Deep Learning Models Could Boost DME Diagnosis Efficiency


The best way to diagnose diabetic macular edema is using optical coherence tomography, but it is costly.

Jeffrey R. Willis, MD, PhD

Jeffrey R. Willis, MD, PhD

Artificial intelligence (AI), specifically deep learning, could help provide cost-effective, widespread eye screenings via telemedicine and assist ophthalmologists for millions of diabetes patients who might not receive regular eye exams, according to a new report.

Diabetic macular edema (DME) is the leading cause of vision impairment in people with diabetes, who may not be getting the exams they need due to cost, technical requirements of the standard test, or a variety of other reasons. Annual exams or those recommended by an ophthalmologist can prevent 95% of diabetes-related vision loss.

Investigators from Roche/Genentech’s “Ophthalmology Personalized Healthcare” initiative used large scale data from a previous phase 3 trial containing nearly 18,000 color fundus photos (CFPs) to detect swelling and its severity in the macula in order to predict and prevent ocular conditions and preserve vision in these diabetic patients.

The study authors noted in a statement that DME may be underreported due to diabetic patients not getting screenings, or when they do, it’s too late to correct the problem. That can lead to irreversible vision loss or blindness.

The best deep learning algorithm used in the analysis was up to 97% accurate in detecting DME severity using only the CFP images. However, investigators pointed out that the model’s performance was not as high when the CFPs included images of poorer quality or had laser scars.

More specifically, the model predicted central subfield thickness (CST) ≥ 250 μm and CFT ≥ 250 μm with an area under the curve (AUC) of 0.97 and 0.91, respectively. The best deep learning model predicted CST ≥ 400 μm and CFT ≥ 400 μm an AUC of 0.94 and 0.96, respectively.

“Deep learning, was able to predict three-dimensional signs/symptoms of DME using two-dimensional color-fundus photographs,” author Jeffrey R. Willis, MD, PhD told MD Magazine®. “This finding adds to the growing evidence that AI has the potential to expand access to screenings for diabetic retinopathy and to help improve patient outcomes by assisting ophthalmologists in managing their practice workflow.”

Roche/Genentech wants to be able to utilize its clinical trial database to continue to develop AI algorithms, Willis said, especially to predict presence of disease, prognosis, and response to treatment. This can lead to more high quality, personalized health care for patients, study authors wrote.

“Our study adds to the growing evidence that AI could improve the efficiency and cost-effectiveness of tele-ophthalmology screening programs for diabetic eye disease,” Willis said. “Specifically, our AI algorithm could potentially get DME patients diagnosed and treated earlier, helping ophthalmologists achieve better vision outcomes in their patients.”

Investigators wrote that their deep learning algorithm could contribute to earlier diagnoses for abnormal macular thickening, a timely referral to specialists, faster recruitment of patients into clinical trials, ad enhanced visual and health outcomes among diabetic patients.

The study, “Deep Learning Predicts OCT Measures of Diabetic Macular Thickening From Color Fundus Photographs,” was published in Investigative Ophthalmology & Visual Science.

Related Videos
A panel of 3 experts on ocular diseases and conditions
Viability of Elamipretide for Geographic Atrophy in Dry AMD | Image Credit: HCPLive
Advancing Elamipretide into Phase 3 ReNEW and ReGAIN Trials | Image Credit: HCPLive
Elamipretide in ReCLAIM-2 Trial | Image Credit: HCPLive
Hong-Uyen Hua, MD; Sruthi Arepalli, MD; and Peter Kaiser, MD
Kelly Nichols, OD; Laura Periman, MD; and Mile Brujic, OD
Cindy X. Cai, MD: Impact of Race, Neighborhood on Diabetic Retinopathy Care | Image Credit: Johns Hopkins University
© 2024 MJH Life Sciences

All rights reserved.