Advances in Screening Technology – Essential Tools for Managing Diabetic Retinopathy

Introduction

As discussed in last month’s OcuTerra Insights, the ongoing diabetes pandemic – and the concurrent increase in diagnoses of diabetic retinopathy (DR) in the US and worldwide – has created the urgent need for safe and effective alternatives to currently available treatment options. Non-invasive treatments that can be administered early (and, in the case of OTT166, administered at home by the patient), could delay or even prevent progression of disease, benefiting millions of patients and transforming the standard of care for this potentially devastating condition.

New treatments represent just one approach to tackling the DR pandemic. On the other side of the coin is the development of new and improved diagnostic tools that can accurately and consistently diagnose disease earlier and less invasively, so safe and effective treatment can begin before disease progresses and vision may become threatened.

Current diagnosis using retinal imaging

DR can be diagnosed by an optometrist or ophthalmologist with a simple examination, and in many cases, diagnosis can occur before any overt symptoms are noticed by the patient; this underscores the need for everyone, especially those with diabetes, to have regular eye exams. Tests include fluorescein angiography, which requires the intravenous injection of a contrast dye to highlight a patient’s blood vessels as broken or leaking vessels may indicate disease, or optical coherence tomography (OCT), which uses light waves to take a cross-section picture to assess the thickness of a patient’s retina. An increased thickness may indicate a buildup of fluid caused by leaky blood vessels. As a readout of treatment effectiveness, retinal thickness measured by OCT is also used in follow-up examinations to determine how disease is progressing1.

Following initial diagnosis, there is at present no way to accurately predict which patients will progress to more severe stages of disease, or how quickly this will occur. Physicians can estimate the risk of progression based on established severity scales – ranging from mild to vision-threatening DR – but patients must be regularly monitored to determine the best time to intervene with currently available treatments.

OcuTerra Vice President of Clinical and Medical Affairs Majid Anderesi, MD, presented at the Retinal Imaging Biomarkers & Endpoints Summit in Boston last month, showcasing the power of imaging modalities including fundus photography, OCT, OCT-A, FA, and wide-field imaging as tools for diagnosis and assessment of disease progression. Dr. Anderesi also covered their use in the identification and development of non-invasive biomarkers of disease severity and progression, and the advantages of adopting multimodal imaging technologies to facilitate clinical advancements – including in the Phase 2 DR:EAM trial evaluating OTT166, OcuTerra’s novel eye drop-based integrin inhibitor therapy2.

Representative images of the retina taken using (A) fluorescein angiography10, optical coherence tomography10 (B), and ultra-wide field color fundus photography11 (C; white circles show extent of standard field).

Artificial intelligence and machine learning 

Research presented at this June’s American Diabetes Association Scientific Sessions in San Diego described the use of machine learning to identify the risk of early progression by assessing retinal images. Out of nearly 10,000 images, the algorithm was able to accurately assign disease progression risk in 91% of cases, demonstrating the feasibility of utilizing machine learning, and its potential to reduce costs and improve patient outcomes3.

Another study, from the Johns Hopkins School of Medicine, looked at over 22,000 patients to compare adherence to annual eye testing guidelines – which has been historically low. The study found that deployment of AI-assisted eye examinations at primary care clinics increased rates of adherence, indicating that incorporation of AI technologies could benefit both effective, early diagnosis of disease, and management of ongoing treatments4.

In a presentation at this year’s Association for Research in Vision and Ophthalmology (ARVO) Annual Meeting, Dr. Tien Wong, Medical Director of the Singapore National Eye Center, outlined the role of AI and machine learning in evolving screening strategies for DR. Effective screening programs for diabetic patients are known to have real benefits in preventing vision loss – Dr. Wong cited a case study in Sweden where a national screening program for all patients with diabetes was implemented between 1989 and 1990 and by 1995, they had reported a 47% reduction in diabetes-related blindness5. However, one reason why so few countries have national screening programs is that they are labor intensive, in part because retinal images must be captured and graded by trained professionals. Moreover, with the increasing prevalence of diabetes, the potential patient population is huge – one study estimated that universal adoption of DR screening in the US would require evaluation of approximately 32 million images per year6.

As already demonstrated, it is possible to accurately grade retinal images with the help of AI and machine learning – but how much could this impact our ability to efficiently screen millions of patients? One study recently published in Nature pitted an AI-based deep learning system against 17 human assessors (ten ophthalmologists and seven professional graders) to compare accuracy and timing. The multicenter study used almost 100,000 images taken of just under 20,000 patients from Australia, China, Hong Kong, Singapore, and the US. While the human assessors needed two years to work through the images, the AI managed it in around one month, and the grading results were very similar. The AI wasn’t perfect, however – around 8% of the images were deemed “ungradable” and would require manual grading by a trained human professional7. As such, the present optimal model for integrating AI into the screening process would be as a triage tool, where the AI screens a large number of images, and potential positives, along with those deemed ‘ungradable’, are referred to human professionals for confirmation. This setup is currently under evaluation in Singapore, using the SELENA+ (Singapore Eye Lesion Analyzer) algorithm to help detect diabetic eye disease, as well as glaucoma and macular degeneration8.

Conclusion

Improvements in imaging technology and the development of better, faster AI tools have the potential to transform the management of DR, assisting with early diagnosis and ongoing treatment evaluation, as well as helping to develop more accurate predictive models for disease progression9.

With the ongoing diabetes pandemic driving increased prevalence of DR, combining earlier detection with the development of early and non-invasive treatment options will help us to move away from the current paradigm of “watch and wait” and delaying therapeutic intervention until a vision-threatening complication arises. OcuTerra is currently evaluating its non-invasive integrin inhibitor OTT166 in the Phase 2 DR:EAM trial – which recently achieved full enrollment of 225 patients – in an effort to provide patients with a safe, effective treatment option to slow or even prevent disease progression. You can find more information on OTT166 and the DR:EAM trial on our website.

References

  1. No authors listed. Grading diabetic retinopathy from stereoscopic color fundus photographs--an extension of the modified Airlie House classification. ETDRS report number 10. Early Treatment Diabetic Retinopathy Study Research Group. Ophthalmology. 1991;98(5 Suppl):786-806.

  2. Anderesi M. Optimizing Approaches to Visualize Diabetic Retinopathy & Clinically Measure Therapeutic-Induced Changes in Blood Flow. Retinal Imaging Biomarkers & Endpoints Summit; Oral Presentation. 2023

  3. Nigam A, Sun J, Subhash V, Silva PS; 26-LB: Identifying the Risk of Diabetic Retinopathy Progression Using Machine Learning on Ultrawide Field Retinal Images. Diabetes 20 June 2023; 72 (Supplement_1): 26–LB.

  4. Ariel Leong, Jiangxia Wang, Risa Wolf, Roomasa Channa, Michael David Abramoff, Harold Lehmann, T. Y. Alvin Liu; Autonomous artificial intelligence (AI) increases health equity for patients who are more at risk for poor visual outcomes due to diabetic eye disease (DED). Invest. Ophthalmol. Vis. Sci. 2023;64(8):243.

  5. Bäcklund LB, Algvere PV, Rosenqvist U. New blindness in diabetes reduced by more than one-third in Stockholm County. Diabet Med. 1997;14(9):732-740. doi:10.1002/(SICI)1096-9136(199709)14:9<732::AID-DIA474>3.0.CO;2-J

  6. Garg S. Diabetic Retinopathy Screening With Telemedicine: A Potential Strategy to Engage Our Youth. JAMA Ophthalmol. 2017;135(5):438-439. doi:10.1001/jamaophthalmol.2017.0150

  7. Ting, D.S.W., Cheung, C.Y., Nguyen, Q. et al. Deep learning in estimating prevalence and systemic risk factors for diabetic retinopathy: a multi-ethnic study. npj Digit. Med. 2, 24 (2019). https://doi.org/10.1038/s41746-019-0097-x

  8. Lupidi, M., Danieli, L., Fruttini, D. et al. Artificial intelligence in diabetic retinopathy screening: clinical assessment using handheld fundus camera in a real-life setting. Acta Diabetol 60, 1083–1088 (2023). https://doi.org/10.1007/s00592-023-02104-0

  9. Artificial intelligence for diabetic retinopathy, Sicong Li et al., Chin Med J (Engl). 2022 Feb 5; 135(3): 253–260., Published online 2021 Dec 8. doi: 10.1097/CM9.0000000000001816

  10. Daruich A, Matet A, Moulin A, et al. Mechanisms of macular edema: Beyond the surface. Prog Retin Eye Res. 2018;63:20-68. doi:10.1016/j.preteyeres.2017.10.006

  11. Silva PS, Cavallerano JD, Sun JK, Soliman AZ, Aiello LM, Aiello LP. Peripheral lesions identified by mydriatic ultrawide field imaging: distribution and potential impact on diabetic retinopathy severity. Ophthalmology. 2013;120(12):2587-2595. doi:10.1016/j.ophtha.2013.05.004

Brad Good