Innovative approaches to ocular biology and technology allow for more efficient diagnosis and monitoring

News conference features four studies presented at ARVO 2019

Vancouver, BC — Four studies presented this week at the Association for Research in Vision and Ophthalmology’s (ARVO) 2019 Annual Meeting in Vancouver, British Columbia highlight how the use of innovative Artificial Intelligence technology and sophisticated biological markers may help to more efficiently diagnose and monitor eye diseases, potentially saving sight for millions of people around the world. The news conference will be held – virtually and onsite – on Tuesday, April 30, at 9am. 

Through technologic advances, researchers and clinicians are increasingly turning to artificial intelligence (AI)-based tools that use noninvasive technology supported by algorithms to identify or even predict disease patterns. Known as “deep learning,” the goal of these algorithms is to create an artificial neural network, capable of “seeing” and making intelligent decisions. At the same time, a wealth of previous data has connected certain biologic changes with specific ophthalmic disorders, and the awareness of these associations has altered the way several ocular diseases are identified, monitored and treated. Current ophthalmic practice involves an understanding of historical knowledge and rapidly advancing technologies, as well as of the roles biology and genetics play in ocular health. The new studies demonstrate how research in these areas may change clinical practice and, ultimately help save sight.  

Novel imaging approach to diabetic retinopathy screening, markers

It is well established that mitochondrial dysfunction is considered to play a crucial role in the development of diabetic retinopathy, but in vivo confirmatory data is lacking. Previous data has shown that mitochondrial dysfunction can be noninvasively assessed by measuring retinal flavoprotein fluorescence (FPF), a substance that is excreted by certain types of proteins when exposed to certain factors (i.e., oxidation) and that serves as a marker for mitochondrial metabolic changes. Raffaele Raimondi, MD and researchers on the team of Rishi P. Singh, director of the Center for Ophthalmic Bioinformatics, Cole Eye Institute and associate professor of ophthalmology, Case Western Reserve University, reviewed the charts of 135 patients without diabetes, with diabetes but without retinopathy and with diabetic retinopathy who underwent FPF imaging.

According to Raimondi, “current standard-of-care devices are often focused on the identification of structural changes rather than metabolic changes.” He explains that metabolic changes occur “earlier in the disease progression, so detection of retinal mitochondrial dysfunction is potentially a crucial tool to anticipate the diagnosis and consequently revolutionize the prognosis.”

Raimondi and his team found that using FPF to measure mitochondrial dysfunction correlated to the existence and severity of diabetic retinopathy. This means that measuring baseline FPF levels for patients may allow the use of increases in FPF intensity to screen for early diagnosis of diabetic retinopathy, before structural damage occurs.

Abstract title: Non-invasive assessment of retinal mitochondrial dysfunction in diabetic retinopathy
Presentation start/end time: Wednesday, May 1, 3 - 4:45pm
Location: West Exhibition Hall 
Abstract number: B0017 - B0048

Using color fundus photographs, deep learning to accurately predict OCT measurements of diabetic macular thickening

Researchers at Genentech/Roche have discovered that AI, specifically deep learning (DL) systems, can automatically detect the severity of diabetic macular edema using color fundus photographs (CFPs) only. Specifically, the DL algorithm was able to accurately estimate optical coherence tomography (OCT) equivalent measures of diabetic macular thickening on CFPs only. This proof-of-concept study, led by Jeffrey R. Willis, MD, PhD, analyzed almost 18,000 color fundus photographs from the phase III RIDE/RISE trials on diabetic macular edema. 

“This study adds to the growing literature that AI in ophthalmology is promising in enhancing tele-ophthalmology screening programs and in assisting ophthalmologists in improving overall vision outcomes,” said Willis, who is the associate medical director of Genentech. 

The next step, according to Willis, is validating the DL algorithm in the real world setting and understanding the potential impact on patient outcomes. In addition, researchers at Genentech/Roche plan on looking into their clinical trial databases to develop other AI algorithms that could benefit ophthalmologists and patients.  

Abstract title: Deep learning predicts OCT measures of diabetic macular thickening from color fundus photographs
Presentation start/end time: Monday, April 29, 8:15 - 10am 
Location: West Exhibition Hall
Abstract number: A0206 - A0225

Deep-learning system used to screen eye for signs of diabetes 

Another study examining the role of AI in the early diagnosis of disease focused specifically on whether a deep-learning system could successfully estimate levels of hemoglobin A1c (Hba1c), an important marker for diabetes, using retinal photos and serum samples. Researchers from the Singapore Eye Research Institute, part of the Singapore National Eye Centre, retrospectively reviewed five population-based and clinical eye studies, with a total of 17,422 participants. 

Fundus images and serum measurements were used for the deep learning system’s training and validation of results. The system generated a slight underestimation of HbA1c for patients with diabetes and a slight overestimation for healthy individuals. However, overall, the results for the deep learning system’s ability to estimate Hba1c levels as a screening and monitoring tool for diabetes is promising. Further study is required.

“This new deep learning-based modality was successful in estimating Hba1c and may offer a paradigm shift in diabetes care,” said lead researcher Yih Chung Tham, PhD. 

Abstract title: Estimation of Haemoglobin A1c from retinal photographs via deep learning
Presentation start/end time: Monday, April 29, 8:15 - 10am  
Location: West Exhibition Hall
Abstract number: A0117 - A0160

A more efficient way to determine volume-related characteristics in AMD 

Age-related macular degeneration (AMD) is a leading cause of vision loss worldwide among people aged 50 and older. Advanced stages of this disease are often accompanied by growth of abnormal blood vessels below and in the deep layers of the retina, a condition known as choroidal neovascularization (CNV). Swept-Source OCT Angiography (SS-OCTA) is a noninvasive imaging technology that enables visualization of these retinal vessels in impressive detail (i.e., three dimensions), allowing for the detection and tracking of CNV. However, volume-related characteristics of CNV are very difficult to quantify in SS-OCTA images due to the complexity and time intensity associated with the manual annotations. 

Researchers at Carl Zeiss Meditec Inc., led by Luis de Sisternes, PhD, and through collaboration with Bascom Palmer Eye Institute and Créteil University Hospital, have developed an AI-based method to automatically detect the presence of CNV within these images and generate quantifiable characteristics describing the CNV volume, area, density, and invasiveness (as defined by measurement of abnormal vessel intrusion within the retina). Using SS-OCTA images for 42 eyes of patients with CNV and 20 eyes of healthy individuals for training and testing purposes, the AI-based method correctly identified CNV eyes from eyes without CNV with 100% accuracy in the tested eyes. 

De Sisternes explained that “identifying the presence and location of CNV using SS-OCTA data is extremely laborious and time-consuming, and annotating such large volumetric data is simply unviable in a clinical setting today.” The AI-based model developed by de Sisternes and colleagues accelerates clinical research and “enables clinicians to push the boundaries of patient management by easily assessing complex data that can help with monitoring patient status and response to treatment over time,” said de Sisternes.

Abstract title: Automated volumetric choroidal neovascularization segmentation and quantification in swept-source OCT angiography using machine learning
Presentation start/end time: Tuesday, April 30, 11:45am - 1:30pm
Location: West Exhibition Hall
Abstract number: A0282 - A0341


The Association for Research in Vision and Ophthalmology (ARVO) is the largest eye and vision research organization in the world. Members include nearly 12,000 eye and vision researchers from over 75 countries. ARVO advances research worldwide into understanding the visual system and preventing, treating and curing its disorders. Learn more at

The 2019 ARVO Annual Meeting will take place in Vancouver, BC from April 28 – May 2. The Meeting is the premiere gathering of nearly 12,000 eye and vision researchers from around the world. During the Meeting, more than 6,600 abstracts will be presented on the latest basic and translational research in eye and vision science.

All abstracts accepted for presentation at the ARVO Annual Meeting represent previously unpublished data and conclusions. This research may be proprietary or may have been submitted for journal publication. Embargo policy: Journalists must seek approval from the presenter(s) before reporting data from paper or poster presentations. Press releases or stories on information presented at the ARVO Annual Meeting may not be released or published until the conclusion of the presentation.

Media contact:
Julene Joy