Different fundus imaging modalities and technical factors in AI screening for diabetic retinopathy: a review

BACKGROUND

Effective screening is a desirable method for the early detection and successful treatment for diabetic retinopathy, and fundus photography is currently the dominant medium for retinal imaging due to its convenience and accessibility. Manual screening using fundus photographs has however involved considerable costs for patients, clinicians and national health systems, which has limited its application particularly in less-developed countries. The advent of artificial intelligence, and in particular deep learning techniques, has however raised the possibility of widespread automated screening.

MAIN TEXT

In this review, we first briefly survey major published advances in retinal analysis using artificial intelligence. We take care to separately describe standard multiple-field fundus photography, and the newer modalities of ultra-wide field photography and smartphone-based photography. Finally, we consider several machine learning concepts that have been particularly relevant to the domain and illustrate their usage with extant works.

CONCLUSIONS

In the ophthalmology field, it was demonstrated that deep learning tools for diabetic retinopathy show clinically acceptable diagnostic performance when using colour retinal fundus images. Artificial intelligence models are among the most promising solutions to tackle the burden of diabetic retinopathy management in a comprehensive manner. However, future research is crucial to assess the potential clinical deployment, evaluate the cost-effectiveness of different DL systems in clinical practice and improve clinical acceptance.

Health Economic and Safety Considerations for Artificial Intelligence Applications in Diabetic Retinopathy Screening

Systematic screening for diabetic retinopathy (DR) has been widely recommended for early detection in patients with diabetes to address preventable vision loss. However, substantial manpower and financial resources are required to deploy opportunistic screening and transition to systematic DR screening programs. The advent of artificial intelligence (AI) technologies may improve access and reduce the financial burden for DR screening while maintaining comparable or enhanced clinical effectiveness. To deploy an AI-based DR screening program in a real-world setting, it is imperative that health economic assessment (HEA) and patient safety analyses are conducted to guide appropriate allocation of resources and design safe, reliable systems. Few studies published to date include these considerations when integrating AI-based solutions into DR screening programs. In this article, we provide an overview of the current state-of-the-art of AI technology (focusing on deep learning systems), followed by an appraisal of existing literature on the applications of AI in ophthalmology. We also discuss practical considerations that drive the development of a successful DR screening program, such as the implications of false-positive or false-negative results and image gradeability. Finally, we examine different plausible methods for HEA and safety analyses that can be used to assess concerns regarding AI-based screening.

Automated and Computer-Assisted Detection, Classification, and Diagnosis of Diabetic Retinopathy

Background: The introduction of artificial intelligence (AI) in medicine has raised significant ethical, economic, and scientific controversies. Introduction: Because an explicit goal of AI is to perform processes previously reserved for human clinicians and other health care personnel, there is justified concern about the impact on patient safety, efficacy, equity, and liability. Discussion: Systems for computer-assisted and fully automated detection, triage, and diagnosis of diabetic retinopathy (DR) from retinal images show great variation in design, level of autonomy, and intended use. Moreover, the degree to which these systems have been evaluated and validated is heterogeneous. We use the term DR AI system as a general term for any system that interprets retinal images with at least some degree of autonomy from a human grader. We put forth these standardized descriptors to form a means to categorize systems for computer-assisted and fully automated detection, triage, and diagnosis of DR. The components of the categorization system include level of device autonomy, intended use, level of evidence for diagnostic accuracy, and system design. Conclusion: There is currently minimal empirical basis to assert that certain combinations of autonomy, accuracy, or intended use are better or more appropriate than any other. Therefore, at the current stage of development of this document, we have been descriptive rather than prescriptive, and we treat the different categorizations as independent and organized along multiple axes.

25-years Trends and Risk factors related to Surgical Outcomes of Giant Retinal Tear-Rhegmatogenous Retinal Detachments

To describe the 25-year surgical trends, long-term outcomes and risk factors affecting the outcomes of giant retinal tear-related rhegmatogenous retinal detachments (GRT-RRD). Patients’ demographics, pre-operative characteristics, risk factors, operative procedures and post-operative outcomes were collected and divided into three groups – Group A: 1991 to 2015 (overall); Group B: 1991 to 2005, and Group C: 2006 to 2015. Functional and anatomical successes were monitored over a 5-year period. Multivariate logistic regression analysis was performed to identify the risk factors related to functional and anatomical success.127 eyes of 127 patients were included in the study. At 5^th year, 69.4% patients had visual acuity (VA) < logMAR 1.0 with 87.5% primary anatomical success rate. While the functional outcome remained the same between group B and C, there was an increase in the anatomical success from 89.7% to 100%, albeit not statistically significant. Patients with worse presenting VA, 150 degrees or more of giant retina tear, macula-detached status and presence of PVR were associated with VA of> logMAR 1.0 (all p < 0.05). The types of surgery (TPPV vs combined SB/TPPV), number of breaks, lens extraction and additional cryotherapy were not associated with the functional or anatomical success. In conclusion, the GRT-RRD functional and structural outcomes were comparable between 1991–2005 and 2006–2015, albeit a statistically insignificant improvement of anatomical outcome over the past 25 years. Worse presenting VA, 150 degrees or more of giant retinal tear, detached macula and presence of PVR were associated with poorer visual outcome.

Technical and imaging factors influencing performance of deep learning systems for diabetic retinopathy

Deep learning (DL) has been shown to be effective in developing diabetic retinopathy (DR) algorithms, possibly tackling financial and manpower challenges hindering implementation of DR screening. However, our systematic review of the literature reveals few studies studied the impact of different factors on these DL algorithms, that are important for clinical deployment in real-world settings. Using 455,491 retinal images, we evaluated two technical and three image-related factors in detection of referable DR. For technical factors, the performances of four DL models (VGGNet, ResNet, DenseNet, Ensemble) and two computational frameworks (Caffe, TensorFlow) were evaluated while for image-related factors, we evaluated image compression levels (reducing image size, 350, 300, 250, 200, 150 KB), number of fields (7-field, 2-field, 1-field) and media clarity (pseudophakic vs phakic). In detection of referable DR, four DL models showed comparable diagnostic performance (AUC 0.936-0.944). To develop the VGGNet model, two computational frameworks had similar AUC (0.936). The DL performance dropped when image size decreased below 250 KB (AUC 0.936, 0.900, p < 0.001). The DL performance performed better when there were increased number of fields (dataset 1: 2-field vs 1-field-AUC 0.936 vs 0.908, p < 0.001; dataset 2: 7-field vs 2-field vs 1-field, AUC 0.949 vs 0.911 vs 0.895). DL performed better in the pseudophakic than phakic eyes (AUC 0.918 vs 0.833, p < 0.001). Various image-related factors play more significant roles than technical factors in determining the diagnostic performance, suggesting the importance of having robust training and testing datasets for DL training and deployment in the real-world settings.

Correction to: Artificial intelligence for diabetic retinopathy screening: a review

An amendment to this paper has been published and can be accessed via a link at the top of the paper.

Real-World Treatment Outcomes of Age-Related Macular Degeneration and Polypoidal Choroidal Vasculopathy in Asians

PURPOSE

To describe the 12-month outcomes of treatment-naïve eyes with choroidal neovascularization (CNV) resulting from age-related macular degeneration (AMD) and polypoidal choroidal vasculopathy (PCV) after initiation of intravitreal anti-vascular endothelial growth factor (VEGF) monotherapy or combination therapy with verteporfin photodynamic therapy (PDT).

DESIGN

A 12-month single-center, retrospective, comparative, nonrandomized cohort study.

PARTICIPANTS

Patients with AMD or PCV who initiated intravitreal anti-VEGF therapy during 2015.

METHODS

Demographics, visual outcomes, OCT, and treatment data were collected at baseline and months 1, 3, 6, and 12 after treatment initiation. Multivariate analysis was performed to identify baseline features predictive of visual maintenance and improvement after 12 months of treatment.

MAIN OUTCOME MEASURES

Primary end point was visual acuity (VA) change from baseline to month 12. Secondary end points were treatment exposure and change in central subfield thickness on OCT.

RESULTS

A total of 364 patients (165 AMD and 199 PCV) were included. Baseline vision was 41 and 43 logarithm of the minimum angle of resolution (logMAR) letters for AMD and PCV patients, respectively. Patients with AMD and PCV received 5.5 and 5.3 injections (5.0 monotherapy vs. 5.6 combination therapy; mean, 1.2 PDT sessions), respectively. Patients with AMD gained 4.7 logMAR letters after 12 months (P = 0.002), whereas PCV patients gained 6.6 logMAR letters (P = 0.001) and 10.8 logMAR letters (P < 0.001) for monotherapy and combination therapy, respectively. Only patients with presenting VA of fewer than 35 letters (Snellen equivalent, 6/60) achieved significant visual improvement (10.4 letters for AMD, 17.1 letters for PCV with monotherapy, and 35.5 letters for PCV with combination therapy). Predictors of VA gain included number of intravitreal injections (AMD and PCV adjusted odds ratio, 12.1 [P = 0.001] and 12.5 [P = 0.004] for ≥7 injections, respectively) and baseline VA of 20 logMAR letters or fewer (adjusted odds ratio, 3.8 and 10.6 for AMD and PCV, respectively). Age, gender, race, use of PDT or focal laser therapy, and central subfield thickness were not predictive of significant visual gain at 12 months.

CONCLUSIONS

In Asian patients, treatment of AMD with anti-VEGF therapy yielded 12-month visual outcomes comparable with those of other real-world studies from Western populations but poorer than those of controlled trials. In contrast, for PCV eyes, anti-VEGF monotherapy and combination therapy with PDT yielded comparable outcomes as those of controlled clinical trials.

Deep learning in estimating prevalence and systemic risk factors for diabetic retinopathy: a multi-ethnic study

In any community, the key to understanding the burden of a specific condition is to conduct an epidemiological study. The deep learning system (DLS) recently showed promising diagnostic performance for diabetic retinopathy (DR). This study aims to use DLS as the grading tool, instead of human assessors, to determine the prevalence and the systemic cardiovascular risk factors for DR on fundus photographs, in patients with diabetes. This is a multi-ethnic (5 races), multi-site (8 datasets from Singapore, USA, Hong Kong, China and Australia), cross-sectional study involving 18,912 patients (n = 93,293 images). We compared these results and the time taken for DR assessment by DLS versus 17 human assessors - 10 retinal specialists/ophthalmologists and 7 professional graders). The estimation of DR prevalence between DLS and human assessors is comparable for any DR, referable DR and vision-threatening DR (VTDR) (Human assessors: 15.9, 6.5% and 4.1%; DLS: 16.1%, 6.4%, 3.7%). Both assessment methods identified similar risk factors (with comparable AUCs), including younger age, longer diabetes duration, increased HbA1c and systolic blood pressure, for any DR, referable DR and VTDR (p > 0.05). The total time taken for DLS to evaluate DR from 93,293 fundus photographs was ~1 month compared to 2 years for human assessors. In conclusion, the prevalence and systemic risk factors for DR in multi-ethnic population could be determined accurately using a DLS, in significantly less time than human assessors. This study highlights the potential use of AI for future epidemiology or clinical trials for DR grading in the global communities.

Ten Emerging Trends in the Epidemiology of Diabetic Retinopathy

PURPOSE

Diabetes is a major public health problem affecting 415 million people worldwide. With the increasing prevalence of diabetes, diabetic retinopathy (DR) is emerging as the leading cause of avoidable blindness worldwide.

METHODS

We reviewed previous and recent literature to provide an overview of emerging trends on the burden, epidemiology, risk factors, and prevention of DR.

RESULTS

First, there is clear evidence of a global increase in the prevalence of diabetes. Second, there is a decline in the incidence of blindness due to DR, particularly in developed countries. Third, diabetic macular edema (DME) rather than proliferative diabetic retinopathy (PDR) is the increasingly common cause of visual impairment. Fourth, DR awareness remains patchy and low in most populations. Fifth, hyperglycemia remains the most consistent risk factor for DR in type 1 diabetes across different studies and populations. Sixth, in contrast, blood pressure is an important risk factor for DR in type 2 diabetes. Seventh, the relationship between dyslipidemia and DR remains unclear, with inconsistent results from different studies and trials. Eighth, the utility of predictive models incorporating multiple risk factors for assessing DR risk requires evaluation. Ninth, photographic screening of DR using tele-ophthalmology platforms is increasingly recognized as being feasible and cost-effective. Finally, DR prevention in low-resource settings cannot follow models developed in high-resource countries and requires different strategies.

CONCLUSIONS

The ten trends we observed in the current review may guide planning of public healthcare strategies for the management of DR and prevention of blindness.

Light and portable novel device for diabetic retinopathy screening

BACKGROUND

To validate the use of an economical portable multipurpose ophthalmic imaging device, EyeScan (Ophthalmic Imaging System, Sacramento, CA, USA), for diabetic retinopathy screening.

DESIGN

Evaluation of a diagnostic device.

PARTICIPANTS

One hundred thirty-six (272 eyes) were recruited from diabetic retinopathy screening clinic of Royal Perth Hospital, Western Australia, Australia.

METHODS

All patients underwent three-field (optic disc, macular and temporal view) mydriatic retinal digital still photography captured by EyeScan and FF450 plus (Carl Zeiss Meditec, North America) and were subsequently examined by a senior consultant ophthalmologist using the slit-lamp biomicroscopy (reference standard). All retinal images were interpreted by a consultant ophthalmologist and a medical officer.

MAIN OUTCOME MEASURES

The sensitivity, specificity and kappa statistics of EyeScan and FF450 plus with reference to the slit-lamp examination findings by a senior consultant ophthalmologist.

RESULTS

For detection of any grade of diabetic retinopathy, EyeScan had a sensitivity and specificity of 93 and 98%, respectively (ophthalmologist), and 92 and 95%, respectively (medical officer). In contrast, FF450 plus images had a sensitivity and specificity of 95 and 99%, respectively (ophthalmologist), and 92 and 96%, respectively (medical officer). The overall kappa statistics for diabetic retinopathy grading for EyeScan and FF450 plus were 0.93 and 0.95 for ophthalmologist and 0.88 and 0.90 for medical officer, respectively.

CONCLUSIONS

Given that the EyeScan requires minimal training to use and has excellent diagnostic accuracy in screening for diabetic retinopathy, it could be potentially utilized by the primary eye care providers to widely screen for diabetic retinopathy in the community.

Diabetic retinopathy management by Australian optometrists

BACKGROUND

To survey the current diabetic retinopathy screening and management practices of Australian optometrists following the release of the 1997 National Health Medical Research Council Diabetic Retinopathy Management Guidelines.

DESIGN

Cross-sectional national survey, primary care setting.

PARTICIPANTS

1000 Australian optometrists across different states.

METHODS

A self-administered questionnaire was sent to 1000 optometrists across all states during 2007/2008.

MAIN OUTCOME MEASURES

Use of retinal camera, screening practices/attitudes and behaviour in diabetic retinopathy management.

RESULTS

568 optometrists (57%) responded to the survey. Patients' unpreparedness to drive post dilation (51%) and the fear of angle closure glaucoma (13%) were the two main barriers to optometrists not performing dilated ophthalmoscopy. Those who had strong desire to screen for diabetic retinopathy were more likely to use a retinal camera (p<0.005). Use of a retinal camera was significantly associated with an increased confidence in detecting clinical signs of diabetic retinopathy including macular oedema (P<0.001). Optometrists who read the guidelines at least once were 2.5-times (P<0.001) more likely to have confidence in detecting macular oedema than those who had never read the guidelines. Although they may be confident in diagnosis, and may use retinal cameras for screening, nearly 60% of optometrists would not refer patients with macular oedema to an ophthalmologist.

CONCLUSIONS

Despite their self-reported desire for involvement in diabetic retinopathy, the management of macular oedema by Australian optometrists needs improvement. The use of retinal cameras and promotion of the 2008 NHMRC guidelines should be encouraged to improve overall optometric diabetic retinopathy management, particularly with macular oedema.

Visual neglect following stroke: current concepts and future focus

Visual neglect is a common, yet frequently overlooked, neurological disorder following stroke characterized by a deficit in attention and appreciation of stimuli on the contralesional side of the body. It has a profound functional impact on affected individuals. A assessment and management of this condition are hindered, however, by the lack of professional awareness and clinical guidelines. Recent evidence suggests that the underlying deficit in visual attention is due to a disrupted internalized representation of the outer world rather than a disorder of sensory inputs. Dysfunction of the cortical domains and white-matter tracts, as well as inter-hemispheric imbalance, have been implicated in the various manifestations of visual neglect. Optimal diagnosis requires careful history-taking from the patient, family, and friends, in addition to clinical assessment with the line bisection test, the star cancellation test, and the Catherine Bergego Scale. Early recognition and prompt rehabilitation employing a multidisciplinary approach is desirable. Although no treatment has been definitively shown to be of benefit, those with promise include prism adaptation, visual scanning therapy, and virtual reality-based techniques. Further high quality research to seek optimum short- and long-term rehabilitative strategies for visual neglect is required.