AI Could Help Track Response to Anti-VEGF Therapy for Diabetic Macular Edema

By Reuters Staff

August 10, 2020

NEW YORK (Reuters Health) - Artificial intelligence (AI) analysis of imaging data could help monitor treatment response in patients with diabetic macular edema (DME) undergoing treatment with anti-vascular endothelial growth factor (VEGF), a new study shows.

"Artificial intelligence may be used to automatically measure fluid volumes in a reliable way and thus may facilitate a comparison of the differential efficacy of anti-VEGF agents," Dr. Ursula M. Schmidt-Erfurth of the Medical University of Vienna and colleagues write in JAMA Ophthalmology.

Dr. Schmidt-Erfurth and her team developed a deep-learning algorithm to quantify volumetric changes in intraretinal fluid (IRF) and subretinal fluid (SRF) by analyzing spectral-domain optical coherence tomographic (SD-OCT) data.

They performed a post hoc analysis on nearly 7,000 scans of 570 eyes from 570 participants in the Diabetic Retinopathy Clinical Research Network's protocol T trial, which compared ranibizumab, bevacizumab and aflibercept. Participants underwent SD-OCT every four weeks for a year.

Mean fluid volumes at baseline were 448.6 nL for IRF and 36.9 nL for SRF and 161.2 nL and 4.4 nL, respectively, at 12 months. Study participants with SRF at the study's outset had mean best visual acuity (BCVA) of 63.2, versus 66.9 for those without SRF at baseline.

IRF volume was reduced significantly more with aflibercept compared with bevacizumab, with a 79.8 nL difference after the first injection and a 10.4 nL difference every four weeks.

The first injection with ranibizumab also reduced IRF significantly more than bevacizumab (difference, 75.2 nL), with a "borderline" difference at subsequent scans (difference, 6.3 nL; P=0.07). There were no difference in the effect of aflibercept and ranabizumab on IRF volume, and SRF volume change was similar for all three drugs.

Participants with SRF at baseline had an average 0.5 point increase in Early Treatment Diabetic Retinopathy Study scores every four weeks, compared to 0.4 for patients with no SRF. By month 12, there was no difference in BCVA or IRF volume between patients with and without SRF at baseline.

"With use of AI, an automated and objective profiling of the efficacy of different substances was possible by automated fluid quantification. Despite the limited correlation with visual function found in our study, the presence and amount of IRF and SRF served as measures of individual treatment response," the authors write. "Thus, automated fluid segmentation could support the clinical applicability of individualized medicine."

In a linked commentary, Dr. Peter Koulen of the University of Missouri-Kansas City and colleagues write, "These findings are in accordance with previous work demonstrating aflibercept's superiority compared with other anti-VEGF treatments in improving functional and anatomical outcomes in DME, particularly in patients with a BCVA of 20/50 or worse."

They conclude: "Although additional prospective trials are needed to clarify these associations, the findings from the current study may facilitate more informed clinical decision-making in the treatment of patients with DME."

One of Dr. Schmidt-Erfurth's coauthors has a patent to analyze OCT images pending.

Neither Dr. Schmidt-Erfurth nor Dr. Koulen was available for an interview by press time.

SOURCE: and JAMA Ophthalmology, online July 31, 2020.