials Register (https://www.clinicaltrials.gov/) and WHO International Clinical Trials Registry Platform for primary research related to DL for diagnosing IK. The search strategy aims to locate both published and unpublished studies. The search will be developed with two concepts built into the search strategy to capture relevant articles: (1) AI and (2) IK. There will be no restriction on the study design, date or language for the search. The search strategy, including all identified keywords and index terms, will be adapted to each included information source. The reference list of all eligible studies will be manually screened for additional studies. An example of the search strategy is provided in table 1.

    View this table:
    Table 1

    A summary of the search strategy using EMBASE for studies related to artificial intelligence in diagnosing infectious keratitis

    Study selection

    Following the search, all identified citations will be loaded into EndNote V.X9 (Clarivate Analytics, Pennsylvania, USA) and duplicates will be removed. The titles and abstract will be screened by two independent reviewers for assessment against the inclusion criteria of the review, using the Rayyan AI platform (Qatar).43 The full text of selected citations will then be assessed in detail against the inclusion criteria by two independent reviewers. Reasons for exclusion of any full-text studies will be recorded and reported in the systematic review. Any disagreements between reviewers at each stage of study selection will be resolved through discussions or consultations with a third reviewer. The results of the search will be reported in full in the final systematic review and presented in a PRISMA-DTA flow diagram.41

    Data collection and data items

    Data will be extracted from the included articles by two independent reviewers using a standardised and pilot-tested data extraction tool, RevMan V.5.4 (Copenhagen: The Nordic Cochrane Centre, Cochrane). The extracted data will include specific details about the name of authors, study title, year of publication, countries of study, populations (including diseased and healthy cases), demographic factors (ie, age, gender, ethnicity), sample size, study methods, types of DL algorithms, decision threshold, types of reference standard (which may include expert consensus, microbiological results confirmed on either smear microscopy, culture or PCR and/or corneal imaging such as IVCM), and diagnostic accuracy (including the sensitivity and specificity) of the index test (ie, DL algorithms) and the comparator (ie, non-expert healthcare professionals), if available. We will extract sufficient information to build 2×2 contingency tables at the reported threshold for each study. The contingency tables will include true positive, false positive, true negative and false negative to calculate the sensitivity and specificity. If various contingency tables were provided for the same or different algorithms in the same study, they are assumed to be independent from each other as the aim of this work is to provide an overview of the results of various studies instead of the precise point estimates.30 Any disagreements will be resolved by group consensus. Authors of eligible studies will be contacted to request any missing data, where required.

    Outcomes

    The primary outcome will be the diagnostic accuracy of DL algorithms in distinguishing IK from healthy eyes and/or those with other types of corneal diseases, as compared with the reference standard. The diagnostic accuracy of each group will be presented in the form of sensitivity and specificity.44 The secondary outcomes for this review will involve a comparison of the accuracy in differentiating various types of IK and in differentiating IK from other types of corneal or ocular surface diseases. For studies that focused on distinguishing IK (any type) from healthy corneas or those with other non-IK corneal pathologies, the reference standard will be the expert consensus and/or microbiological results. For studies that focused on differentiating the subtypes of organisms (eg, bacteria vs fungi), the microbiological results or expert consensus (if microscopy or culture results were not available) will be the reference standard. Other potential outcomes for this review will include the accuracy of DL in predicting the culture positivity and clinical outcomes of IK based on the initial presenting images. These secondary outcomes may not be feasible if these more specific questions are not ascertained in the included studies, but they are nonetheless of interest.

    Risk of bias assessment

    Eligible studies will be critically appraised by two independent reviewers at the outcome level for methodological quality in the review using the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool.45 QUADAS-AI (an AI-specific extension to QUADAS-2) tool46 will be used if available by the time of the conduct of this systematic review. Specifically, we will assess the Risk of Bias for our primary outcomes (ie, accuracy of DL vs reference standard for IK). The questions used in these tools are split into four domains: patient selection, index test, reference test, and flow and timing. Each of these domain help assess the risk of bias created by patient selection, the conduct and interpretation of index test and reference test and the sequence and timing of the study, respectively. We will also assess whether the AI systems have been tested on an externally validated test set. Authors of papers will be contacted to request for additional data for clarification, where necessary. Any disagreements will be resolved by discussion or by seeking the advice of a third reviewer. All studies, regardless of the results of their methodological quality, will undergo data extraction and synthesis. The results of critical appraisal will be reported in the final systematic review, in both narrative and tabular formats.

    Data synthesis and analysis

    The analysis will be conducted at two levels: (1) a systematic synthesis of all eligible studies and (2) a meta-analysis of all relevant studies with similar outcome reporting. For the meta-analysis, the intervention group (ie, the ‘index test’) will refer to the image-based DL algorithms for diagnosing or differentiating IK from other ocular diseases. The reference group will be the expert consensus and/or microbiological results, also known as the gold standard or ‘ground truth’ for the DL algorithms whereas the comparator group, if available, will be the non-expert healthcare professionals.

    Where possible, we will pool similar measures of accuracy from studies with statistical meta-analysis using RevMan V.5.4 software. ‘Paired’ forest plots, with one forest plot for sensitivity and the other for specificity, will be used and presented side by side.47 The means and 95% CIs of each selected primary studies will be presented alongside the number of true positives, false positives, true negatives and false negatives, wherever appropriate. Summary receiver operating characteristic (SROC) curves will also be plotted using the sensitivity and specificity of each primary study. χ2 or Fisher’s exact tests will be used to assess the heterogeneity objectively, if needed.47 We expect heterogeneity in the types of DL systems and algorithms used across studies and we will consider all to be acceptable ‘interventions’ for analysis as our question is meant to assess the general accuracy of any DL system. In view of the anticipated interstudies heterogeneity, a random-effects model will be used for the meta-analysis to determine the pooled sensitivity and specificity of the included studies. A fixed-effect model may be used if there is no significant heterogeneity. Subgroup analyses will be conducted where there are sufficient data to investigate different types of IK and other ocular surface diseases, as per our prespecified secondary outcomes. Additional subgroup analysis will also be performed on the diagnostic performance of the DL systems based on different imaging modalities, including slit-lamp photography, IVCM, AS-OCT and others.

    The heterogeneity between studies will be assessed through the graphic display of paired forest plots or SROC curves. We will evaluate potential publication bias of the pooled data using Deek’s funnel plot, and a p<0.05 will be considered of significant publication bias.48 49 The Grading of Recommendations, Assessment, Development and Evaluation approach for grading the certainty of evidence will be followed.40 A Summary of Findings (SoF) table, created using GRADEpro software, will be presented. Where appropriate, the following information will be included in the SoF: number and type(s) of studies contributing to the outcome, total sample size contributing to the outcome, ranking of the certainty of the evidence based on the risk of bias, heterogeneity, directness, publication bias and precision of the review results. We will include the following outcomes in the SoF table: AUC, sensitivity and specificity for IK overall (ie, primary outcome).

    Patient and public involvement

    DSJT had previously involved patients who were affected by IK to help identify the research need and priority in relation to IK. Many of the patients with IK have highlighted the importance of timely and accurate diagnosis of IK as the delay in diagnosis has negatively affected their visual outcomes. This serves as one of the key reasons for conducting this systematic review and meta-analysis, which aims to improve the diagnosis of IK in clinical settings.

    Clinical relevance of this systematic review

    The results of this systematic review and meta-analysis will provide high-quality evidence on the diagnostic accuracy of DL in IK. This study will help identify the gaps (if any) in the current clinical evidence, which may be related to study design, quality of the research methodologies, setting of reference standard, risk of bias and outcome reporting. The identification of these issues can help refine the study design of any future clinical trials evaluating the diagnostic accuracy of DL in IK in a real-world setting. These findings will also help inform the clinicians, researchers, policy-makers and regulatory bodies on the clinical applicability of DL in diagnosing IK, with an aim to develop more accessible investigations for IK in the future, including in both HIC and LMICs.

    Ethics statements

    Patient consent for publication

    References

    1. ↵
      1. Burton MJ,
      2. Ramke J,
      3. Marques AP, et al
      . The Lancet global health Commission on global eye health: vision beyond 2020. Lancet Glob Health 2021;9:e489–551. doi:10.1016/S2214-109X(20)30488-5
      OpenUrlCrossRefPubMed
    2. ↵
      1. Flaxman SR,
      2. Bourne RRA,
      3. Resnikoff S, et al
      . Global causes of blindness and distance vision impairment 1990-2020: a systematic review and meta-analysis. Lancet Glob Health 2017;5:e1221–34. doi:10.1016/S2214-109X(17)30393-5
      OpenUrlPubMed
    3. ↵
      1. Ting DSJ,
      2. Ho CS,
      3. Deshmukh R, et al
      . Infectious keratitis: an update on epidemiology, causative microorganisms, risk factors, and antimicrobial resistance. Eye 2021;35:1084–101. doi:10.1038/s41433-020-01339-3
      OpenUrlCrossRefPubMed
    4. ↵
      1. Durand ML,
      2. Barshak MB,
      3. Chodosh J
      . Infectious keratitis in 2021. JAMA 2021;326:1319–20. doi:10.1001/jama.2021.0424
      OpenUrl
    5. ↵
      1. Ung L,
      2. Acharya NR,
      3. Agarwal T, et al
      . Infectious corneal ulceration: a proposal for neglected tropical disease status. Bull World Health Organ 2019;97:854–6. doi:10.2471/BLT.19.232660
      OpenUrl
    6. ↵
      1. Brown L,
      2. Leck AK,
      3. Gichangi M, et al
      . The global incidence and diagnosis of fungal keratitis. Lancet Infect Dis 2021;21:e49–57. doi:10.1016/S1473-3099(20)30448-5
      OpenUrlCrossRefPubMed
    7. ↵
      1. Ting DSJ,
      2. Ho CS,
      3. Cairns J, et al
      . 12-year analysis of incidence, microbiological profiles and in vitro antimicrobial susceptibility of infectious keratitis: the Nottingham infectious keratitis study. Br J Ophthalmol 2021;105:328–33. doi:10.1136/bjophthalmol-2020-316128
      OpenUrlAbstract/FREE Full Text
    8. ↵
      1. Collier SA,
      2. Gronostaj MP,
      3. MacGurn AK, et al
      . Estimated burden of keratitis -- United States, 2010. MMWR Morb Mortal Wkly Rep 2014;63:1027–30.
      OpenUrlPubMed
    9. ↵
      1. Gopinathan U,
      2. Sharma S,
      3. Garg P, et al
      . Review of epidemiological features, microbiological diagnosis and treatment outcome of microbial keratitis: experience of over a decade. Indian J Ophthalmol 2009;57:273–9. doi:10.4103/0301-4738.53051
      OpenUrlCrossRefPubMedWeb of Science
    10. ↵
      1. Chidambaram JD,
      2. Venkatesh Prajna N,
      3. Srikanthi P, et al
      . Epidemiology, risk factors, and clinical outcomes in severe microbial keratitis in South India. Ophthalmic Epidemiol 2018;25:297–305. doi:10.1080/09286586.2018.1454964
      OpenUrlPubMed
    11. ↵
      1. Ting DSJ,
      2. Cairns J,
      3. Gopal BP, et al
      . Risk factors, clinical outcomes, and prognostic factors of bacterial keratitis: the Nottingham infectious keratitis study. Front Med (Lausanne) 2021;8:715118. doi:10.3389/fmed.2021.715118
      OpenUrl
    12. ↵
      1. Ting DSJ,
      2. Galal M,
      3. Kulkarni B, et al
      . Clinical characteristics and outcomes of fungal keratitis in the United Kingdom 2011-2020: a 10-year study. J Fungi (Basel) 2021;7:966. doi:10.3390/jof7110966
    13. ↵
      1. Khor W-B,
      2. Prajna VN,
      3. Garg P, et al
      . The Asia cornea Society infectious keratitis study: a prospective multicenter study of infectious keratitis in Asia. Am J Ophthalmol 2018;195:161–70. doi:10.1016/j.ajo.2018.07.040
      OpenUrlCrossRefPubMed
    14. ↵
      1. Ting DSJ,
      2. Mckenna M,
      3. Sadiq SN, et al
      . Arthrographis kalrae keratitis complicated by endophthalmitis: a case report with literature review. Eye Contact Lens 2020;46:e59–65. doi:10.1097/ICL.0000000000000713
      OpenUrl
    15. ↵
      1. Ting DSJ,
      2. Gopal BP,
      3. Deshmukh R, et al
      . Diagnostic armamentarium of infectious keratitis: a comprehensive review. Ocul Surf 2022;23:27–39. doi:10.1016/j.jtos.2021.11.003
      OpenUrl
    16. ↵
      1. Bakken IM,
      2. Jackson CJ,
      3. Utheim TP, et al
      . The use of in vivo confocal microscopy in fungal keratitis-progress and challenges. Ocul Surf 2022;24:103–18. doi:10.1016/j.jtos.2022.03.002
      OpenUrl
    17. ↵
      1. Ting DSJ,
      2. Said DG,
      3. Dua HS
      . Interface haze after Descemet stripping automated endothelial keratoplasty. JAMA Ophthalmol 2019;137:1201–2. doi:10.1001/jamaophthalmol.2019.2745
      OpenUrl
    18. ↵
      1. Khoo P,
      2. Cabrera-Aguas MP,
      3. Nguyen V, et al
      . Microbial keratitis in Sydney, Australia: risk factors, patient outcomes, and seasonal variation. Graefes Arch Clin Exp Ophthalmol 2020;258:1745–55. doi:10.1007/s00417-020-04681-0
      OpenUrl
    19. ↵
      1. Ting DSJ,
      2. Bignardi G,
      3. Koerner R, et al
      . Polymicrobial keratitis with Cryptococcus curvatus, Candida parapsilosis, and Stenotrophomonas maltophilia after penetrating keratoplasty: a rare case report with literature review. Eye Contact Lens 2019;45:e5–10. doi:10.1097/ICL.0000000000000517
      OpenUrlPubMed
    20. ↵
      1. Ting DSJ,
      2. Chodosh J,
      3. Mehta JS
      . Achieving diagnostic excellence for infectious keratitis: a future roadmap. Frontiers in Microbiology 2022;13:.:1020198. doi:10.3389/fmicb.2022.1020198
    21. ↵
      1. Schwalbe N,
      2. Wahl B
      . Artificial intelligence and the future of global health. Lancet 2020;395:1579–86. doi:10.1016/S0140-6736(20)30226-9
      OpenUrlPubMed
    22. ↵
      1. LeCun Y,
      2. Bengio Y,
      3. Hinton G
      . Deep learning. Nature 2015;521:436–44. doi:10.1038/nature14539
      OpenUrlCrossRefPubMed
    23. ↵
      1. Ting DSW,
      2. Cheung CY-L,
      3. Lim G, et al
      . Development and validation of a deep learning system for diabetic retinopathy and related eye diseases using retinal images from multiethnic populations with diabetes. JAMA 2017;318:2211–23. doi:10.1001/jama.2017.18152
      OpenUrlCrossRefPubMed
    24. ↵
      1. De Fauw J,
      2. Ledsam JR,
      3. Romera-Paredes B, et al
      . Clinically applicable deep learning for diagnosis and referral in retinal disease. Nat Med 2018;24:1342–50. doi:10.1038/s41591-018-0107-6
      OpenUrlPubMed
    25. ↵
      1. Seah JCY,
      2. Tang CHM,
      3. Buchlak QD, et al
      . Effect of a comprehensive deep-learning model on the accuracy of chest X-ray interpretation by radiologists: a retrospective, multireader multicase study. Lancet Digit Health 2021;3:e496–506. doi:10.1016/S2589-7500(21)00106-0
      OpenUrl
    26. ↵
      1. Ting DSJ,
      2. Deshmukh R,
      3. Ting DSW, et al
      . Big data in corneal diseases and cataract: current applications and future directions. Frontiers in Big Data 2023;6:1017420.:1017420. doi:10.3389/fdata.2023.1017420
      OpenUrl
    27. ↵
      1. Soenksen LR,
      2. Kassis T,
      3. Conover ST, et al
      . Using deep learning for dermatologist-level detection of suspicious pigmented skin lesions from wide-field images. Sci Transl Med 2021;13:581. doi:10.1126/scitranslmed.abb3652
      OpenUrl
    28. ↵
      1. Sabanayagam C,
      2. Xu D,
      3. Ting DSW, et al
      . A deep learning algorithm to detect chronic kidney disease from retinal Photographs in community-based populations. Lancet Digit Health 2020;2:e295–302. doi:10.1016/S2589-7500(20)30063-7
      OpenUrlPubMed
    29. ↵
      1. Ting DSJ,
      2. Ang M,
      3. Mehta JS, et al
      . Artificial intelligence-assisted telemedicine platform for cataract screening and management: a potential model of care for global eye health. Br J Ophthalmol 2019;103:1537–8. doi:10.1136/bjophthalmol-2019-315025
      OpenUrlFREE Full Text
    30. ↵
      1. Liu X,
      2. Faes L,
      3. Kale AU, et al
      . A comparison of deep learning performance against health-care professionals in detecting diseases from medical imaging: a systematic review and meta-analysis. Lancet Digit Health 2019;1:e271–97. doi:10.1016/S2589-7500(19)30123-2
      OpenUrl
    31. ↵
      1. Gulshan V,
      2. Peng L,
      3. Coram M, et al
      . Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus Photographs. JAMA 2016;316:2402–10. doi:10.1001/jama.2016.17216
      OpenUrlCrossRefPubMed
    32. ↵
      1. Taylor S,
      2. Brown JM,
      3. Gupta K, et al
      . Monitoring disease progression with a quantitative severity scale for retinopathy of prematurity using deep learning. JAMA Ophthalmol 2019;137:1022. doi:10.1001/jamaophthalmol.2019.2433
      OpenUrl
    33. ↵
      1. Li J-P,
      2. Liu H,
      3. Ting DSJ, et al
      . Digital technology, tele-medicine and artificial intelligence in ophthalmology: a global perspective. Prog Retin Eye Res 2021;82:100900. doi:10.1016/j.preteyeres.2020.100900
      OpenUrlCrossRefPubMed
    34. ↵
      1. Li W,
      2. Yang Y,
      3. Zhang K, et al
      . Dense anatomical annotation of slit-lamp images improves the performance of deep learning for the diagnosis of ophthalmic disorders. Nat Biomed Eng 2020;4:767–77. doi:10.1038/s41551-020-0577-y
      OpenUrl
    35. ↵
      1. Tiwari M,
      2. Piech C,
      3. Baitemirova M, et al
      . Differentiation of active corneal infections from healed scars using deep learning. Ophthalmology 2022;129:139–46. doi:10.1016/j.ophtha.2021.07.033
      OpenUrl
    36. ↵
      1. Koyama A,
      2. Miyazaki D,
      3. Nakagawa Y, et al
      . Determination of probability of causative pathogen in infectious keratitis using deep learning algorithm of slit-lamp images. Sci Rep 2021;11:22642. doi:10.1038/s41598-021-02138-w
    37. ↵
      1. Ting DSJ,
      2. Foo VH,
      3. Yang LWY, et al
      . Artificial intelligence for anterior segment diseases: emerging applications in ophthalmology. Br J Ophthalmol 2021;105:158–68. doi:10.1136/bjophthalmol-2019-315651
      OpenUrlAbstract/FREE Full Text
    38. ↵
      1. Li Z,
      2. Jiang J,
      3. Chen K, et al
      . Preventing corneal blindness caused by keratitis using artificial intelligence. Nat Commun 2021;12. doi:10.1038/s41467-021-24116-6
    39. ↵
      1. Hung N,
      2. Shih AK-Y,
      3. Lin C, et al
      . Using slit-lamp images for deep learning-based identification of bacterial and fungal keratitis: model development and validation with different convolutional neural networks. Diagnostics (Basel) 2021;11:1246. doi:10.3390/diagnostics11071246
    40. ↵
      1. Shamseer L,
      2. Moher D,
      3. Clarke M, et al
      . Preferred reporting items for systematic review and meta-analysis protocols (PRISMA-P) 2015: elaboration and explanation. BMJ 2015;350:g7647. doi:10.1136/bmj.g7647
    41. ↵
      1. Salameh J-P,
      2. Bossuyt PM,
      3. McGrath TA, et al
      . Preferred reporting items for systematic review and meta-analysis of diagnostic test accuracy studies (PRISMA-DTA): explanation, elaboration, and checklist. BMJ 2020;370:m2632. doi:10.1136/bmj.m2632
    42. ↵
      1. Moons KGM,
      2. de Groot JAH,
      3. Bouwmeester W, et al
      . Critical appraisal and data extraction for systematic reviews of prediction modelling studies: the charms checklist. PLoS Med 2014;11:e1001744. doi:10.1371/journal.pmed.1001744
    43. ↵
      1. Ouzzani M,
      2. Hammady H,
      3. Fedorowicz Z, et al
      . Rayyan-a web and mobile APP for systematic reviews. Syst Rev 2016;5:210. doi:10.1186/s13643-016-0384-4
      OpenUrlCrossRefPubMed
    44. ↵
      1. Rampat R,
      2. Deshmukh R,
      3. Chen X, et al
      . Artificial intelligence in cornea, refractive surgery, and cataract: basic principles, clinical applications, and future directions. Asia Pac J Ophthalmol (Phila) 2021;10:268–81. doi:10.1097/APO.0000000000000394
      OpenUrl
    45. ↵
      1. Whiting PF,
      2. Rutjes AWS,
      3. Westwood ME, et al
      . QUADAS-2: a revised tool for the quality assessment of diagnostic accuracy studies. Ann Intern Med 2011;155:529–36. doi:10.7326/0003-4819-155-8-201110180-00009
      OpenUrlCrossRefPubMedWeb of Science
    46. ↵
      1. Sounderajah V,
      2. Ashrafian H,
      3. Rose S, et al
      . A quality assessment tool for artificial intelligence-centered diagnostic test accuracy studies: QUADAS-AI. Nat Med 2021;27:1663–5. doi:10.1038/s41591-021-01517-0
      OpenUrlCrossRefPubMed
    47. ↵
      1. Campbell JM,
      2. Klugar M,
      3. Ding S, et al
      . Diagnostic test accuracy: methods for systematic review and meta-analysis. Int J Evid Based Healthc 2015;13:154–62. doi:10.1097/XEB.0000000000000061
      OpenUrlCrossRefPubMed
    48. ↵
      1. Deeks JJ,
      2. Macaskill P,
      3. Irwig L
      . The performance of tests of publication bias and other sample size effects in systematic reviews of diagnostic test accuracy was assessed. J Clin Epidemiol 2005;58:882–93. doi:10.1016/j.jclinepi.2005.01.016
      OpenUrlCrossRefPubMedWeb of Science
    49. ↵
      1. van Enst WA,
      2. Ochodo E,
      3. Scholten R, et al
      . Investigation of publication bias in meta-analyses of diagnostic test accuracy: a meta-epidemiological study. BMC Med Res Methodol 2014;14:70. doi:10.1186/1471-2288-14-70

    Footnotes

    • Contributors DSJT designed and conceptualised the study. ZZO, YS, XL, RQ, S-HL, TL, VS, HA, DT, DS, JM, MJB, HSD and DSJT developed the study protocol. ZZO and YS collected the data. ZZO, YS, XL, RQ, S-HL, TL, VS, HA, DT, DS, JM, MJB, HSD and DSJT interpreted the data. ZZO and DSJT drafted the initial manuscript. ZZO, YS, XL, RQ, S-HL, TL, VS, HA, DT, DS, JM, MJB, HSD and DSJT critically reviewed, revised and approved the final manuscript. DSJT acts as the guarantor of the study.

    • Funding MJB is supported by the Wellcome Trust (grant number 207472/Z/17/Z). DSJT is supported by Medical Research Council/Fight for Sight Clinical Research Fellowship (MR/T001674/1).

    • Disclaimer The funder has no role in developing this protocol.

    • Competing interests None declared.

    • Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.

    • Provenance and peer review Not commissioned; externally peer reviewed.