Diabetic macular edema (DME) is a known cause of vision loss in patients with diabetic retinopathy. Despite the current treatment options, some patients do not have access to treatment, or for those who are able to receive intravitreal injections of the approved medications, do not respond to therapy and are left with inadequately treated illness and the possibility of total blindness. Research published in late January 2020 in the Optical Society journal Biomedical Optics Express, describes a new method of employing artificial intelligence (AI) to evaluate retinal images that can algorithmically guide optimal treatment strategies in the future.

Research team leader Sina Farsiu from Duke University and colleagues explored opportunities to predict the efficacy of antivascular endothelial growth factor (VEGF) treatment of DME using optical coherence tomography images. Anti-VEGF agents are considered first-line interventions for DME, but they are not effective for all patients. This potential nonresponse sparked interest among the team to develop a method to predict and potentially avoid a useless and ineffective trial in order to facilitate a more appropriate treatment strategy and select more effective first-line therapy.

“We developed an algorithm that can be used to automatically analyze optical coherence tomography (OCT) images of the retina to predict whether a patient is likely to respond to anti-VEGF treatments,” said Dr. Farsiu, adding, “This research represents a step toward precision medicine, in which such predictions help clinicians better select first-line therapies for patients based on specific disease conditions.” 

The team conducted a retrospective analysis that included 127 persons who had received three consecutive anti-VEGF treatments for DME. Using their algorithm, they compared the OCT images from before the anti-VEGF treatments with those taken after the treatments, examining total retinal thickness pre- and post anti-VEGF treatment. The researchers used what they described as a “a novel deep convolutional neural network that was designed and evaluated using pre-treatment OCT scans as input and differential retinal thickness as output, with 5-foldcross-validation.” Response to anti-VEGF treatment was defined as a minimum decrease of 10% in retinal thickness, and the predictive abilities of the system were evaluated based on sensitivity, specificity, precision, and area under the receiver operating characteristic curve (AUC) calculations.  The team reported, “The algorithm achieved an average AUC of 0.866 in discriminating responsive from non-responsive patients with an average precision, sensitivity, and specificity of 85.5%, 80.1%, and 85.0% respectively.” The researchers also reported that classification precision increased significantly when comparing very responsive patients to very unresponsive patients. 

The authors concluded that their pilot study is a “critical step” in encouraging the use of noninvasive imaging and artificial automated analyzing processes to determine the best course of treatment for patients with specific disease states. They concluded that “this proposed automatic algorithm accurately predicts the response to anti-VEGF treatment in DME patients based on OCT imaging.” Dr. Farsiu, the lead study author, added, “Our approach could potentially be used in eye clinics to prevent unnecessary and costly trial-and-error treatments and thus alleviate a substantial treatment burden for patients.”

 « Click here to return to Diabetes Update.

​