Artificial intelligence in diabetic retinopathy screening: from idea to a medical device in clinical practice.

With the growing significance of artificial intelligence in healthcare, new perspectives are emerging in primary care. Diabetic retinopathy, a microvascular complication of diabetes mellitus, often remains unnoticed until patient is facing complications. Artificial intelligence presents a promising solution that can enhance the accessibility of diabetic retinopathy screening for a broader range of patients. The key challenge lies in successfully integrating the solution into clinical practice, a demanding process with multiple phases to ensure the resulting medical device is effective and safe for patient use. Aireen software uses artificial intelligence to perform diabetic retinopathy screening on retinal images captured by optical fundus cameras. The medical device complies with European Medical Device Regulation 2017/745 and was introduced to the market in 2023. Collaboration between physicians and the development team played a crucial role throughout the entire lifecycle of the medical device. Physicians were engaged in defining the intended use of the medical device, risk analysis, data annotation for training and software validation, as well as throughout a clinical trial. A clinical trial was conducted on 1,274 patients with type 1 and type 2 diabetes mellitus, where Aireen medical device achieved a sensitivity of 94.0% and a specificity of 90.7% compared to the reference evaluation. This clinical trial confirmed the potential of Aireen to enhance the availability of diabetic retinopathy screening and early disease detection.

Comparison of 2-Field and 5-Field Mydriatic Handheld Retinal Imaging in a Community-Based Diabetic Retinopathy Screening Program.

INTRODUCTION: The purpose of this study was to compare 2-field (2F) and 5-field (5F) mydriatic handheld retinal imaging for the assessment of diabetic retinopathy (DR) severity in a community-based DR screening program (DRSP). METHODS: This was a prospective, cross-sectional diagnostic study, evaluating images of 805 eyes from 407 consecutive patients with diabetes acquired from a community-based DRSP. Mydriatic standardized 5F imaging (macula, disc, superior, inferior, temporal) with handheld retinal camera was performed. 2F (disc, macula), and 5F images were independently assessed using the International DR classification at a centralized reading center. Simple (K) and weighted (Kw) kappa statistics were calculated for DR. Sensitivity and specificity for referable DR ([refDR] moderate nonproliferative DR [NPDR] or worse) and vision-threatening DR ([vtDR] severe NPDR or worse) for 2F compared to 5F imaging were calculated. RESULTS: Distribution of DR severity by 2F/5F images (%): no DR 66.0/61.7, mild NPDR 10.7/14.4, moderate NPDR 7.9/8.1, severe NPDR 3.3/5.6, proliferative DR 5.6/4.6, ungradable 6.5/5.6. Exact agreement of DR grading between 2F and 5F was 81.7%, within 1-step 97.1% (K = 0.64, Kw = 0.78). Sensitivity/specificity for 2F compared 5F was refDR 0.80/0.97, vtDR 0.73/0.98. The ungradable images rate with 2F was 16.1% higher than with 5F (6.5 vs. 5.6%, p < 0.001). CONCLUSIONS: Mydriatic 2F and 5F handheld imaging have substantial agreement in assessing severity of DR. However, the use of mydriatic 2F handheld imaging only meets the minimum standards for sensitivity and specificity for refDR but not for vtDR. When using handheld cameras, the addition of peripheral fields in 5F imaging further refines the referral approach by decreasing ungradable rate and increasing sensitivity for vtDR.

Relative Importance of Glaucoma-Referral Indicators in Retinal Images in a Diabetic Retinopathy Screening Programme in Slovenia: A Cross-Sectional Study.

Background and Objectives: Glaucoma is a major cause of irreversible visual impairment and blindness, so its timely detection is crucial. Retinal images from diabetic retinopathy screening programmes (DRSP) provide an opportunity to detect undiagnosed glaucoma. Our aim was to find out which retinal image indicators are most suitable for referring DRSP patients for glaucoma assessment and to determine the glaucoma detection potential of Slovenian DRSP. Materials and Methods: We reviewed retinal images of patients from the DRSP at the University Medical Centre Ljubljana (November 2019-January 2020, May-August 2020). Patients with at least one indicator and some randomly selected patients without indicators were invited for an eye examination. Suspect glaucoma and glaucoma patients were considered accurately referred. Logistic regression (LOGIT) with patients as statistical units and generalised estimating equation with logistic regression (GEE) with eyes as statistical units were used to determine the referral accuracy of indicators. Results: Of the 2230 patients reviewed, 209 patients (10.1%) had at least one indicator on a retinal image of either one eye or both eyes. A total of 149 (129 with at least one indicator and 20 without) attended the eye exam. Seventy-nine (53.0%) were glaucoma negative, 54 (36.2%) suspect glaucoma, and 16 (10.7%) glaucoma positive. Seven glaucoma patients were newly detected. Neuroretinal rim notch predicted glaucoma in all cases. The cup-to-disc ratio was the most important indicator for accurate referral (odds ratio 7.59 (95% CI 3.98-14.47; p < 0.001) and remained statistically significant multivariably. Family history of glaucoma also showed an impact (odds ratio 3.06 (95% CI 1.02-9.19; p = 0.046) but remained statistically significant only in the LOGIT multivariable model. Other indicators and confounders were not statistically significant in the multivariable models. Conclusions: Our results suggest that the neuroretinal rim notch and cup-to-disc ratio are the most important for accurate glaucoma referral from retinal images in DRSP. Approximately half of the glaucoma cases in DRSPs may be undiagnosed.

Mixed diabetic retinopathy screening coverage results in Indigenous Australian primary care settings: A nurse-led model of integrated diabetes care.

AIMS: To determine eye screening coverage and adherence to national eye screening recommendations of a nurse-led retinal image-based model of diabetes education and eye screening in Indigenous primary care clinics. DESIGN: A pre-post study. METHODS: During January 2018-March 2020 Indigenous Australians with diabetes at three regional Australian clinics were offered eye screening by a nurse-diabetes educator/retinal imager. At the main site the nurse recruited/scheduled participants, and at satellite sites local clinic staff did so. Visual acuity was tested and digital retinal images acquired and graded. Participants were offered rescreening at or before 12-months based on baseline results. RESULTS: In total 203 adults with diabetes attending Indigenous primary care clinics were screened, with divergent results based on the recruitment methods. At the main clinic 135 of 172 eligible adults (79%) were screened. At the satellite sites, 15 of 85 (18%) and 21 of 77 (27%) diabetes patients were screened. Combined coverage 51%. CONCLUSION: A credentialed nurse-educator implemented a model of retinal image-based diabetes education, measured eye screening coverage and adherence to national eye screening guidelines, met the 'acceptable 75% eye screening coverage' benchmark and improved patient eye screening guideline adherence at the one site where the nurse-educator had access to patient recruitment and scheduling. IMPACT: This novel nurse-led primary care iDEES model of retinal image-based diabetes education can improve the currently low Indigenous diabetes eye screening coverage in Australia. Importantly, the nurse-managed iDEES model of integrated diabetes care is readily adaptable to other settings and populations where access to and/or uptake of eye care is suboptimal. CLINICAL TRIAL REGISTRATION: ANZCTRN1261800120435.

A model of culturally-informed integrated diabetes education and eye screening in indigenous primary care services and specialist diabetes clinics: Study protocol.

AIMS: To improve diabetes management in Indigenous Australians using an integrated nurse-led model of diabetes education and eye screening in indigenous primary care and specialist diabetes clinics. DESIGN: A pre-post study. METHODS: This study will be implemented in indigenous primary care and specialist diabetes clinics in Victoria, Australia. Participants recruited to the study will be existing adult patient with diagnosed diabetes attending study sites. A nurse-credentialled diabetes educator and certified retinal imager will deliver three study components: (a) retinal photography as a diabetic retinopathy screening and patient engagement tool; (b) lifestyle and behaviour surveys, administered at baseline and at the final visit, in 12 months. Findings from the surveys and participants' retinal images will be used to guide; and (c) personalized diabetes education. The primary outcomes are participant adherence to diabetic eye screening recommendations and health service diabetic retinopathy screening coverage. Secondary outcomes are baseline DR prevalence and changes in clinical and lifestyle risk factor levels, diabetes knowledge and satisfaction with diabetes care. DISCUSSION: Compared with non-indigenous Australians, Indigenous Australians have a high prevalence of diabetic retinopathy and blindness, low adherence to eye screening recommendations and suboptimal health literacy. Nurse-credentialled diabetes educators can be trained to incorporate retinal imaging and eye screening into their clinical practice to give image-based diabetes education to facilitate diabetic retinopathy management. IMPACT: Credentialled nurse diabetes educators who integrate eye screening and diabetes education can facilitate timelier diabetic retinopathy screening, referral pathways and treatment of sight-threatening retinopathy. We believe that this model of integrated diabetes education and eye screening will also improve adherence to eye screening recommendations, population screening coverage, health literacy, risk factor levels and diabetes self-care. CLINICAL TRIAL REGISTRATION: ANZCTRN1261800120435.

Screening for diabetic retinopathy in primary care: Future prospects in low-middle income countries.

Diabetic retinopathy (DR) is an urgent public health concern, and it is prudent to prevent DR than face the burden of blindness arising from it. This requires committed physicians, ophthalmologists, patients as well as policy makers. Blindness rate due to DR has shown a decline in developed countries which is not only due to advanced and accessible tertiary care, but also large-scale screening programmes. Currently, the most common method of screening in low- and middle-income countries is still an opportunistic model. A more practical, cost-effective and systematic screening model is needed, utilizing advances in telemedicine and artificial intelligence.

Personalising screening of sight-threatening diabetic retinopathy - qualitative evidence to inform effective implementation.

BACKGROUND: Internationally, systematic screening for sight-threatening diabetic retinopathy (STDR) usually includes annual recall. Researchers and policy-makers support extending screening intervals, citing evidence from observational studies with low incidence rates. However, there is little research around the acceptability to people with diabetes (PWD) and health care professionals (HCP) about changing eye screening intervals. METHODS: We conducted a qualitative study to explore issues surrounding acceptability and the barriers and enablers for changing from annual screening, using in-depth, semistructured interviews analysed using the constant comparative method. PWD were recruited from general practices and HCP from eye screening networks and related specialties in North West England using purposive sampling. Interviews were conducted prior to the commencement of and during a randomised controlled trial (RCT) comparing fixed annual with variable (6, 12 or 24 month) interval risk-based screening. RESULTS: Thirty PWD and 21 HCP participants were interviewed prior to and 30 PWD during the parallel RCT. The data suggests that a move to variable screening intervals was generally acceptable in principle, though highlighted significant concerns and challenges to successful implementation. The current annual interval was recognised as unsustainable against a backdrop of increasing diabetes prevalence. There were important caveats attached to acceptability and a need for clear safeguards around: the safety and reliability of calculating screening intervals, capturing all PWD, referral into screening of PWD with diabetic changes regardless of planned interval. For PWD the 6-month interval was perceived positively as medical reassurance, and the 12-month seen as usual treatment. Concerns were expressed by many HCP and PWD that a 2-year interval was too lengthy and was risky for detecting STDR. There were also concerns about a negative effect upon PWD care and increasing non-attendance rates. Amongst PWD, there was considerable conflation and misunderstanding about different eye-related appointments within the health care system. CONCLUSIONS: Implementing variable-interval screening into clinical practice is generally acceptable to PWD and HCP with important caveats, and misconceptions must be addressed. Clear safeguards against increasing non-attendance, loss of diabetes control and alternative referral pathways are required. For risk calculation systems to be safe, reliable monitoring and clear communication is required.

Artificial Intelligence Screening for Diabetic Retinopathy: the Real-World Emerging Application.

PURPOSE OF REVIEW: This paper systematically reviews the recent progress in diabetic retinopathy screening. It provides an integrated overview of the current state of knowledge of emerging techniques using artificial intelligence integration in national screening programs around the world. Existing methodological approaches and research insights are evaluated. An understanding of existing gaps and future directions is created. RECENT FINDINGS: Over the past decades, artificial intelligence has emerged into the scientific consciousness with breakthroughs that are sparking increasing interest among computer science and medical communities. Specifically, machine learning and deep learning (a subtype of machine learning) applications of artificial intelligence are spreading into areas that previously were thought to be only the purview of humans, and a number of applications in ophthalmology field have been explored. Multiple studies all around the world have demonstrated that such systems can behave on par with clinical experts with robust diagnostic performance in diabetic retinopathy diagnosis. However, only few tools have been evaluated in clinical prospective studies. Given the rapid and impressive progress of artificial intelligence technologies, the implementation of deep learning systems into routinely practiced diabetic retinopathy screening could represent a cost-effective alternative to help reduce the incidence of preventable blindness around the world.

Integrating diabetic retinopathy screening within diabetes education services in Australia's diabetes and indigenous primary care clinics.

As diabetes occurs in all ethnicities and regions it is essential that retinopathy screening be widely available. Screening rates are lower in Indigenous than in non-Indigenous Australians. Technological advances and Medicare rebates should facilitate improved outcomes. Use of non-ophthalmic clinicians, (general practitioners, diabetes educators, health-workers and endocrinologists) to supplement coverage by ophthalmologists and optometrists would extend retinopathy screening capacity. Diabetes educators are an integral part of diabetes management. Integrating ocular screening and diabetes education in primary care settings has potential to improve synergistically retinopathy screening coverage, patient self-management, risk factor control, care satisfaction, health economics and sustainability of under-resourced services.

Diabetic eye disease and screening attendance by ethnicity in New Zealand: A systematic review.

To examine differences in incidence, prevalence and screening for diabetic retinopathy in New Zealand, we searched MEDLINE, EMBASE and CINAHL up to 6 December 2018 for observational studies reporting diabetic eye disease or attendance at retinal screening, disaggregated by ethnicity. Two authors separately screened and selected studies, and extracted data. None of the 11 included studies reported data on visual impairment from diabetic retinopathy. All nine studies reporting diabetic eye disease by ethnicity found Pacific people and Maori had higher rates of sight-threatening disease and lower rates of screening attendance compared to Europeans. Data for Asian people were infrequently reported, but when they were, they also fared worse than Europeans. This review highlights that equity-focused strategies are needed to address ethnic disparities in eye health among New Zealanders with diabetes. The review also identifies how research methods can be strengthened to enable future calculation of robust disease prevalence estimates.

Cost of diabetic retinopathy and macular oedema in a population, an eight year follow up.

BACKGROUND: Prospective, population-based study of an 8-year follow up. To determine the direct cost of diabetic retinopathy [DR], evaluating our screening programme and the cost of treating DR, focusing on diabetic macular oedema [DMO] after anti-vascular endothelial growth factor [anti-VEGF] treatment. METHODS: A total of 15,396 diabetes mellitus [DM] patients were studied. We determined the cost-effectiveness of our screening programme against an annual programme by applying the Markov simulation model. We also compared the cost-effectiveness of anti-VEGF treatment to laser treatment for screened patients with DMO. RESULTS: The cost of our 2.5-year screening programme was as follows: per patient with any-DR, €482.85 ± 35.14; per sight-threatening diabetic retinopathy [STDR] patient, €1528.26 ± 114.94; and €1826.98 ± 108.26 per DMO patient. Comparatively, an annual screening programme would result in increases as follows: 0.77 in QALY per patient with any-DR and 0.6 and 0.44 per patient with STDR or DMO, respectively, with an incremental cost-effective ratio [ICER] of €1096.88 for any-DR, €4571.2 for STDR and €7443.28 per DMO patient. Regarding diagnosis and treatment, the mean annual total cost per patient with DMO was €777.09 ± 49.45 for the laser treated group and €7153.62 ± 212.15 for the anti-VEGF group, with a QALY gain of 0.21, the yearly mean cost was €7153.62 ± 212.15 per patient, and the ICER was €30,361. CONCLUSIONS: Screening for diabetic retinopathy every 2.5 years is cost-effective, but should be adjusted to a patient's personal risk factors. Treatment with anti-VEGF for DMO has increased costs, but the cost-utility increases to 0.21 QALY per patient.

Screening for diabetic retinopathy with artificial intelligence: a real world evaluation.

AIM: Periodic screening for diabetic retinopathy (DR) is effective for preventing blindness. Artificial intelligence (AI) systems could be useful for increasing the screening of DR in diabetic patients. The aim of this study was to compare the performance of the DAIRET system in detecting DR to that of ophthalmologists in a real-world setting. METHODS: Fundus photography was performed with a nonmydriatic camera in 958 consecutive patients older than 18 years who were affected by diabetes and who were enrolled in the DR screening in the Diabetes and Endocrinology Unit and in the Eye Unit of ULSS8 Berica (Italy) between June 2022 and June 2023. All retinal images were evaluated by DAIRET, which is a machine learning algorithm based on AI. In addition, all the images obtained were analysed by an ophthalmologist who graded the images. The results obtained by DAIRET were compared with those obtained by the ophthalmologist. RESULTS: We included 958 patients, but only 867 (90.5%) patients had retinal images sufficient for evaluation by a human grader. The sensitivity for detecting cases of moderate DR and above was 1 (100%), and the sensitivity for detecting cases of mild DR was 0.84 ± 0.03. The specificity of detecting the absence of DR was lower (0.59 ± 0.04) because of the high number of false-positives. CONCLUSION: DAIRET showed an optimal sensitivity in detecting all cases of referable DR (moderate DR or above) compared with that of a human grader. On the other hand, the specificity of DAIRET was low because of the high number of false-positives, which limits its cost-effectiveness.

Feasibility and accuracy of the screening for diabetic retinopathy using a fundus camera and an artificial intelligence pre-evaluation application.

AIMS: Periodical screening for diabetic retinopathy (DR) by an ophthalmologist is expensive and demanding. Automated DR image evaluation with Artificial Intelligence tools may represent a clinical and cost-effective alternative for the detection of retinopathy. We aimed to evaluate the accuracy and reliability of a machine learning algorithm. METHODS: This was an observational diagnostic precision study that compared human grader classification with that of DAIRET®, an algorithm nested in an electronic medical record powered by Retmarker SA. Retinal images were taken from 637 consecutive patients attending a routine annual diabetic visit between June 2021 and February 2023. They were manually graded by an ophthalmologist following the International Clinical Diabetic Retinopathy Severity Scale and the results were compared with those of the AI responses. The main outcome measures were screening performance, such as sensitivity and specificity and diagnostic accuracy by 95% confidence intervals. RESULTS: The rate of cases classified as ungradable was 1.2%, a figure consistent with the literature. DAIRET® sensitivity in the detection of cases of referable DR (moderate and above, "sight-threatening" forms of retinopathy) was equal to 1 (100%). The specificity, that is the true negative rate of absence of DR, was 80 ± 0.04. CONCLUSIONS: DAIRET® achieved excellent sensitivity for referable retinopathy compared with that of human graders. This is undoubtedly the key finding of the study and translates into the certainty that no patient in need of the ophthalmologist is misdiagnosed as negative. It also had sufficient specificity to represent a cost-effective alternative to manual grade alone.

Diabetic retinopathy screening with confocal fundus camera and artificial intelligence - assisted grading.

PURPOSE: Screening for diabetic retinopathy (DR) by ophthalmologists is costly and labour-intensive. Artificial Intelligence (AI) for automated DR detection could be a clinically and economically alternative. We assessed the performance of a confocal fundus imaging system (DRSplus, Centervue SpA), coupled with an AI algorithm (RetCAD, Thirona B.V.) in a real-world setting. METHODS: 45° non-mydriatic retinal images from 506 patients with diabetes were graded both by an ophthalmologist and by the AI algorithm, according to the International Clinical Diabetic Retinopathy severity scale. Less than moderate retinopathy (DR scores 0, 1) was defined as non-referable, while more severe stages were defined as referable retinopathy. The gradings were then compared both at eye-level and patient-level. Key metrics included sensitivity, specificity all measured with a 95% Confidence Interval. RESULTS: The percentage of ungradable eyes according to the AI was 2.58%. The performances of the AI algorithm for detecting referable DR were 97.18% sensitivity, 93.73% specificity at eye-level and 98.70% sensitivity and 91.06% specificity at patient-level. CONCLUSIONS: DRSplus paired with RetCAD represents a reliable DR screening solution in a real-world setting. The high sensitivity of the system ensures that almost all patients requiring medical attention for DR are referred to an ophthalmologist for further evaluation.

Improving Access to Eye Care Through Community Health Screenings Using Artificial Intelligence.

PURPOSE: To the best of our knowledge, implementation of artificial intelligence (AI)-based vision screening in community health fair settings has not been previously studied. This prospective cohort study explored the incorporation of AI in a community health fair setting to improve access to eyecare. METHODS: Vision screening was implemented during a community health fair event using an AI-based non-mydriatic fundus camera. In addition, a questionnaire was provided to survey the various barriers to eyecare and assess eye health literacy. RESULTS: A total of 53 individuals were screened at this event. Notably, about 88% of participants had follow-up appointments scheduled accordingly with an approximate 62% attendance rate. The most reported barrier to eyecare was lack of health insurance followed by transportation. CONCLUSION: The addition of AI-based vision screening in community health fairs may ultimately help improve access to eye care.

CHILDSTAR: CHIldren Living With Diabetes See and Thrive with AI Review.

Background: Artificial intelligence (AI) appears capable of detecting diabetic retinopathy (DR) with a high degree of accuracy in adults; however, there are few studies in children and young adults. Methods: Children and young adults (3-26 years) with type 1 diabetes mellitus (T1DM) or type 2 diabetes mellitus (T2DM) were screened at the Dhaka BIRDEM-2 hospital, Bangladesh. All gradable fundus images were uploaded to Cybersight AI for interpretation. Two main outcomes were considered at a patient level: 1) Any DR, defined as mild non-proliferative diabetic retinopathy (NPDR or more severe; and 2) Referable DR, defined as moderate NPDR or more severe. Diagnostic test performance comparing Orbis International's Cybersight AI with the reference standard, a fully qualified optometrist certified in DR grading, was assessed using the Matthews correlation coefficient (MCC), area under the receiver operating characteristic curve (AUC-ROC), area under the precision-recall curve (AUC-PR), sensitivity, specificity, positive and negative predictive values. Results: Among 1274 participants (53.1% female, mean age 16.7 years), 19.4% (n = 247) had any DR according to AI. For referable DR, 2.35% (n = 30) were detected by AI. The sensitivity and specificity of AI for any DR were 75.5% (CI 69.7-81.3%) and 91.8% (CI 90.2-93.5%) respectively, and for referable DR, these values were 84.2% (CI 67.8-100%) and 98.9% (CI 98.3%-99.5%). The MCC, AUC-ROC and the AUC-PR for referable DR were 63.4, 91.2 and 76.2% respectively. AI was most successful in accurately classifying younger children with shorter duration of diabetes. Conclusions: Cybersight AI accurately detected any DR and referable DR among children and young adults, despite its algorithms having been trained on adults. The observed high specificity is particularly important to avoid over-referral in low-resource settings. AI may be an effective tool to reduce demands on scarce physician resources for the care of children with diabetes in low-resource settings.

Geographic variation in diabetic retinopathy screening within the Veterans Health Administration.

AIMS: Diabetic retinopathy (DR) remains the leading cause of vision impairment among working-age adults in the United States. The Veterans Health Administration (VA) supplemented its DR screening efforts with teleretinal imaging in 2006. Despite its scale and longevity, no national data on the VA's screening program exists since 1998. Our objective was to determine the influence of geography on diabetic retinopathy screening adherence. METHODS: Setting: VA national electronic medical records. STUDY POPULATION: A national cohort of 940,654 veterans with diabetes (defined as two or more diabetes ICD-9 codes (250.xx)) without a history of DR. EXPOSURES: 125 VA Medical Center catchment areas, demographics, comorbidity burden, mean HbA1c levels, medication use and adherence, as well as utilization and access metrics. MAIN OUTCOME MEASURE: Screening for diabetic retinopathy within the VA medical system within a 2-year period. RESULTS: Within a 2-year time frame 74 % of veterans without a history of DR received retinal screenings within the VA system. After adjustment for age, gender, race-ethnic group, service-connected disability, marital status, and the van Walraven Elixhauser comorbidity score, the prevalence of DR screening varied by VA catchment area with values ranging from 27 % to 86 %. These differences persisted after further adjusting for mean HbA1c level, medication use and adherence as well as utilization and access metrics. CONCLUSIONS: The wide variability in DR screening across 125 VA catchment areas indicates the presence of unmeasured determinants of DR screening. These results are relevant to clinical decision making in DR screening resource allocation.

Accuracy of Low-Cost, Smartphone-Based Retinal Photography for Diabetic Retinopathy Screening: A Systematic Review.

Purpose: Diabetic retinopathy (DR) is a leading cause of blindness. Early DR screening is essential, but the infrastructure can be less affordable in low resource countries. This study aims to review the accuracy of low-cost smartphone-based fundus cameras for DR screening in adult patients with diabetes. Methods: We performed a systematic literature search to find studies that reported the sensitivity and specificity of low-cost smartphone-based devices for fundus photography in adult patients with diabetes. We searched three databases (MEDLINE, Google Scholar, Scopus) and one register (Cochrane CENTRAL). We presented the accuracy values by grouping the diagnosis into three: any DR, referrable DR, and diabetic macular oedema (DMO). Risk of bias and applicability of the studies were assessed using QUADAS-2. Results: Five out of 294 retrieved records were included with a total of six smartphone-based devices reviewed. All of the reference diagnostic methods used in the included studies were either indirect ophthalmoscopy or slit-lamp examinations and all smartphone-based devices' imaging protocols used mydriatic drops. The reported sensitivity and specificity for any DR were 52-92.2% and 73.3-99%; for referral DR were 21-91.4% and 64.9-100%; and for DMO were 29.4-81% and 95-100%, respectively. Conclusion: Sensitivity available low-cost smartphone-based devices for DR screening were acceptable and their specificity particularly for detecting referrable DR and DMO were considerably good. These findings support their potential utilization for DR screening in a low resources setting.

Artificial Intelligence for Diabetic Retinopathy Screening Using Color Retinal Photographs: From Development to Deployment.

Diabetic retinopathy (DR), a leading cause of preventable blindness, is expected to remain a growing health burden worldwide. Screening to detect early sight-threatening lesions of DR can reduce the burden of vision loss; nevertheless, the process requires intensive manual labor and extensive resources to accommodate the increasing number of patients with diabetes. Artificial intelligence (AI) has been shown to be an effective tool which can potentially lower the burden of screening DR and vision loss. In this article, we review the use of AI for DR screening on color retinal photographs in different phases of application, ranging from development to deployment. Early studies of machine learning (ML)-based algorithms using feature extraction to detect DR achieved a high sensitivity but relatively lower specificity. Robust sensitivity and specificity were achieved with the application of deep learning (DL), although ML is still used in some tasks. Public datasets were utilized in retrospective validations of the developmental phases in most algorithms, which require a large number of photographs. Large prospective clinical validation studies led to the approval of DL for autonomous screening of DR although the semi-autonomous approach may be preferable in some real-world settings. There have been few reports on real-world implementations of DL for DR screening. It is possible that AI may improve some real-world indicators for eye care in DR, such as increased screening uptake and referral adherence, but this has not been proven. The challenges in deployment may include workflow issues, such as mydriasis to lower ungradable cases; technical issues, such as integration into electronic health record systems and integration into existing camera systems; ethical issues, such as data privacy and security; acceptance of personnel and patients; and health-economic issues, such as the need to conduct health economic evaluations of using AI in the context of the country. The deployment of AI for DR screening should follow the governance model for AI in healthcare which outlines four main components: fairness, transparency, trustworthiness, and accountability.

Is Artificial Intelligence the Cost-Saving Lens to Diabetic Retinopathy Screening in Low- and Middle-Income Countries?

Diabetes is a rapidly growing global health crisis disproportionately affecting low- and middle-income countries (LMICs). The emergence of diabetes as a global pandemic is one of the major challenges to human health, as long-term microvascular complications such as diabetic retinopathy (DR) can lead to irreversible blindness. Leveraging artificial intelligence (AI) technology may improve the diagnostic accuracy, efficiency, and accessibility of DR screenings across LMICs. However, there is a gap between the potential of AI technology and its implementation in clinical practice. The main objective of this systematic review is to summarize the currently available literature on the health economic assessments of AI implementation for DR screening in LMICs. The review was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines. We conducted an extensive systematic search of PubMed/MEDLINE, Scopus, and the Web of Science on July 15, 2023. Our review included full-text English-language articles from any publication year. The Joanna Briggs Institute's (JBI) critical appraisal checklist for economic evaluations was used to rate the quality and rigor of the selected articles. The initial search generated 1,423 records and was narrowed to five full-text articles through comprehensive inclusion and exclusion criteria. Of the five articles included in our systematic review, two used a cost-effectiveness analysis, two used a cost-utility analysis, and one used both a cost-effectiveness analysis and a cost-utility analysis. Across the five articles, LMICs such as China, Thailand, and Brazil were represented in the economic evaluations and models. Overall, three out of the five articles concluded that AI-based DR screening was more cost-effective in comparison to standard-of-care screening methods. Our systematic review highlights the need for more primary health economic analyses that carefully evaluate the economic implications of adopting AI technology for DR screening in LMICs. We hope this systematic review will offer valuable guidance to healthcare providers, scientists, and legislators to support appropriate decision-making regarding the implementation of AI algorithms for DR screening in healthcare workflows.